<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Matthew Armstrong</title>
    <description>The latest articles on DEV Community by Matthew Armstrong (@open-madscience).</description>
    <link>https://dev.to/open-madscience</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4171433%2F8deeb383-d1eb-4ec0-955e-612d43f55925.png</url>
      <title>DEV Community: Matthew Armstrong</title>
      <link>https://dev.to/open-madscience</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/open-madscience"/>
    <language>en</language>
    <item>
      <title>Spicey Electrons</title>
      <dc:creator>Matthew Armstrong</dc:creator>
      <pubDate>Thu, 19 Nov 2026 15:12:33 +0000</pubDate>
      <link>https://dev.to/open-madscience/spicey-electrons-57gg</link>
      <guid>https://dev.to/open-madscience/spicey-electrons-57gg</guid>
      <description>&lt;p&gt;&lt;em&gt;(Formerly called Sim Sala Bim.)&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;New:&lt;/strong&gt; capacitors and MOSFETs show a 5-segment charge bar ("C1 charge", "M1 gate") that fills as you scrub or press Play.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;A circuit simulator a 14-year-old can figure out, with a real SPICE engine underneath.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Drag parts onto the board, wire their legs together, press ▶ Run, then clip scope probes onto&lt;br&gt;
wires (voltage) or parts (current) and watch the waveforms. Wires glow green for + and red for −,&lt;br&gt;
LEDs light up, and if you'd burn something out it tells you, in plain English, before your&lt;br&gt;
fingers find out.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fmizbvk3a64z07u7lj0vg.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fmizbvk3a64z07u7lj0vg.webp" alt="The transistor-switch example: a switch, a 10 kΩ resistor, an NPN transistor and a green LED, with the scope showing base and collector current" width="799" height="482"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;📘 &lt;strong&gt;&lt;a href="https://github.com/The-Dorkknight/spicey-electrons/blob/main/docs/SpiceyElectrons-manual.pdf" rel="noopener noreferrer"&gt;The whole manual and research notes as one PDF&lt;/a&gt;&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Licence:&lt;/strong&gt; source-available under the &lt;a href="https://github.com/The-Dorkknight/spicey-electrons/blob/main/LICENSE.md" rel="noopener noreferrer"&gt;PolyForm Noncommercial License 1.0.0&lt;/a&gt;. Free&lt;br&gt;
for personal, educational and other non-commercial use; commercial use needs permission. It comes&lt;br&gt;
with &lt;strong&gt;no warranty and no liability&lt;/strong&gt;: it's a learning tool, not a design sign-off, and the&lt;br&gt;
mains-voltage examples are deliberately low-voltage.&lt;/p&gt;

&lt;p&gt;Project started 2026-11-18 14:12.&lt;/p&gt;




&lt;h2&gt;
  
  
  Why another simulator?
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;SPICE (LTspice, ngspice)&lt;/th&gt;
&lt;th&gt;Falstad CircuitJS&lt;/th&gt;
&lt;th&gt;Tinkercad Circuits&lt;/th&gt;
&lt;th&gt;&lt;strong&gt;Spicey Electrons&lt;/strong&gt;&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Real SPICE models and maths&lt;/td&gt;
&lt;td&gt;✅&lt;/td&gt;
&lt;td&gt;❌ own simplified engine&lt;/td&gt;
&lt;td&gt;❓ closed&lt;/td&gt;
&lt;td&gt;✅ ngspice&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Easy for beginners&lt;/td&gt;
&lt;td&gt;❌ netlists, &lt;code&gt;.tran&lt;/code&gt;, cryptic errors&lt;/td&gt;
&lt;td&gt;✅&lt;/td&gt;
&lt;td&gt;✅&lt;/td&gt;
&lt;td&gt;✅&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Runs offline on your own PC&lt;/td&gt;
&lt;td&gt;✅&lt;/td&gt;
&lt;td&gt;✅&lt;/td&gt;
&lt;td&gt;❌ online account&lt;/td&gt;
&lt;td&gt;✅&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Plain-English errors and "it'll burn" warnings&lt;/td&gt;
&lt;td&gt;❌&lt;/td&gt;
&lt;td&gt;partly&lt;/td&gt;
&lt;td&gt;partly&lt;/td&gt;
&lt;td&gt;✅&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Temperature (0 / 20 / 40 °C)&lt;/td&gt;
&lt;td&gt;✅ (if you know how)&lt;/td&gt;
&lt;td&gt;❌&lt;/td&gt;
&lt;td&gt;❌&lt;/td&gt;
&lt;td&gt;✅ one click&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The research behind this (how accurate Falstad really is, licences, why not fork it) is in&lt;br&gt;
&lt;a href="https://github.com/The-Dorkknight/spicey-electrons/blob/main/DESIGN_NOTES.md" rel="noopener noreferrer"&gt;DESIGN_NOTES.md&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  How to use it
&lt;/h2&gt;

&lt;h3&gt;
  
  
  1. Add parts
&lt;/h3&gt;

&lt;p&gt;Parts are grouped on the left (Power, Basics, Semiconductors, Chips). Point at one to see what it&lt;br&gt;
does. &lt;strong&gt;Click it, then click the board&lt;/strong&gt; to drop it (or drag it straight on). Press &lt;strong&gt;R&lt;/strong&gt; while&lt;br&gt;
placing to turn it.&lt;/p&gt;

&lt;h3&gt;
  
  
  2. Wire them up
&lt;/h3&gt;

&lt;p&gt;&lt;strong&gt;Drag from a leg dot to another leg dot.&lt;/strong&gt; The wire bends into an L. You can also pick 〰️ Wire and&lt;br&gt;
click corner by corner (Esc to stop). A wire that ends on the middle of another wire joins it (a&lt;br&gt;
dot appears); wires that just cross don't connect, just like a real circuit diagram.&lt;/p&gt;

&lt;h3&gt;
  
  
  3. Press ▶ Run
&lt;/h3&gt;

&lt;p&gt;No ground symbol? It uses the battery's minus side as 0 V and tells you so. Legs that aren't&lt;br&gt;
connected show up as red dots, with a message saying which one.&lt;/p&gt;

&lt;h3&gt;
  
  
  4. Probe it
&lt;/h3&gt;

&lt;p&gt;Pick 🔎 Probe and:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;click a wire&lt;/strong&gt; to see its &lt;strong&gt;voltage&lt;/strong&gt; (round clip, solid line on the scope)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;click a part&lt;/strong&gt; to see the &lt;strong&gt;current through it&lt;/strong&gt; (square clip, dashed line on the scope)&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Up to six probes, each in its own colour. You can add probes after a run without running again:&lt;br&gt;
every voltage and current is already worked out. Click a probe's letter to remove it.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ftvdu4h9g2xs2rt4kolnl.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ftvdu4h9g2xs2rt4kolnl.webp" alt="Hovering a part shows its current and the voltage across it; the scope cursor is at 1 time constant" width="799" height="482"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  5. Explore
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Drag on the scope&lt;/strong&gt; to move the time cursor. The wires recolour and the LEDs brighten to match
that moment. &lt;strong&gt;▶ Play&lt;/strong&gt; runs it in slow motion.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Hover&lt;/strong&gt; any wire or part for its voltage or current at the cursor time.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Change a value:&lt;/strong&gt; double-click a part (or click its blue value). A little editor pops up on the board; type a number, pick the size from the list (k kilo, µ micro, n nano, p pico...) and it re-runs by itself.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Click a switch&lt;/strong&gt; to flip it. &lt;strong&gt;Slide a volume knob (pot)&lt;/strong&gt; and watch the result move.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Temperature:&lt;/strong&gt; ❄️ 0 °C, 20 °C or ☀️ 40 °C. Diodes, LEDs and transistors change with temperature
(about −2 mV per °C on a diode, as in real life).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Show:&lt;/strong&gt; how long a slice of time to simulate. "auto" picks about 5 cycles of the slowest signal
or about 5 time constants.&lt;/li&gt;
&lt;li&gt;Under each result, &lt;strong&gt;"For the curious"&lt;/strong&gt; shows the exact SPICE netlist that was run, so you can
learn real SPICE when you're ready.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F9wkij5z4c11bbyo9htfe.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F9wkij5z4c11bbyo9htfe.webp" alt="Op-amp inverting amplifier: 0.5 V in, 5 V out, upside down" width="" height=""&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Transformers, coils and very fast circuits (new in 0.2.0)
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Transformer&lt;/strong&gt;: two coils on one core. Set the &lt;em&gt;turns ratio&lt;/em&gt; (secondary ÷ primary; 0.5 steps the
voltage down by half), the primary coil size and how well the coils share (0.99 is typical). The
dots mark the ends that go up together. It only passes changing power, so DC stays on the primary side.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Show / From&lt;/strong&gt;: &lt;em&gt;Show&lt;/em&gt; is how long a slice of time you see, &lt;em&gt;From&lt;/em&gt; skips ahead first (like the
delay dial on a scope). Fast circuits get fine time steps automatically, and the scope draws a
shaded min/max band where the wiggles are too close together to draw one by one.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Hz button&lt;/strong&gt;: on a voltage probe whose wire is oscillating, the &lt;strong&gt;Hz&lt;/strong&gt; chip switches the trace to
&lt;em&gt;frequency&lt;/em&gt;, zoomed in, so you can watch it swing. That is how you see FM.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fh90q9luwdxs2uw1wo7vp.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fh90q9luwdxs2uw1wo7vp.webp" alt="A one-transistor FM bug: a 20 kHz voice sound swings the 86 MHz carrier" width="800" height="536"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fnfw29xzpjzd89vxlqsti.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fnfw29xzpjzd89vxlqsti.webp" alt="A Joule thief lighting a 1.8 V LED from a 1.5 V battery" width="800" height="536"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F6zqo6ff0jb186c1sxm00.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F6zqo6ff0jb186c1sxm00.webp" alt="A transformer, diode and capacitor power supply" width="800" height="536"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Safety warnings
&lt;/h3&gt;

&lt;p&gt;Results are checked against what real parts can take: LEDs over about 30 mA, resistors over 0.25 W,&lt;br&gt;
a battery nearly short-circuited, transistors and diodes past their rating. The part turns red with&lt;br&gt;
a 🔥 and the message says how to fix it. A short switch-on surge (like a capacitor filling) does not&lt;br&gt;
count unless it's ten times over the limit.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Frlsxfh9kz3y7j0b1sm0x.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Frlsxfh9kz3y7j0b1sm0x.webp" alt="A 10 Ω resistor in series with an LED: burn-out warnings" width="799" height="482"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Parts
&lt;/h3&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Group&lt;/th&gt;
&lt;th&gt;Parts&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Power&lt;/td&gt;
&lt;td&gt;Battery (with 0.1 Ω internal resistance), Wave maker (sine), Clock (square), Ground&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Basics&lt;/td&gt;
&lt;td&gt;Resistor, Capacitor, Coil, Transformer, Volume knob (potentiometer), Switch&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Semiconductors&lt;/td&gt;
&lt;td&gt;Zener diode, Diode (1N4148, 1N4007), LED (red, yellow, green, blue, white), NPN (2N3904, 2N2222), PNP (2N3906), N-MOSFET (2N7000)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Lights, sound &amp;amp; motion&lt;/td&gt;
&lt;td&gt;Lamp (heats up, so it surges at switch-on), Speaker, Relay (clicks, with hysteresis), Motor (makes its own back-voltage)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Sensors&lt;/td&gt;
&lt;td&gt;Light sensor (LDR), Thermistor (gets low when warm)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Chips&lt;/td&gt;
&lt;td&gt;Op-amp (with its ±12 V supply wired up for you), 555 timer (reset and control pins tied off for you)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Logic gates&lt;/td&gt;
&lt;td&gt;NOT, AND, NAND, OR, NOR, XOR (high level 3.3, 5, 9 or 12 V; they have no supply pins to wire)&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h3&gt;
  
  
  Examples (📚 menu): 41 of them
&lt;/h3&gt;

&lt;p&gt;Every one is a lesson, and each has a page in the &lt;strong&gt;Experimenters' Workbook&lt;/strong&gt; (&lt;a href="https://github.com/The-Dorkknight/spicey-electrons/blob/main/docs/SpiceyElectrons-workbook.pdf" rel="noopener noreferrer"&gt;docs/SpiceyElectrons-workbook.pdf&lt;/a&gt;),&lt;br&gt;
a jargon-light guide and exercise book for all ages, with questions, answers and six open challenges.&lt;/p&gt;

&lt;p&gt;First sparks: LED + resistor · Dim an LED with a volume knob. Ohm and Kirchhoff: Ohm's law · Voltage divider · Kirchhoff's loop · Current divider. Storing energy: Capacitor charging ·&lt;br&gt;
Rounding a square wave · RC high-pass · Coil current builds slowly · Ringing coil + capacitor (RLC).&lt;br&gt;
One-way streets: Rectifier · Zener regulator · Diode clipper. Switches: Transistor switch · Transistor inverter ·&lt;br&gt;
Emitter follower. Amplifiers: Op-amp ×10 · Common-emitter (Class A) · Class B · Class AB · Class D.&lt;br&gt;
Things that tick: 555 flasher · 555 one-shot · 555 tone maker · Two-transistor flasher · Astable multivibrator&lt;br&gt;
(waveforms) · Monostable multivibrator · Ring oscillator. Decisions: Schmitt trigger (hysteresis) · Heat alarm ·&lt;br&gt;
Night light · Logic gates: half adder. Power, light and motion: Transformer supply · Light bulb surge · Motor&lt;br&gt;
spin-up · Relay with a flyback diode · Joule thief. Curiosities: MOSFET Miller plateau · MOSFET false turn-on · FM bug.&lt;/p&gt;

&lt;h3&gt;
  
  
  Moving current dots and charge bars
&lt;/h3&gt;

&lt;p&gt;Tick &lt;strong&gt;Current dots&lt;/strong&gt; in the scope bar: dots stream along every wire in the direction the current flows, faster and&lt;br&gt;
bigger for more current. The wire currents come from the part currents using Kirchhoff's current law, so the dots&lt;br&gt;
appear only on wire networks without loops (a loop of wire gets no dots rather than wrong ones). Capacitors and&lt;br&gt;
MOSFET gates show a 5-box battery-style charge bar.&lt;/p&gt;

&lt;h2&gt;
  
  
  What exists and what doesn't (yet)
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Works now&lt;/th&gt;
&lt;th&gt;Not yet&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Drag-and-drop editor, wires, junctions, undo&lt;/td&gt;
&lt;td&gt;Live "turn the volume knob while it runs" (it re-runs instead)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Real ngspice transient simulation, 0/20/40 °C, animated current dots&lt;/td&gt;
&lt;td&gt;AC frequency sweeps (Bode plots), DC sweeps&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Voltage and current probes, 6-trace scope, cursor, slow-motion play&lt;/td&gt;
&lt;td&gt;Export to SoloTrace for a real PCB&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Plain-English errors and burn-out warnings&lt;/td&gt;
&lt;td&gt;A no-install version in the browser (ngspice as WebAssembly)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Save / open / auto-keep, 41 examples, workbook PDF&lt;/td&gt;
&lt;td&gt;Photo-realistic parts&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Linux tested for real&lt;/td&gt;
&lt;td&gt;Windows and Mac launchers not run yet&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  More
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;a href="https://github.com/The-Dorkknight/spicey-electrons/blob/main/HOW_IT_WORKS.md" rel="noopener noreferrer"&gt;HOW_IT_WORKS.md&lt;/a&gt;: what happens when you press Run&lt;/li&gt;
&lt;li&gt;
&lt;a href="https://github.com/The-Dorkknight/spicey-electrons/blob/main/DESIGN_NOTES.md" rel="noopener noreferrer"&gt;DESIGN_NOTES.md&lt;/a&gt;: research and design decisions&lt;/li&gt;
&lt;li&gt;
&lt;a href="https://github.com/The-Dorkknight/spicey-electrons/blob/main/CHANGELOG.md" rel="noopener noreferrer"&gt;CHANGELOG.md&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/spicey-electrons/blob/main/THIRD_PARTY.md" rel="noopener noreferrer"&gt;THIRD_PARTY.md&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/spicey-electrons/blob/main/CONTRIBUTING.md" rel="noopener noreferrer"&gt;CONTRIBUTING.md&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/spicey-electrons/blob/main/SECURITY.md" rel="noopener noreferrer"&gt;SECURITY.md&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Tests: &lt;code&gt;python3 -m unittest discover -s tests -v&lt;/code&gt; (the simulations are checked against hand maths)&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;&lt;strong&gt;Source code:&lt;/strong&gt; &lt;a href="https://github.com/The-Dorkknight/spicey-electrons" rel="noopener noreferrer"&gt;The-Dorkknight/spicey-electrons on GitHub&lt;/a&gt;&lt;/p&gt;

</description>
      <category>education</category>
      <category>software</category>
      <category>tools</category>
    </item>
    <item>
      <title>Icy Ewe: 3D scanning on the cheap with a scratched depth camera and a slow laptop</title>
      <dc:creator>Matthew Armstrong</dc:creator>
      <pubDate>Sat, 14 Nov 2026 12:27:00 +0000</pubDate>
      <link>https://dev.to/open-madscience/icy-ewe-3d-scanning-on-the-cheap-with-a-scratched-depth-camera-and-a-slow-laptop-1498</link>
      <guid>https://dev.to/open-madscience/icy-ewe-3d-scanning-on-the-cheap-with-a-scratched-depth-camera-and-a-slow-laptop-1498</guid>
      <description>&lt;p&gt;Second-hand depth cameras (an Intel RealSense SR305, the D415 hiding inside a Prowise MOVE 3in1) see depth 30 times a&lt;br&gt;
second for the price of a takeaway. Turning one into a 3D scanner is easy right up to the point where you line the&lt;br&gt;
frames up. Then the object grows a second nose.&lt;/p&gt;

&lt;p&gt;I'm working on about 50 projects at once, so I don't have the time or brain capacity for humour; it cuts into my four&lt;br&gt;
hours of sleep-eating time (efficiency). So, straight to it, apart from the title: &lt;strong&gt;Icy Ewe&lt;/strong&gt; (say it out loud) is the write-up of how my app,&lt;br&gt;
Scantastic, keeps a scan together on bad cameras and slow computers:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Picture features check every step&lt;/strong&gt; the 3D shape suggests, and veto it when the shape is fooling itself.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Loop closures can't lie:&lt;/strong&gt; one the pictures can't confirm may only move a frame as far as the chain could really
have drifted. On a plain test blob, 40 of 42 closures had been wrong.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Turntable mode&lt;/strong&gt; holds every frame to one fitted axis; a printed marker ring is optional.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;TRACKING LOST&lt;/strong&gt; tells you while you can still do something about it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Slow PCs record now and stitch afterwards&lt;/strong&gt;, in a separate process that just tries again on fewer frames if it
runs out of memory.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Speed profiles, side-by-side alignment with identical answers, and a graphics card that has to pass a test first.&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A sensor check&lt;/strong&gt; that maps where your camera's window is scratched.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The honest bit: all the measured results come from made-up scans with known answers, and the new methods haven't been&lt;br&gt;
through a real camera yet. There's a test plan in section 8. If you have one of these cameras, please break it and tell&lt;br&gt;
me what happened.&lt;/p&gt;

&lt;p&gt;It's also a &lt;strong&gt;defensive publication&lt;/strong&gt;: section 9 dedicates the methods and their variations to the public, so they stay&lt;br&gt;
free for hobbyists.&lt;/p&gt;

&lt;p&gt;Paper: &lt;a href="https://github.com/The-Dorkknight/icy-ewe" rel="noopener noreferrer"&gt;https://github.com/The-Dorkknight/icy-ewe&lt;/a&gt; · App: &lt;a href="https://github.com/The-Dorkknight/scantastic" rel="noopener noreferrer"&gt;https://github.com/The-Dorkknight/scantastic&lt;/a&gt;&lt;/p&gt;




&lt;h2&gt;
  
  
  The paper
&lt;/h2&gt;




&lt;p&gt;title: "Icy Ewe"&lt;br&gt;
subtitle: "3D scanning on the cheap: turning a scratched, second-hand depth camera and a slow laptop into a 3D scanner that doesn't lose the plot. A hobbyist's field guide and a defensive publication."&lt;br&gt;
author: "Matthew Armstrong (The-Dorkknight)"&lt;/p&gt;

&lt;h2&gt;
  
  
  date: "Version 1.1, 13 November 2026 (first published as \"Depth Perception\", 2 November 2026)"
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffwqkvo9x2vg0ecvveu9s.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffwqkvo9x2vg0ecvveu9s.jpg" alt="**&amp;quot;Icy Ewe&amp;quot;**, the cover art and the title of this paper: a sheep in a parka points a scanner at an icy rune stone. Say it out loud: *icy ewe* sounds like **&amp;quot;I see you&amp;quot;**, which is a depth camera's whole job, and this one is doing it on the cheap, in the cold, with second-hand kit. (The ewe is wearing glasses because the camera's window is scratched.)" width="799" height="436"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7bwo0502lsyz0b7vyv56.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7bwo0502lsyz0b7vyv56.jpg" alt="**A real scan on real hardware.** A face, scanned with the depth camera inside a Prowise MOVE 3in1 (an Intel RealSense D415) and Scantastic 0.5.1 on 11 October 2026: 31 of 32 frames merged. Its owner called it &amp;quot;a lot better, still gloopy&amp;quot;. Most of this paper is about what happened next." width="800" height="600"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  The short version
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;The goal:&lt;/strong&gt; a free desktop app that turns a cheap depth camera into a tabletop 3D scanner. The cameras are an Intel
RealSense SR305 and the D415 inside a Prowise MOVE 3in1, both second-hand, one with a scratched window. The computer
is anything down to a 7th-generation Core i3 with no graphics card. The app is called &lt;strong&gt;Scantastic&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The hard part isn't the camera, it's lining the frames up.&lt;/strong&gt; A scan is dozens of snapshots taken from different
places. Put one snapshot a few degrees out and the object grows a second nose. Over 90% of the time a build takes is
spent lining frames up.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;What this paper describes&lt;/strong&gt; (in enough detail to rebuild it):

&lt;ul&gt;
&lt;li&gt;picture features (the patterns on the object and the room) used to &lt;strong&gt;check every step&lt;/strong&gt; the 3D shape suggests;&lt;/li&gt;
&lt;li&gt;a rule that only lets a &lt;strong&gt;loop closure&lt;/strong&gt; move a frame as far as the chain could really have drifted;&lt;/li&gt;
&lt;li&gt;a &lt;strong&gt;turntable mode&lt;/strong&gt; that holds every frame to one fitted axis, with an optional printed marker ring;&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;live tracking&lt;/strong&gt; that says "TRACKING LOST" before you waste a scan;&lt;/li&gt;
&lt;li&gt;a &lt;strong&gt;"record now, stitch afterwards"&lt;/strong&gt; style for slow PCs, where the slow part runs in its own process with a memory
budget and simply tries again on fewer frames if it runs out;&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;speed profiles&lt;/strong&gt;, alignments &lt;strong&gt;side by side&lt;/strong&gt; in helper processes, and an &lt;strong&gt;exactly-the-same-answer shortcut&lt;/strong&gt;
for the final polish;&lt;/li&gt;
&lt;li&gt;a &lt;strong&gt;sensor check&lt;/strong&gt; that points the camera at a wall and maps where the window is scratched.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;How well it works:&lt;/strong&gt; on made-up scans with known answers, the version before this work failed every handheld and
turntable scan. Now 0 of 40 turntable scans fail, and 1 of 20 held-out handheld scans does. &lt;strong&gt;No real camera has
been through the new methods yet.&lt;/strong&gt; Section 8 is a test plan.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Why a paper:&lt;/strong&gt; it's a &lt;strong&gt;defensive publication&lt;/strong&gt;. Section 9 describes the methods and many variations of them and
dedicates them to the public, so that they are prior art and nobody can easily patent them and charge hobbyists for
them later.&lt;/li&gt;
&lt;/ul&gt;

&lt;h1&gt;
  
  
  1. Why this exists
&lt;/h1&gt;

&lt;p&gt;Proper 3D scanners cost hundreds of pounds. Second-hand depth cameras turn up in odd places: the RealSense SR305 was&lt;br&gt;
sold for face login and gesture control, and a D415 sits inside a classroom document camera (the Prowise MOVE 3in1).&lt;br&gt;
They see depth, 30 times a second, for the price of a takeaway.&lt;/p&gt;

&lt;p&gt;The catch is everything else:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;The software that came with them was made for faces and gestures, not scanning. Intel's own SDK is a toolbox, not a
scanner.&lt;/li&gt;
&lt;li&gt;The cameras are noisy. One of them has a scratched window that makes part of the view flicker.&lt;/li&gt;
&lt;li&gt;The computer it has to run on is a 7th-gen i3 laptop: 2 cores, 4 threads, 8 GB, no graphics card worth the name.&lt;/li&gt;
&lt;li&gt;The existing free scanning software that does this well (SLAM systems such as ORB-SLAM3) is under the GPL. Scantastic
is under a noncommercial licence (free to use, not to sell), and the two licences can't be mixed (section 3).&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;I'm working on about 50 programming projects at the same time, so I don't have the time or the brain capacity for&lt;br&gt;
humour. Each joke cuts into my four hours of sleep-eating time (efficiency). This paper will therefore keep the jokes&lt;br&gt;
to a minimum and the pictures to a maximum.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fd3lvl2xa407ka3lotthp.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fd3lvl2xa407ka3lotthp.jpg" alt="**The first live picture**, 1 October 2026: the colour view and, beside it, the depth the camera sees, with a &amp;quot;nothing in the middle of the view&amp;quot; warning. Everything after this started here." width="800" height="1067"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h1&gt;
  
  
  2. How a depth-camera scanner works, in two minutes
&lt;/h1&gt;

&lt;p&gt;A depth camera gives you a picture where each pixel says &lt;strong&gt;how far away&lt;/strong&gt; that bit of the world is, as well as&lt;br&gt;
&lt;strong&gt;what colour&lt;/strong&gt; it is. One such picture is a &lt;strong&gt;frame&lt;/strong&gt;: a patch of 3D surface seen from one side.&lt;/p&gt;

&lt;p&gt;To scan a whole object you take frames from all round it (moving the camera, or turning the object), then:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Clean each frame.&lt;/strong&gt; Throw away the unreliable pixels: the "flying pixels" along outlines, readings too near or too
far, and the room behind the object.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Line the frames up.&lt;/strong&gt; Work out where the camera was for each frame, so all the patches land in the same place.
This is the hard part.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Fuse them.&lt;/strong&gt; Every frame "votes" on where the surface is. The noise averages out and one smooth surface is left.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Save a mesh&lt;/strong&gt; (STL, OBJ or PLY) for a slicer, Blender or FreeCAD.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ff3v4hxa6dqeo9s17wg3m.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ff3v4hxa6dqeo9s17wg3m.png" alt="**The whole pipeline.** The lining-up stages (purple, red, teal) are where almost all of the time goes, and where almost all of the clever bits in this paper live." width="800" height="271"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Lining up, in one paragraph.&lt;/strong&gt; The standard tool is &lt;strong&gt;ICP&lt;/strong&gt; (Iterative Closest Point): slide one frame over another&lt;br&gt;
until each point sits as close as possible to the other frame's surface. It's excellent when the frames start close&lt;br&gt;
together, and it's fooled by smooth or symmetric shapes. A ball fits a ball in any orientation. A plain box fits itself&lt;br&gt;
four ways round. You chain frame 1 to 2, 2 to 3 and so on, then add &lt;strong&gt;loop closures&lt;/strong&gt; (frame 20 also overlaps frame 1:&lt;br&gt;
tie them together) and let a &lt;strong&gt;pose graph&lt;/strong&gt; spread out the leftover error. Each wrong link in that chain is a doubled&lt;br&gt;
surface in the final model.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fbbvo3pdmrmq96wyzuz02.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fbbvo3pdmrmq96wyzuz02.png" alt="**Where the time goes**, measured on one made-up 24-frame handheld scan (Scantastic 0.11.0, a 2-core computer). Lining up is 111 seconds; making the mesh is 16." width="800" height="315"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h1&gt;
  
  
  3. What already exists
&lt;/h1&gt;

&lt;p&gt;This is a well-studied problem. What follows is a hobbyist's search, not a professional one, and &lt;strong&gt;not legal advice&lt;/strong&gt;.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;What&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;th&gt;Why Scantastic didn't just use it&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;KinectFusion (Newcombe et al., 2011) [1]&lt;/td&gt;
&lt;td&gt;Live tracking against a fused volume, on a graphics card&lt;/td&gt;
&lt;td&gt;Needs a GPU; the volume is the depth fusion Scantastic uses &lt;em&gt;after&lt;/em&gt; lining up&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;ElasticFusion, BundleFusion and similar [2]&lt;/td&gt;
&lt;td&gt;Dense live reconstruction with loop closure&lt;/td&gt;
&lt;td&gt;GPU-only; research or non-commercial licences&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;ORB-SLAM3 [3]&lt;/td&gt;
&lt;td&gt;Feature-based SLAM: keyframes, relocalisation, loop closing&lt;/td&gt;
&lt;td&gt;GPLv3, which can't be combined with a noncommercial licence. Its authors offer a paid commercial licence&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Open3D's reconstruction system (Choi, Zhou, Koltun, 2015) [4], colored ICP (Park et al., 2017) [5]&lt;/td&gt;
&lt;td&gt;Fragments, pose graphs, loop closures checked by geometry; ICP that uses colour as well as shape&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Used&lt;/strong&gt;: Open3D (MIT licence) is the engine underneath. The checks in section 4 are built on top of it&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Visual-plus-geometric loop-closure checks (e.g. Aalto, ICIP 2016) [6]&lt;/td&gt;
&lt;td&gt;Accept a loop closure only if image features and shape agree&lt;/td&gt;
&lt;td&gt;Same family of idea as section 4.5; the details differ&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Commercial turntable scanners and photogrammetry rigs&lt;/td&gt;
&lt;td&gt;Turntables, often with printed marker mats, to know each turn&lt;/td&gt;
&lt;td&gt;Closed software; the marker idea itself is old and well known&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Patents found with titles close to this work&lt;/td&gt;
&lt;td&gt;e.g. scanning an object turned in the hand (US 9,639,943 B1) [7]; power-efficient depth sensing for live reconstruction (US 9,443,350 B2) [8]; turntable scanner calibration (US 2003/0202691) [9]&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Not assessed.&lt;/strong&gt; Listed so readers can look. Nothing here is a claim about what they cover&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;strong&gt;The gap Scantastic fills&lt;/strong&gt; is not a new algorithm. It's a set of plain, checkable rules that make standard tools&lt;br&gt;
(ICP, ORB features, pose graphs, TSDF fusion) behave on &lt;strong&gt;bad cameras and slow computers&lt;/strong&gt;, in a program anyone can&lt;br&gt;
run by double-clicking. Section 9 writes those rules down as prior art.&lt;/p&gt;

&lt;h1&gt;
  
  
  4. The method, in enough detail to rebuild it
&lt;/h1&gt;

&lt;p&gt;All numbers are Scantastic's defaults (0.9.0 to 0.11.0). Distances are in metres at the camera unless they say&lt;br&gt;
otherwise.&lt;/p&gt;

&lt;h2&gt;
  
  
  4.1 Capturing a clean frame
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Depth confidence&lt;/strong&gt; can be set stricter for a worn or scratched window ("Strict (worn / scratched glass)").&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Edge trim.&lt;/strong&gt; Pixels next to a sudden depth jump are dropped: 2 pixels on the SR305, 5 on the D400s, whose outlines
are softer.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cut out the subject&lt;/strong&gt; (on by default). Keep the surface nearest the camera in the middle of the view, plus
everything joined to it without a depth jump, within a slab of depth. The room goes. &lt;em&gt;Why:&lt;/em&gt; without this, a still
room dominates the alignment. Turn yourself in front of a fixed camera and the room "doesn't move", so the frames get
lined up to the room and the face comes out doubled. (This is the "second nose" in section 6.)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Average when still.&lt;/strong&gt; If the camera has hardly moved (under 1 pixel of shift and 1.5 mm of depth change), up to 5
frames are averaged into one capture, pixel by pixel. A pixel from an earlier frame is only used where it agrees.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Speed hints&lt;/strong&gt; on the live picture: "Steady", "Moving quickly" (over 150 pixels a second at 640 wide) and "Too fast -
slow down" (over 350). Coded-light cameras smear when you rush them.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Distance colours.&lt;/strong&gt; The live depth picture is green in the best band (30 to 50 cm for the SR305, 35 to 65 cm for
the D400s) and fades to red outside it.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  4.2 Picture features: a second opinion on every step
&lt;/h2&gt;

&lt;p&gt;The 3D shape can't tell a ball from a rotated ball, but the &lt;strong&gt;pattern on it&lt;/strong&gt; can. So every frame also gets:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;up to &lt;strong&gt;1,500 ORB keypoints&lt;/strong&gt; (corners in the picture), found on a contrast-boosted greyscale copy (CLAHE);&lt;/li&gt;
&lt;li&gt;keypoints kept only where there's depth (in turntable mode, also at least 6 pixels inside the depth outline:
outline points sit half on the still background and say "nothing moved");&lt;/li&gt;
&lt;li&gt;each keypoint &lt;strong&gt;lifted into 3D&lt;/strong&gt; with its depth.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;For two frames, keypoints are matched with &lt;strong&gt;Lowe's ratio test at 0.9&lt;/strong&gt; (0.8 kept too few right matches on plain&lt;br&gt;
objects), then a &lt;strong&gt;RANSAC rigid fit&lt;/strong&gt;: the 3D motion that the most matches agree with, to within 6 mm. That gives a&lt;br&gt;
camera move with a &lt;strong&gt;confidence&lt;/strong&gt; that rises from 0 at 12 agreeing matches to 1.0 at 52.&lt;/p&gt;

&lt;p&gt;That estimate is the second opinion. It's used three ways:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;As &lt;strong&gt;one more starting guess&lt;/strong&gt; for ICP.&lt;/li&gt;
&lt;li&gt;As a &lt;strong&gt;veto&lt;/strong&gt;: if the pictures are confident (0.5 or more) and ICP's answer is more than &lt;strong&gt;2.5° or 12 mm&lt;/strong&gt; away from
theirs, ICP is wrong (a smooth surface fitting itself in the wrong place), and the frame is placed where the pictures
say, then fine-tuned from there.&lt;/li&gt;
&lt;li&gt;As a &lt;strong&gt;vote&lt;/strong&gt; in loop closures (section 4.5).&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  4.3 Chaining frames, with guards
&lt;/h2&gt;

&lt;p&gt;Each new frame is aligned to the last frame that worked:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Starting guesses:&lt;/strong&gt; the tracker's or pictures' guess (if any), "the same move as last time", and "no move". ICP
(colour ICP, matching within 8.7 mm at the default quality) runs from each; the one with the best two-way overlap
wins. Two-way means checking that frame A's points find frame B's surface &lt;strong&gt;and&lt;/strong&gt; B's find A's: a frame full of
noise, or of the whole room, finds &lt;em&gt;something&lt;/em&gt; near almost every point one way, but not the other way round.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;If that's weak&lt;/strong&gt; (under 60% overlap), the slower &lt;strong&gt;feature-based global alignment&lt;/strong&gt; (FPFH + RANSAC) is tried too.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;If it's suspicious&lt;/strong&gt; (overlap under 60%, or the surfaces sit more than 1.6 times further apart than usual) &lt;strong&gt;and the
pictures don't vouch for the step&lt;/strong&gt;, the frame is also tried against 6 earlier frames spread across the scan. The
closest fit wins. This is how a quick swing back to the start re-attaches there.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;If nothing fits but the pictures are sure,&lt;/strong&gt; the picture estimate is used and fine-tuned.&lt;/li&gt;
&lt;li&gt;A frame that fits nothing (under 35% overlap) is &lt;strong&gt;skipped&lt;/strong&gt;, not glued on somewhere wrong.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  4.4 The room check, for plain objects
&lt;/h2&gt;

&lt;p&gt;A plain white box or mug gives the pictures almost nothing to match. But &lt;strong&gt;the room behind it doesn't move&lt;/strong&gt;. So in&lt;br&gt;
handheld mode, each frame also keeps a thinned (6 mm) copy of the whole view, before the subject was cut out. When the&lt;br&gt;
pictures can't vouch for a step:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;the two whole-room views are aligned coarse to fine (30, 15, then 8 mm), from each starting guess;&lt;/li&gt;
&lt;li&gt;a fit counts only with at least 60% overlap and under 4 mm average error;&lt;/li&gt;
&lt;li&gt;tables and walls are flat and slide over each other, so among the fits that pass, the one &lt;strong&gt;most picture matches
agree with&lt;/strong&gt; wins;&lt;/li&gt;
&lt;li&gt;if the room says the object's own fit is more than 2.5° or 12 mm out, &lt;strong&gt;and&lt;/strong&gt; at least as many picture matches side
with the room, the room wins.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;The same pictures catch a classic mistake.&lt;/strong&gt; If they say the camera hardly moved for most steps (under 1° and 3 mm)&lt;br&gt;
while the shape says it did, the object was turned by hand in front of a still camera. The pictures are right about the&lt;br&gt;
room and wrong about the object. They're not allowed to overrule, and the report suggests Turntable mode.&lt;/p&gt;

&lt;h2&gt;
  
  
  4.5 Loop closures on a leash
&lt;/h2&gt;

&lt;p&gt;Loop closures are where the 0.9.0 work found its worst errors. On a plain made-up blob, &lt;strong&gt;40 of 42 accepted closures&lt;br&gt;
were wrong&lt;/strong&gt;, by tens of degrees: a smooth surface happily fitting itself in the wrong place, with plenty of overlap.&lt;/p&gt;

&lt;p&gt;The candidates are: frames 2 to 4 steps apart, every frame against the first, and up to 4 earlier frames seen from&lt;br&gt;
within 30° and 20 cm (a "revisit"). Each is aligned (coarse plain ICP at 4 times the match distance, then colour ICP),&lt;br&gt;
and kept only with at least 50% two-way overlap and a surface error under 1.6 times the scan's typical step. Then:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Confirmed by the pictures?&lt;/strong&gt; At least 12 picture matches agree with the closure &lt;strong&gt;and&lt;/strong&gt; the pictures don't back
their own best estimate clearly better than this closure (0.10.0: a dozen chance matches had "confirmed" closures
18-24° out). Then it's in.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Not confirmed?&lt;/strong&gt; It may only move a frame as far as the chain could have drifted:

&lt;ul&gt;
&lt;li&gt;at most &lt;strong&gt;2° + 0.25° for every step in between that the pictures vouched for + 5° for every step they couldn't&lt;/strong&gt;;&lt;/li&gt;
&lt;li&gt;at most &lt;strong&gt;8 mm + 2 mm per vouched step + 30 mm per unvouched step&lt;/strong&gt;.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Either way&lt;/strong&gt;, if the two frames' own picture features disagree with the closure by more than 2.5° or 12 mm, and
either the tracker's or the pictures' estimate for that pair was confident, or the pictures clearly prefer their
own answer, it's out.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F8k7xmq1l5qllgpdvl1j6.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F8k7xmq1l5qllgpdvl1j6.png" alt="**The loop-closure rule.** Steps the pictures vouched for drift only a fraction of a degree each, so a closure that wants to move a frame by much more than that is a false fit." width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  4.6 Polish, and leaving out what moved
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Pose graph&lt;/strong&gt; (Open3D's Levenberg-Marquardt optimiser) spreads the leftover error over the chain.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Polish (handheld):&lt;/strong&gt; every frame is aligned again against the model made from all frames together, in two
passes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Frames that disagree&lt;/strong&gt; with their neighbours (median point-to-plane offset over 2 times the scan's typical offset,
and more than 0.8 mm) are left out, never more than 40% of them.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Areas that moved&lt;/strong&gt; (a blink, a mouth, a hand passing through, or a sensor defect that puts a bump at a fixed place in
the image) are cut out of just that frame: a patch at least 25 points big, over 3 times the typical offset and 3 mm
away, but under 35% of the frame. The other frames still cover it.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  4.7 Turntable mode: one axis to rule them all
&lt;/h2&gt;

&lt;p&gt;With the camera on a stand and the object on a turntable, &lt;strong&gt;every step is a turn about the same line&lt;/strong&gt;. Scantastic&lt;br&gt;
uses that hard:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Remove what stands still.&lt;/strong&gt; Either record the empty turntable first (any pixel within 8 mm of the empty scene is
background), or find the table plane and, on top of it, the &lt;strong&gt;turntable's top&lt;/strong&gt;: a flat layer parallel to the table
and wider than the object. Left in, the turntable top looks identical at every turn and pulls every step short.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Picture features on the object only.&lt;/strong&gt; Keypoints near the object's outline are dropped in turntable mode (they sit
on the still background and said "nothing moved").&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Fit the axis to the steps.&lt;/strong&gt; Each step's rotation axis is fitted to one line. A step that doesn't fit is tried
again at the angle the pictures or neighbouring steps suggest. &lt;strong&gt;Each frame's pose is then just its angle&lt;/strong&gt;: the sum
of the turns before it. One bad step can't bend the rest of the scan.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Fine-tune axis and turns together.&lt;/strong&gt; Each small step measures its turn well but its axis badly (a 12° turn about a
line 5 mm away looks almost the same), and on a made-up scan that grew to about 6 mm on the far side of the object. So the axis (4
numbers) and every frame's turn are fitted together by Gauss-Newton on point-to-plane distances, using each frame
and its 4 nearest frames within 45°, with each turn held near what its step measured.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Close the full turn.&lt;/strong&gt; When the scan goes all the way round, the last frames are also matched to the first and the
angles are solved again with those closures in, so a little error per step can't add up to a gap.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;If the steps don't fit one axis&lt;/strong&gt; (more than 3 mm from it), the scan is treated like a handheld one instead.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;No free polish.&lt;/strong&gt; The handheld polish lets each frame move freely in any direction. On turntable scans of plain
objects that undid the axis fit (0.6 mm became 3.5 mm), so turntable scans skip it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Optional marker ring.&lt;/strong&gt; A printed ring of ArUco markers lies on the turntable around the object. The object turns
with the ring, so the ring's pose in each frame gives the turn directly.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fmnb2npmwfjurck4veasg.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fmnb2npmwfjurck4veasg.png" alt="**Turntable mode from above.** Each frame sits on the fitted axis at the sum of the turns before it; a step that doesn't fit is re-tried." width="800" height="714"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fwr4hcnlmnq5zmm824r79.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fwr4hcnlmnq5zmm824r79.png" alt="**A marker ring of the kind Scantastic prints** (an illustration drawn for this paper with OpenCV's ArUco markers; the app's own ring has its own layout and a 50 mm check bar)." width="800" height="800"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  4.8 Live tracking: knowing you're lost while you can still do something about it
&lt;/h2&gt;

&lt;p&gt;Version 0.10.0 added the parts of SLAM that help here (keyframes, and finding the way back when lost), written from&lt;br&gt;
scratch on OpenCV's ORB features, so the licence stays the same:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Keyframes.&lt;/strong&gt; Each live picture is matched to the current keyframe, not the frame before, so small errors don't
pile up. A new keyframe is taken after 8°, 4 cm, or when fewer than 60 features still match. 40 recent keyframes are
kept.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Finding the way back.&lt;/strong&gt; When a picture can't be matched (camera covered, swung away), the tracker tries recent
keyframes, then a spread of older ones.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;TRACKING LOST&lt;/strong&gt; appears on the live picture while it's lost.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Each capture starts where the tracker said the camera was.&lt;/strong&gt; These guesses carry where they came from and which
&lt;strong&gt;track&lt;/strong&gt; (one START to STOP, or one recording) they belong to. Guesses from two different tracks are never compared:
the object may have been turned over in between. Live guesses may overrule ICP at 2.5° / 12 mm; an older recording's
chained guesses drift more and keep a looser 6° / 4 cm.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The camera thread only copies and queues.&lt;/strong&gt; Captures are worked out on their own thread, at most 2 waiting. If the
PC falls behind, a capture is skipped with a message rather than freezing the live picture.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fuz6m61mfkz96mdk7vz1l.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fuz6m61mfkz96mdk7vz1l.png" alt="**Keyframe tracking.** Blue arcs: each frame is matched to its keyframe. Red: the camera was covered; the tracker found its way back by matching an older keyframe." width="800" height="284"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fulonwilhmj2r8hf8cjrw.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fulonwilhmj2r8hf8cjrw.png" alt="**The real banner**, drawn by the app's own function over a made-up frame." width="640" height="480"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  4.9 Slow PCs: slow, but never falling over
&lt;/h2&gt;

&lt;p&gt;On a slow PC such as a 7th-gen i3, stitching while you scan should starve the camera: frames drop and the picture&lt;br&gt;
freezes. (Expected from timings on a 2-core test computer; not yet tried on a real i3.) So there are two &lt;strong&gt;scan&lt;br&gt;
styles&lt;/strong&gt;:&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0mu0v52cwuh64drk85w8.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0mu0v52cwuh64drk85w8.png" alt="**Two scan styles.** On a slow PC, scanning only records; the heavy work runs afterwards in its own process." width="800" height="327"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;The speed check.&lt;/strong&gt; The first time it starts, the app times one tracking step on a made-up 640 × 480 picture. With
4 processor threads or fewer, or a step slower than 150 ms, it picks &lt;strong&gt;"Record now, stitch afterwards"&lt;/strong&gt;. You can
change it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Recording&lt;/strong&gt; saves the camera's frames to disk at most 10 a second, using the same cleaned depth a live capture
would get. The frame list (&lt;code&gt;frames.json&lt;/code&gt;) is rewritten every 10 frames (every 50 on long recordings), to a temporary
file that then replaces the old one, so a crash leaves a usable recording minus its last few frames.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Stitching&lt;/strong&gt; runs in a &lt;strong&gt;separate process&lt;/strong&gt; after STOP. It cleans the frames, follows the camera through all of them
with the live tracker, and keeps &lt;strong&gt;one frame per 6° or 3 cm&lt;/strong&gt; of movement. On a made-up 116-frame recording,
aligning every frame took 14 minutes on the 2-core test computer; the 48 frames worth keeping took 4.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Memory budget:&lt;/strong&gt; at most 30% of free memory at 5 MB a frame, never fewer than 40 or more than 300 frames. A longer
recording is thinned to every n-th frame to fit.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;If it dies&lt;/strong&gt;, only the stitching process dies. It's started again on every 2nd frame, twice at most. A process that
dies before doing any work isn't blamed on memory. The recording stays on disk whatever happens. Build Mesh turns
into Cancel while it runs, and the status bar shows the minutes left.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fbmb0qjqnfz6ahuoczvsw.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fbmb0qjqnfz6ahuoczvsw.png" alt="**Crash-proof stitching.** Left: how many frames a stitch may hold. Right: what happens if it still runs out." width="800" height="318"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fu6e5w5ec02z02t5tyem5.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fu6e5w5ec02z02t5tyem5.png" alt="**Thinning a recording by motion.** Frames that mostly repeat each other are skipped before the slow part." width="800" height="552"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  4.10 Speed profiles, side-by-side alignment and the graphics card
&lt;/h2&gt;

&lt;p&gt;Version 0.11.0 brought over the speed profiles from my STL-to-STEP converter, then measured what actually helps here.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F90qdy94owo8hmtkck9zg.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F90qdy94owo8hmtkck9zg.png" alt="**Speed profiles.** Auto picks one from the computer's cores and memory." width="800" height="259"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Priority.&lt;/strong&gt; Building and stitching run at &lt;strong&gt;low priority&lt;/strong&gt; on Potato ("nice 10" on Linux and macOS, "idle" on
Windows), so the computer should stay usable. It makes nothing faster; it stops the build hogging the computer.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Alignments side by side&lt;/strong&gt; (Moderate, and Extreme when no graphics card is in use: with a card, alignments queue on
it one at a time). The loop-closure alignments are independent of each other, and so are
the starting guesses of one chain step. Helper processes, each holding its own copy of the frames, run them while the
main build process works through them too. Answers are put back in the original order and the checks then run in that
order, so &lt;strong&gt;the result is the same&lt;/strong&gt; (largest pose difference measured: 3 × 10^−13^, which is Open3D's own last-digit
noise).

&lt;ul&gt;
&lt;li&gt;A helper ends when its pipe breaks or a watchdog sees the builder gone (Cancel).&lt;/li&gt;
&lt;li&gt;A helper that dies or fails a task only costs time: the builder does that task itself.&lt;/li&gt;
&lt;li&gt;The number of helpers is cut to fit in half the free memory (300 MB plus 1.5 times the frames each).&lt;/li&gt;
&lt;li&gt;Each process gets two Open3D threads per core's share.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Exactly-the-same polish, quicker.&lt;/strong&gt; Colour ICP works out a colour gradient for every point of the target, from its
neighbours within 2 times the match distance, &lt;strong&gt;on every call&lt;/strong&gt;. In the polish the target is the whole model, and each
frame only sees part of it. So only the model within &lt;strong&gt;5 times&lt;/strong&gt; the match distance of the frame is used. Points
further than 3 times can't change the answer. If the frame moved more than the spare 2 times, or the answer scores
differently against the whole model, it's redone on the whole model. Poses matched the uncropped run to 10^−14^; the
polish took 18 seconds instead of 24 on one made-up scan.

&lt;ul&gt;
&lt;li&gt;
&lt;em&gt;What the three margins are for:&lt;/em&gt; matches are only made within 1× of a point; each point's colour gradient
looks 2× further; so nothing beyond 3× can change the answer. 5× leaves 2× to spare for the frame moving.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Repeatable fallback.&lt;/strong&gt; The feature-based fallback alignment uses random sampling. It's now seeded, so the same
frames always give the same answer.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The graphics card, with a gate.&lt;/strong&gt; Open3D's newer "tensor" ICP and depth fusion can run on an Nvidia card (CUDA). The
standard Open3D for Linux already contains that part. Before a card is used, a separate process runs &lt;strong&gt;the actual
steps&lt;/strong&gt; (colour ICP between two small made-up surfaces, and a depth fusion) on the card and on the processor, and
they must agree to 0.5 mm. During a build, any card error (out of memory, a driver fault) switches the card off and
the processor redoes that step; an error is never mistaken for "these frames don't overlap".

&lt;ul&gt;
&lt;li&gt;Not used: Nvidia on Windows (Open3D's Windows CUDA build is a preview), Apple graphics (Open3D has no Metal
support), AMD and Intel graphics.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fdxd5hi2kbo2uz35ui5qj.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fdxd5hi2kbo2uz35ui5qj.png" alt="**Side by side.** The answers come back in the original order, so the result is the same as one process." width="800" height="341"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2aibphhkyvi4kvyqadzw.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2aibphhkyvi4kvyqadzw.png" alt="**The exactly-the-same crop.** Only the part of the model that can change the answer is used." width="800" height="423"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  4.11 The sensor check: mapping a scratched window
&lt;/h2&gt;

&lt;p&gt;Point the camera at a plain flat wall, hold still, and the app collects 30 frames. A flat wall seen by a pinhole camera&lt;br&gt;
has &lt;strong&gt;1/depth changing in a straight line across the picture&lt;/strong&gt;, so a smooth low-order fit of 1/depth is the "true" wall.&lt;br&gt;
That fit also absorbs lens distortion and a slightly bowed wall. (The fit is repeated with the worst pixels left out;&lt;br&gt;
the written breakdown in the report uses an even more outlier-proof fit from 2,000 random trials.) A scratch then&lt;br&gt;
stands out from it in one of three ways. These are the limits for pixels that get &lt;strong&gt;ignored while scanning&lt;/strong&gt;:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Fault&lt;/th&gt;
&lt;th&gt;Ignored when (scaled to 40 cm)&lt;/th&gt;
&lt;th&gt;Shown as&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Flicker&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;a pixel's noise is over 2.5 times the typical noise, and over 1.5 mm&lt;/td&gt;
&lt;td&gt;red on the noise map&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Steady offset&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;a pixel is off the wall by over 5 times the typical spread, and over 2 mm&lt;/td&gt;
&lt;td&gt;red (too far) or blue (too near) on the offset map&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;No depth&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;enclosed holes up to 5% of the view, and pixels that lose their depth more than a quarter of the time&lt;/td&gt;
&lt;td&gt;purple&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The report's written diagnosis is more sensitive: it already counts an offset of 3 times the typical spread.&lt;/p&gt;

&lt;p&gt;The marked pixels can then be &lt;strong&gt;ignored while scanning&lt;/strong&gt;, all of them or only the no-depth ones, and shown live in&lt;br&gt;
magenta.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ftnv5tnsz71vod8rktwl1.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ftnv5tnsz71vod8rktwl1.jpg" alt="**A real sensor check**, on the D415 at 50 cm (11 October 2026): &amp;quot;noticeable damage&amp;quot;, 8.1% of the view to ignore. Left, flicker; right, steady offset." width="800" height="486"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fcqv1tsj8er9jk7mazmft.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fcqv1tsj8er9jk7mazmft.jpg" alt="**The same check on a made-up wall** with one scratch (flicker), one dent (steady offset) and one hole, as drawn by the app." width="800" height="749"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  4.12 From frames to a mesh
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Smooth surface (default):&lt;/strong&gt; TSDF fusion in an Open3D VoxelBlockGrid, sized from the scan itself because it reserves
its memory up front. On the SR305 the grid follows the Quality slider (2.5 mm at Low, about 1.5 mm in the middle,
1 mm at High) and is smoothed with 5 Taubin passes; the D400s start at a fixed 1 mm grid and 2 passes. Surface seen in
only one frame is dropped when there are enough frames.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Watertight:&lt;/strong&gt; Poisson reconstruction, trimmed back to within 2.5 grid cells of a scanned point.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The 3D preview&lt;/strong&gt; is coloured by how many frames saw each area, so thin spots show before you print.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cut and make solid:&lt;/strong&gt; draw a line across the model to cut it, or let the app find the table. "Make solid" closes an
open scan by inside-outside voting along 9 directions.&lt;/li&gt;
&lt;/ul&gt;

&lt;h1&gt;
  
  
  5. A tour of the app
&lt;/h1&gt;

&lt;p&gt;&lt;strong&gt;These screenshots come from the app itself, running on made-up data&lt;/strong&gt;, because the camera wasn't at the computer the&lt;br&gt;
paper was written on. They show version 0.10.0's window; 0.11.0 adds the Speed and graphics card box.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Forneedz6s3ab9tjr352g.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Forneedz6s3ab9tjr352g.jpg" alt="**The main window** in its default &amp;quot;retro TV&amp;quot; look: live picture on the left (static when no camera is plugged in), settings on the right, and two big buttons." width="800" height="646"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fa6cng1i8gc4c5m0md7x3.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fa6cng1i8gc4c5m0md7x3.jpg" alt="**A made-up scan, coloured by height**, with the view cube top right." width="800" height="646"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F46e7l45ixcqu24vym844.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F46e7l45ixcqu24vym844.jpg" alt="**Cutting:** &amp;quot;Find the table&amp;quot; found the flat base; the part that will go is red, and the yellow handle drags the cut." width="800" height="646"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Faq0krcaevjyepg6cap3r.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Faq0krcaevjyepg6cap3r.jpg" alt="**Turntable mode** in the Scan panel: record the empty turntable, save the marker ring, and the scan-style box." width="800" height="646"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0r2inlpqzjezm2qkgshn.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0r2inlpqzjezm2qkgshn.jpg" alt="**&amp;quot;Record now, stitch afterwards&amp;quot;**, chosen for a slow PC, with the honest expectations spelled out in the panel." width="800" height="646"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ferh3thggdcxh8ek3nxli.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ferh3thggdcxh8ek3nxli.jpg" alt="**The plain look**, for anyone who doesn't want a wooden television." width="800" height="687"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h1&gt;
  
  
  6. The mistakes, for free
&lt;/h1&gt;

&lt;p&gt;Every one of these happened. Most were caught by measuring, a few by a separate reviewer, and some by the owner of the&lt;br&gt;
camera.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;What went wrong&lt;/th&gt;
&lt;th&gt;What it looked like&lt;/th&gt;
&lt;th&gt;The fix&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;The room was in every frame&lt;/td&gt;
&lt;td&gt;A still room "won" the alignment, so turning in front of the camera gave a &lt;strong&gt;second nose&lt;/strong&gt;
&lt;/td&gt;
&lt;td&gt;Cut out the subject (4.1)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;A setting from an older Open3D&lt;/td&gt;
&lt;td&gt;The fallback alignment's RANSAC got a "confidence" of 500, so it never stopped early: 400,000 iterations, every time&lt;/td&gt;
&lt;td&gt;Proper (iterations, confidence) values&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Open3D 0.20's one-line hole filling&lt;/td&gt;
&lt;td&gt;Handed back vertices from memory that had already been freed: garbage in 19 of 30 tries&lt;/td&gt;
&lt;td&gt;Keep every step alive and copy the arrays out: 0 of 30&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;An outline trim made for the D415&lt;/td&gt;
&lt;td&gt;Silently cut 20% of the points from every SR305 frame&lt;/td&gt;
&lt;td&gt;Trim per camera (2 px SR305, 5 px D400)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Loop closures on smooth shapes&lt;/td&gt;
&lt;td&gt;40 of 42 accepted closures on the plain blob were tens of degrees out&lt;/td&gt;
&lt;td&gt;The drift rule (4.5)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Picture keypoints on the outline (turntable)&lt;/td&gt;
&lt;td&gt;Background corners said "nothing moved", so every turn came out short&lt;/td&gt;
&lt;td&gt;Drop outline keypoints, turntable only (it broke handheld scans when done there too)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;The turntable's own top&lt;/td&gt;
&lt;td&gt;Looks the same at every turn: pulled every step short&lt;/td&gt;
&lt;td&gt;Find it and remove it (4.7)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;The axis from small steps&lt;/td&gt;
&lt;td&gt;About 7 mm out, 6 mm of error on the far side&lt;/td&gt;
&lt;td&gt;Fit axis and turns together (4.7)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;The free polish on turntable scans&lt;/td&gt;
&lt;td&gt;0.6 mm became 3.5 mm on the plain scan target&lt;/td&gt;
&lt;td&gt;Turntable scans skip it&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;A dozen chance picture matches&lt;/td&gt;
&lt;td&gt;"Confirmed" loop closures 18-24° out&lt;/td&gt;
&lt;td&gt;Confirmed only if the pictures don't prefer their own answer (4.5)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;The older recording tracker&lt;/td&gt;
&lt;td&gt;One bad link put every later frame 0.7 m out, while reporting 0.65 confidence&lt;/td&gt;
&lt;td&gt;Stitching uses the keyframe tracker&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Aligning every recorded frame&lt;/td&gt;
&lt;td&gt;14 minutes for one 116-frame recording&lt;/td&gt;
&lt;td&gt;Keep one frame per 6° or 3 cm: 4 minutes&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Open3D 0.20 on Apple-silicon Macs&lt;/td&gt;
&lt;td&gt;Needed Homebrew's libusb: "Library not loaded"&lt;/td&gt;
&lt;td&gt;Macs use Open3D 0.19 (figure below)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;A graphics-card error&lt;/td&gt;
&lt;td&gt;Treated as "no overlap", so frames could be dropped silently&lt;/td&gt;
&lt;td&gt;Switch the card off, redo on the processor&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Helpers after Cancel&lt;/td&gt;
&lt;td&gt;A busy helper kept running&lt;/td&gt;
&lt;td&gt;A watchdog ends it within a second&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;A result file for 0.10.0&lt;/td&gt;
&lt;td&gt;Said 0 of 20 failed. Run again, the shipped code fails 1 of 20&lt;/td&gt;
&lt;td&gt;Corrected in 0.11.0's notes; the rerun is published next to it&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fthw6387b4wrdhvrcnntq.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fthw6387b4wrdhvrcnntq.jpg" alt="**Mistake, live:** the first real run on an M5 Mac (24 October 2026). Open3D 0.20 couldn't find a library that only Homebrew installs. Fixed the same day." width="800" height="600"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fwqm5jcsf8zse8yms0chy.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fwqm5jcsf8zse8yms0chy.jpg" alt="**Building the camera driver from source** on Linux (15 October 2026), before 0.8.0 switched to Intel's ready-made one. It took a while." width="800" height="600"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h1&gt;
  
  
  7. The numbers
&lt;/h1&gt;

&lt;p&gt;&lt;strong&gt;Everything in this section comes from made-up scans&lt;/strong&gt; with known answers (a benchmark that renders 5 objects with&lt;br&gt;
rich and plain patterns, with an SR305-like noise model, from known camera paths). A scan &lt;strong&gt;fails&lt;/strong&gt; if any kept frame&lt;br&gt;
lands more than 3 mm from where it should. Tuning used seeds 0-1; the held-out seeds 1000-1001 were only used to check&lt;br&gt;
finished code. Times are on a 2-core cloud computer.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fglingn4gk4tlqb09amwy.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fglingn4gk4tlqb09amwy.png" alt="**Results.** Left: handheld held-out scans that failed, by version. Right: how far out the worst frame typically was, on scans that passed." width="800" height="344"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Version, mode&lt;/th&gt;
&lt;th&gt;Failed&lt;/th&gt;
&lt;th&gt;Typical worst frame&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;0.8.1 (before this work), handheld and turntable&lt;/td&gt;
&lt;td&gt;every scan&lt;/td&gt;
&lt;td&gt;—&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;0.9.0 turntable (tuning + held-out)&lt;/td&gt;
&lt;td&gt;0 of 40&lt;/td&gt;
&lt;td&gt;0.6 mm&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;0.9.0 handheld, held-out&lt;/td&gt;
&lt;td&gt;0 of 20 (as recorded then; not re-run)&lt;/td&gt;
&lt;td&gt;0.9 mm&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;0.10.0 live tracking (tuning + held-out)&lt;/td&gt;
&lt;td&gt;0 of 20 each&lt;/td&gt;
&lt;td&gt;1.0 mm&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;0.10.0 record now, stitch later, held-out&lt;/td&gt;
&lt;td&gt;2 of 20 (one stray frame each, 6-8 mm)&lt;/td&gt;
&lt;td&gt;0.9 mm&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;0.10.0 handheld held-out, re-run&lt;/td&gt;
&lt;td&gt;1 of 20 (the plain blob, 3.1 mm)&lt;/td&gt;
&lt;td&gt;0.93 mm&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;0.11.0 handheld held-out&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;1 of 20, the same scan, identical answers on all 20&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;0.93 mm&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;0.11.0, graphics-card code run on the processor&lt;/td&gt;
&lt;td&gt;1 of 20 (the same scan)&lt;/td&gt;
&lt;td&gt;0.88 mm&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;strong&gt;Speed, honestly.&lt;/strong&gt; On a 7th-gen i3, expect &lt;strong&gt;several minutes&lt;/strong&gt; from STOP to a mesh (estimated from the 2-core test&lt;br&gt;
computer, which is in the same range; not timed on a real i3). The 0.11.0 speed-ups are small&lt;br&gt;
on 2 cores: the polish about a quarter quicker, loop closures about 10% quicker with one helper. More cores and a&lt;br&gt;
graphics card should help more, but &lt;strong&gt;neither has been measured&lt;/strong&gt;.&lt;/p&gt;

&lt;h1&gt;
  
  
  8. A test plan, for anyone with a depth camera
&lt;/h1&gt;

&lt;p&gt;If you have an SR305, SR300, a D400-series camera or a Prowise MOVE 3in1, please try these and report back (an issue&lt;br&gt;
or discussion on the repository, with photos):&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Sensor check&lt;/strong&gt; on a plain wall at 40-50 cm. Post both maps and the percentage.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Scan target.&lt;/strong&gt; Print the app's 70 mm test block, scan it in Turntable mode, and use "Compare mesh with scan target".
It reports the error in millimetres.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Turntable, with and without the printed ring&lt;/strong&gt;, on a plain object (a mug) and a patterned one.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Handheld, both scan styles&lt;/strong&gt;, on the same object. Note the time from STOP to mesh and whether TRACKING LOST
appeared.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A slow PC.&lt;/strong&gt; On a 7th-gen i3 or similar: does "Record now, stitch afterwards" get picked, does it finish, and does
the computer stay usable while it works?&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A many-core PC or an Nvidia card on Linux:&lt;/strong&gt; Potato against Extreme, same scan, time to mesh.&lt;/li&gt;
&lt;/ol&gt;

&lt;h1&gt;
  
  
  9. Defensive disclosure
&lt;/h1&gt;

&lt;p&gt;This section is written to be read as prior art. It describes methods implemented in Scantastic (versions 0.9.0 to&lt;br&gt;
0.11.0, source published at the repository below), and variations of them. &lt;strong&gt;All of them are dedicated to the&lt;br&gt;
public.&lt;/strong&gt; Anyone may use them, free, for any purpose. The purpose of publishing them is to make it hard for anyone to&lt;br&gt;
patent them and restrict their use. Where a number is given, any other value that does the same job is also disclosed.&lt;br&gt;
&lt;strong&gt;Not legal advice.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;D1. Choosing between live stitching and record-then-stitch by measuring the computer.&lt;/strong&gt; On first start (or on&lt;br&gt;
demand) the software times a representative step of its own live processing (feature extraction and matching on a&lt;br&gt;
synthetic image; or any other step of the real pipeline) and reads the processor thread count (or, as a variation,&lt;br&gt;
memory too). Below a threshold it selects a mode where scanning only records frames (optionally pre-cleaned, rate-limited) and all&lt;br&gt;
alignment runs afterwards. The choice is shown to the user with an expected-time warning, and the user can override it.&lt;br&gt;
Variations: re-measuring periodically or while scanning; switching modes mid-scan when frames start to drop;&lt;br&gt;
thresholds on battery state, temperature or other load; recording at reduced resolution; recording only frames that&lt;br&gt;
pass a motion or sharpness test.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;D2. Crash-tolerant deferred stitching.&lt;/strong&gt; The deferred alignment runs in a separate operating-system process with a&lt;br&gt;
frame budget derived from free memory (a share of it divided by a per-frame cost, clamped to limits). If the process&lt;br&gt;
dies, it is restarted on a subsampled set of frames (every 2nd, then every 4th; or any decreasing subset), a bounded&lt;br&gt;
number of times. A process that dies before reporting any progress is treated as a different failure from running out&lt;br&gt;
of memory. The recording's index is written atomically (to a temporary file, then renamed) at a decreasing frequency&lt;br&gt;
as the recording grows, so a crash loses only the last few frames. Variations: splitting the recording into&lt;br&gt;
overlapping chunks that are stitched and then joined; spilling to disk; any other subset rule (by motion, by time, by&lt;br&gt;
sharpness).&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;D3. Motion-based thinning before global alignment.&lt;/strong&gt; A lightweight tracker follows the camera through every&lt;br&gt;
recorded frame; only frames that have turned or moved more than a threshold (6° or 3 cm; or any other threshold,&lt;br&gt;
including one adapted to the object size or the overlap) from the last kept frame are passed to the expensive&lt;br&gt;
alignment, together with their tracked poses as starting guesses. On a turntable, a fixed number of frames spread&lt;br&gt;
evenly over the turn.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;D4. Track-scoped, source-weighted pose priors.&lt;/strong&gt; Each starting guess carries its source (live tracker, recording&lt;br&gt;
tracker, picture features, marker ring) and a track identifier. Guesses from different tracks are never compared or&lt;br&gt;
chained. The tolerance with which a guess may overrule a shape-based alignment depends on its source (tight for a&lt;br&gt;
keyframe tracker, loose for a drifting frame-to-frame tracker). A guess with low confidence is replaced, for checking&lt;br&gt;
purposes, by an independent pairwise estimate from the two frames' own picture features.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;D5. Drift-bounded loop-closure acceptance.&lt;/strong&gt; A loop closure between two frames is accepted without image-feature&lt;br&gt;
confirmation only if it moves the frame by no more than a bound that grows with the number of chain steps between the&lt;br&gt;
frames, with a small allowance per step that image features vouched for and a large allowance per step they couldn't&lt;br&gt;
(2° + 0.25°/vouched + 5°/unvouched, and 8 mm + 2 mm/vouched + 30 mm/unvouched; or any monotonic function of those&lt;br&gt;
counts, step lengths or step confidences). Image-feature confirmation requires both a minimum number of agreeing&lt;br&gt;
matches and that the features do not clearly prefer their own best estimate; and any closure the two frames' own&lt;br&gt;
features contradict is rejected. Variations: the same bound applied to chain steps or relocalisation; bounds expressed&lt;br&gt;
as covariance in the pose graph instead of accept/reject.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;D6. The still room as a referee.&lt;/strong&gt; In handheld scanning of an isolated object, a down-sampled copy of the whole&lt;br&gt;
scene (before the object is cut out) is kept per frame. Where image features can't vouch for a step, the scene copies&lt;br&gt;
are aligned coarse to fine from several guesses. Among fits that pass overlap and error thresholds, the one most&lt;br&gt;
image-feature matches agree with is chosen, and it overrules the object's own fit only if the features side with it at&lt;br&gt;
least as much. Also: when image features say the camera stood still while the object's shape says it moved, the&lt;br&gt;
software concludes the object was moved in front of a still camera, stops the features from overruling, and suggests a&lt;br&gt;
turntable mode.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;D7. Turntable poses as one axis plus angles.&lt;/strong&gt; With a still camera, every pairwise step is treated as a rotation&lt;br&gt;
about one common axis. The axis is fitted to all steps, off-axis steps are re-tried at an angle from image features or&lt;br&gt;
neighbouring steps, and each frame's pose is the cumulative angle. Axis (tilt and position) and all angles are then&lt;br&gt;
refined jointly by least squares over pairs of nearby frames (each frame's nearest few within 45°; or any other choice of pairs), together with full-turn closures where the scan goes all the way round, with each angle held near its measured&lt;br&gt;
step. A free 6-degree-of-freedom refinement is skipped in this mode, and if the steps don't fit one axis the scan falls back to free handheld-style alignment. Static scenery is removed by an empty-scene&lt;br&gt;
snapshot or by detecting the table plane and a turntable top (a flat layer parallel to and above the table, wider than&lt;br&gt;
the object). Image keypoints near the object outline are excluded in this mode only. Variations: a printed marker ring&lt;br&gt;
giving angles directly (ArUco, AprilTag, coded dots, a printed angle scale); a motorised turntable reporting angles; any&lt;br&gt;
mix of these as priors.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;D8. Exact cropping of the target in colour ICP.&lt;/strong&gt; When aligning a frame to a much larger target (a fused model, a&lt;br&gt;
previous scan), the target is cropped to the region within a margin of the frame's starting position, where the margin&lt;br&gt;
is at least the colour-gradient neighbourhood radius plus the match distance plus an allowance for movement (5 times the&lt;br&gt;
match distance in the implementation). The result is accepted only if the frame moved less than the allowance and its&lt;br&gt;
fitness and error, re-scored against the whole target, are unchanged; otherwise the alignment is redone on the whole&lt;br&gt;
target. Variations: caching target gradients instead; tiling the target; any margin derived the same way for other&lt;br&gt;
neighbourhood-based ICP variants.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;D9. Order-preserving parallel alignment with fail-safe helpers.&lt;/strong&gt; Independent alignment tasks (loop-closure&lt;br&gt;
candidates, multiple starting guesses for one step) are run in helper processes holding copies of the frames, with the&lt;br&gt;
main process also working, and answers re-ordered to the original task order so the result equals a single-process&lt;br&gt;
run. Helpers are tied to the main process by a pipe and a watchdog; a task that fails in a helper is run only by the&lt;br&gt;
main process thereafter; the helper count is limited by free memory; random number generators used in alignment are&lt;br&gt;
seeded per call so results don't depend on which process ran which task.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;D10. A graphics-card gate made of the real work.&lt;/strong&gt; Before an accelerator is used, a separate process runs the same&lt;br&gt;
alignment and fusion steps on it and on the processor with synthetic input, and the accelerator is enabled only if the&lt;br&gt;
answers agree within a tolerance. During use, any accelerator error disables it for the rest of the job and the&lt;br&gt;
affected step is redone on the processor; an accelerator error is never interpreted as a negative result (such as "no&lt;br&gt;
overlap"). The card list shown to the user gives, per card, whether it can be used and why not.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;D11. A flat-wall sensor map for depth cameras.&lt;/strong&gt; The camera is held still in front of a flat wall; a smooth low-order&lt;br&gt;
surface is fitted to 1/depth across the image, robustly; pixels are classified as flicker (temporal noise well above&lt;br&gt;
typical), steady offset (bias well above typical spread) or missing (enclosed holes), each with an absolute floor scaled&lt;br&gt;
to a reference distance. The classes are shown as separate maps and can be excluded from scanning, together or&lt;br&gt;
individually. Variations: other reference targets (a sphere, a printed pattern); repeating the check over time to track&lt;br&gt;
wear.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;D12. Live-scanning feedback.&lt;/strong&gt; A "tracking lost" banner on the live picture when a keyframe tracker can't place the&lt;br&gt;
current frame; speed warnings from image motion and depth change; distance colouring towards a best working band;&lt;br&gt;
automatic averaging of frames when the camera is still; and captures queued off the camera's thread, bounded, with a&lt;br&gt;
generation counter so cleared scans don't receive stale captures.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;D13. Per-frame removal of areas that moved.&lt;/strong&gt; After alignment, a coherent patch of one frame (a minimum point count&lt;br&gt;
and maximum share of the frame) that sits much further from the other frames than the scan's typical offset, while the&lt;br&gt;
rest of that frame agrees, is excluded from that frame only, and the corresponding pixels are excluded from fusion.&lt;/p&gt;

&lt;h1&gt;
  
  
  10. Limits, and what hasn't been tested
&lt;/h1&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;None of the 0.9.0 to 0.11.0 methods has been run on a real camera yet.&lt;/strong&gt; The real photos in this paper are from
earlier versions.&lt;/li&gt;
&lt;li&gt;All measured results are from made-up scans with a made-up noise model and texture.&lt;/li&gt;
&lt;li&gt;Not run: a real 7th-gen i3, a computer with more than 2 cores, any graphics card, a real out-of-memory death of the
stitching process, the Windows launcher, a printed marker ring, a printed scan target.&lt;/li&gt;
&lt;li&gt;1 of 20 held-out handheld scans still fails (a plain blob), and record-now-stitch-later had 2 of 20 with a stray
frame.&lt;/li&gt;
&lt;/ul&gt;

&lt;h1&gt;
  
  
  11. Credits and how this was made
&lt;/h1&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Built on:&lt;/strong&gt; Open3D (MIT) [10], OpenCV (Apache-2.0) [11], Intel's librealsense (Apache-2.0), NumPy, Pillow and
Python's tkinter.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Written with help from Claude&lt;/strong&gt; (Anthropic's AI assistant), which wrote much of the code, ran the benchmarks and
drafted this paper. The camera, the scratches, the bug reports from real hardware and the decisions are mine. Every
number above came from code that was run, and the made-up ones say so.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Not affiliated&lt;/strong&gt; with Intel, Prowise, or anyone else named here. RealSense is a trademark of Intel.&lt;/li&gt;
&lt;li&gt;Everyone who sends a test result, a correction or prior art I missed will be credited in the next version.&lt;/li&gt;
&lt;/ul&gt;

&lt;h1&gt;
  
  
  References
&lt;/h1&gt;

&lt;ol&gt;
&lt;li&gt;R. A. Newcombe et al., "KinectFusion: Real-time dense surface mapping and tracking", ISMAR 2011.&lt;/li&gt;
&lt;li&gt;T. Whelan et al., "ElasticFusion: Dense SLAM without a pose graph", RSS 2015; A. Dai et al., "BundleFusion", ACM TOG 2017.&lt;/li&gt;
&lt;li&gt;C. Campos et al., "ORB-SLAM3", IEEE T-RO 2021. &lt;a href="https://github.com/UZ-SLAMLab/ORB_SLAM3" rel="noopener noreferrer"&gt;https://github.com/UZ-SLAMLab/ORB_SLAM3&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;S. Choi, Q.-Y. Zhou, V. Koltun, "Robust Reconstruction of Indoor Scenes", CVPR 2015. &lt;a href="https://openaccess.thecvf.com/content_cvpr_2015/html/Choi_Robust_Reconstruction_of_2015_CVPR_paper.html" rel="noopener noreferrer"&gt;https://openaccess.thecvf.com/content_cvpr_2015/html/Choi_Robust_Reconstruction_of_2015_CVPR_paper.html&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;J. Park, Q.-Y. Zhou, V. Koltun, "Colored Point Cloud Registration Revisited", ICCV 2017.&lt;/li&gt;
&lt;li&gt;"Robust loop closures for scene reconstruction by combining odometry and visual correspondences", ICIP 2016. &lt;a href="https://users.aalto.fi/%7Ekannalj1/publications/icip2016b.pdf" rel="noopener noreferrer"&gt;https://users.aalto.fi/~kannalj1/publications/icip2016b.pdf&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;US 9,639,943 B1, "Scanning of a handheld object for 3-dimensional reconstruction". &lt;a href="https://patents.google.com/patent/US9639943" rel="noopener noreferrer"&gt;https://patents.google.com/patent/US9639943&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;US 9,443,350 B2, "Real-time 3D reconstruction with power efficient depth sensor usage". &lt;a href="https://patents.google.com/patent/US9443350" rel="noopener noreferrer"&gt;https://patents.google.com/patent/US9443350&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;US 2003/0202691, "Calibration of multiple cameras for a turntable-based 3D scanner". &lt;a href="https://patents.justia.com/patent/20030202691" rel="noopener noreferrer"&gt;https://patents.justia.com/patent/20030202691&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Q.-Y. Zhou, J. Park, V. Koltun, "Open3D: A modern library for 3D data processing", 2018. &lt;a href="https://www.open3d.org" rel="noopener noreferrer"&gt;https://www.open3d.org&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;E. Rublee et al., "ORB: an efficient alternative to SIFT or SURF", ICCV 2011; S. Garrido-Jurado et al., "Automatic generation and detection of highly reliable fiducial markers under occlusion" (ArUco), Pattern Recognition 2014.&lt;/li&gt;
&lt;/ol&gt;




&lt;p&gt;&lt;strong&gt;How to cite:&lt;/strong&gt; Armstrong, M. (2026). &lt;em&gt;Icy Ewe: 3D scanning on the cheap. A hobbyist's field guide and a defensive&lt;br&gt;
publication.&lt;/em&gt; Version 1.1, 13 November 2026 (first published as "Depth Perception", 2 November 2026).&lt;br&gt;
&lt;a href="https://github.com/The-Dorkknight/icy-ewe" rel="noopener noreferrer"&gt;https://github.com/The-Dorkknight/icy-ewe&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Software:&lt;/strong&gt; Scantastic, &lt;a href="https://github.com/The-Dorkknight/scantastic" rel="noopener noreferrer"&gt;https://github.com/The-Dorkknight/scantastic&lt;/a&gt; (PolyForm Noncommercial 1.0.0;&lt;br&gt;
source-available).&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Licence:&lt;/strong&gt; this paper's text and figures are &lt;a href="https://creativecommons.org/licenses/by-sa/4.0/" rel="noopener noreferrer"&gt;CC BY-SA 4.0&lt;/a&gt;. The&lt;br&gt;
methods in section 9 are dedicated to the public.&lt;/p&gt;

</description>
      <category>3dscanning</category>
      <category>python</category>
      <category>opensource</category>
      <category>computervision</category>
    </item>
    <item>
      <title>Handheld Ops</title>
      <dc:creator>Matthew Armstrong</dc:creator>
      <pubDate>Sat, 14 Nov 2026 12:21:12 +0000</pubDate>
      <link>https://dev.to/open-madscience/handheld-ops-52p9</link>
      <guid>https://dev.to/open-madscience/handheld-ops-52p9</guid>
      <description>&lt;p&gt;Copyright Matthew Armstrong. Source-available under the&lt;br&gt;
&lt;a href="https://github.com/The-Dorkknight/handheld-ops/blob/main/LICENSE.md" rel="noopener noreferrer"&gt;PolyForm Noncommercial License 1.0.0&lt;/a&gt;: free to use and change for&lt;br&gt;
non-commercial purposes. Provided as is, with &lt;strong&gt;no warranty and no liability&lt;/strong&gt;.&lt;br&gt;
Not affiliated with Activision, Infinity Ward, Treyarch, Valve, ASUS, the&lt;br&gt;
Bazzite project or any of the mod teams below.&lt;/p&gt;

&lt;p&gt;One tap-friendly app that gets three classic shooters ready for handheld PCs&lt;br&gt;
(ROG Ally, Steam Deck, Legion Go) on &lt;strong&gt;Bazzite&lt;/strong&gt; or SteamOS (and Windows):&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Game&lt;/th&gt;
&lt;th&gt;Campaign&lt;/th&gt;
&lt;th&gt;Multiplayer&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Call of Duty: Black Ops (2010)&lt;/td&gt;
&lt;td&gt;Hidden gamepad mode switched on&lt;/td&gt;
&lt;td&gt;Same, plus Plutonium's configs&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Call of Duty 4: Modern Warfare (2007)&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;IW3SP-MOD&lt;/strong&gt; installed: gamepad + aim assist&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;CoD4x&lt;/strong&gt;: steps and official page (assisted)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Call of Duty: Modern Warfare 2 (2009)&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;IW-Pad&lt;/strong&gt; installed: full controller support&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;IW4x&lt;/strong&gt; installed: native controller + aim assist&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Plus, for all three:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Steam launch options written for you&lt;/strong&gt; (with Steam closed), so the normal
Steam Play button starts the modded game under Proton. Your own launch
options are never overwritten without asking.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Full HD locked at every start&lt;/strong&gt; (or 1440p / 4K): set in the game's config
files and again on the command line, so a game that falls back to "safe" or
"recommended" settings (MW2's 640-pixel habit) comes straight back.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Five one-tap graphics tiers&lt;/strong&gt; per game (Off, Battery saver, Balanced,
Quality, Ultra): the game's own settings plus a matching SweetFX preset, so
you never have to dig through the effects overlay.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Look&lt;/strong&gt;: ReShade/SweetFX-style sharpening and colour through vkBasalt on
Linux, switched on per game.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Touchscreen&lt;/strong&gt; help for the menus.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Every replaced game file is backed up and &lt;strong&gt;Remove&lt;/strong&gt; puts it back.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F8kjhxh09kn7psm6qo18f.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F8kjhxh09kn7psm6qo18f.webp" alt="MW2 page" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;h2&gt;
  
  
  Step by step
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Install the games in Steam and start each once (Black Ops needs that to
write its config files).&lt;/li&gt;
&lt;li&gt;Open Handheld Ops and pick a game tab.&lt;/li&gt;
&lt;li&gt;Press &lt;strong&gt;Install&lt;/strong&gt; (or &lt;strong&gt;Controller ON&lt;/strong&gt; for Black Ops). Downloads come
straight from each mod's official release page; the log shows every step.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Quit Steam&lt;/strong&gt; (Steam menu &amp;gt; Exit), then press &lt;strong&gt;Write to Steam&lt;/strong&gt;. Or press
&lt;strong&gt;Copy&lt;/strong&gt; and paste each line into Steam &amp;gt; game &amp;gt; Properties &amp;gt; General &amp;gt;
Launch Options yourself.&lt;/li&gt;
&lt;li&gt;Start Steam again and play from your library as usual. In Steam Input, use a
&lt;strong&gt;Gamepad&lt;/strong&gt; layout for these games.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fdmdjxr7q2nzwhaph9aji.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fdmdjxr7q2nzwhaph9aji.webp" alt="CoD4 page" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;h3&gt;
  
  
  What the launch options look like
&lt;/h3&gt;


&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;bash -c 'exec "${@/iw4mp.exe/iw4x.exe}"' -- %command%
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;


&lt;p&gt;Steam fills in &lt;code&gt;%command%&lt;/code&gt; with its own Proton command; bash swaps the stock&lt;br&gt;
.exe for the mod's. The same form is documented by IW-Pad. With the look on,&lt;br&gt;
&lt;code&gt;ENABLE_VKBASALT=1 VKBASALT_CONFIG_FILE="..."&lt;/code&gt; goes in front.&lt;/p&gt;
&lt;h3&gt;
  
  
  Graphics tiers
&lt;/h3&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fxkrddx9swtngvy0iiiy4.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fxkrddx9swtngvy0iiiy4.webp" alt="Graphics tiers" width="800" height="142"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Tap a tier on the game's page. It is applied straight away (quit the game&lt;br&gt;
first) and again whenever you press &lt;strong&gt;Write to Steam&lt;/strong&gt;.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Tier&lt;/th&gt;
&lt;th&gt;Game settings (only those the game already has)&lt;/th&gt;
&lt;th&gt;Effects&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Off&lt;/td&gt;
&lt;td&gt;left as they are&lt;/td&gt;
&lt;td&gt;none&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Battery saver&lt;/td&gt;
&lt;td&gt;60 fps cap, vsync off, AA off, shadows off, 4x texture filtering, depth of field off, textures one step lower&lt;/td&gt;
&lt;td&gt;light CAS sharpening&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Balanced (default)&lt;/td&gt;
&lt;td&gt;60 fps cap, vsync off, 2x AA, shadows on, 8x filtering, best textures&lt;/td&gt;
&lt;td&gt;SMAA + CAS sharpening&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Quality&lt;/td&gt;
&lt;td&gt;90 fps cap, vsync off, 4x AA, shadows on, 16x filtering, best textures&lt;/td&gt;
&lt;td&gt;SMAA + vibrance + luma sharpen&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Ultra (SweetFX)&lt;/td&gt;
&lt;td&gt;120 fps cap, vsync off, 4x AA, shadows on, 16x filtering, best textures&lt;/td&gt;
&lt;td&gt;SMAA + vibrance + contrast curve + luma sharpen&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Bazzite / Linux:&lt;/strong&gt; each game gets its own vkBasalt file; the Steam launch
option points at it. Quality and Ultra download the SweetFX shaders the first
time (automatically).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Windows:&lt;/strong&gt; press &lt;strong&gt;Install ReShade essentials&lt;/strong&gt; for the game on the
&lt;strong&gt;Look&lt;/strong&gt; tab. That installs ReShade without its long effect-package list,
adds only the 5 effects the tiers use (10 small files) and applies the tier.
ReShade's own list then shows just those 5, the start-up tutorial is off,
and &lt;strong&gt;Home&lt;/strong&gt; still opens it. Effects you installed before stay on disk,
hidden; &lt;strong&gt;Show my other effects&lt;/strong&gt; brings them back.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Already have ReShade?&lt;/strong&gt; It is never reinstalled or overwritten: the
button keeps your &lt;code&gt;d3d9.dll&lt;/code&gt;, your own presets and effect packages, and only
changes five lines of &lt;code&gt;ReShade.ini&lt;/code&gt; (which preset, the two shader folders,
skip unused effects, tutorial done). The original &lt;code&gt;ReShade.ini&lt;/code&gt; is saved
the first time; &lt;strong&gt;Restore my ReShade&lt;/strong&gt; puts it back exactly. A &lt;code&gt;d3d9.dll&lt;/code&gt;
that is not ReShade (another mod) is left alone and reported.&lt;/li&gt;
&lt;li&gt;The fps caps suit a 120 Hz handheld screen; in Steam Gaming Mode you can
also cap frames in the Quick Access menu. Leave the Steam scaling filter (FSR)
alone: with the Full HD lock the game already runs at the screen's resolution,
and FSR only helps when a game renders below it.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;
  
  
  Conflicting settings are removed
&lt;/h3&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3qc1y89u273eda4gnh2e.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3qc1y89u273eda4gnh2e.webp" alt="Conflict warning, then Put back" width="800" height="167"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Picking a tier (or Write to Steam) first clears anything that would undo it:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Extra game config files&lt;/strong&gt; such as an &lt;code&gt;autoexec.cfg&lt;/code&gt; in the game's
&lt;code&gt;players&lt;/code&gt; or &lt;code&gt;main&lt;/code&gt; folder: lines that set vsync, the fps cap,
anti-aliasing, shadows, texture settings or the resolution are turned off
(commented out with a &lt;code&gt;// handheld-ops off:&lt;/code&gt; marker, backup first). Other
lines, key binds, the game's own &lt;code&gt;config.cfg&lt;/code&gt; and files a mod installed are
left alone. The Graphics card warns before and shows &lt;strong&gt;Put back&lt;/strong&gt; after.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;ReShade settings that override a tier&lt;/strong&gt; (Windows): a start-up preset,
&lt;code&gt;NoReloadOnInit&lt;/code&gt;, preset-switch hotkeys and preset shortcut keys are set to
off; Handheld Ops' own presets for other tiers are removed. Your own preset
files are kept. &lt;strong&gt;Restore my ReShade&lt;/strong&gt; undoes everything in &lt;code&gt;ReShade.ini&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Old launch options&lt;/strong&gt; that contain only vkBasalt, an exe swap or the
resolution are replaced (and remembered for &lt;strong&gt;Put back&lt;/strong&gt;). Launch options
with anything else of yours still ask first.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;
  
  
  Aim feel (Black Ops)
&lt;/h3&gt;

&lt;p&gt;If aim feels floaty: every graphics tier turns &lt;strong&gt;vsync off&lt;/strong&gt;, the usual cause&lt;br&gt;
of input lag. Then check Steam Input for the game (Controller settings &amp;gt; right&lt;br&gt;
stick: smaller deadzone, Linear or Aggressive response curve). The Black Ops&lt;br&gt;
page also has an experimental &lt;strong&gt;Look speed&lt;/strong&gt; button that changes the game's&lt;br&gt;
&lt;code&gt;input_viewSensitivity&lt;/code&gt; (1, 1.25, 1.5, 2); it is not confirmed that the PC&lt;br&gt;
game uses that setting for the gamepad.&lt;/p&gt;
&lt;h3&gt;
  
  
  Screen resolution (Full HD as the minimum)
&lt;/h3&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fx3xed72o1n8whyh7rd7y.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fx3xed72o1n8whyh7rd7y.webp" alt="Screen resolution card" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;On by default for all three games at &lt;strong&gt;1920x1080&lt;/strong&gt;; tap to choose 2560x1440 or&lt;br&gt;
3840x2160, or switch it off. It is set in two places:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;The config files&lt;/strong&gt;: &lt;code&gt;r_mode&lt;/code&gt; in every engine config under the game's
&lt;code&gt;players&lt;/code&gt; folder (CoD4 keeps them in &lt;code&gt;players/profiles/&amp;lt;name&amp;gt;/&lt;/code&gt;), plus
&lt;code&gt;r_fullscreen 1&lt;/code&gt;, &lt;code&gt;r_aspectRatio "wide 16:9"&lt;/code&gt; and &lt;code&gt;com_recommendedSet 1&lt;/code&gt;
where the game already has those lines. &lt;code&gt;com_recommendedSet 1&lt;/code&gt; tells the
game its recommended settings were already applied, so it stops replacing
yours with auto-detected ones.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The launch option&lt;/strong&gt;: &lt;code&gt;+set r_mode 1920x1080 +set r_fullscreen 1&lt;/code&gt; after
&lt;code&gt;%command%&lt;/code&gt;. These engines apply command-line settings after reading the
config, so the resolution comes back even if the config was reset.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;If a game asks "Run in safe mode?" after a crash, answer &lt;strong&gt;No&lt;/strong&gt;. In Gaming&lt;br&gt;
Mode, also check the game's Steam properties for a Game Resolution option and&lt;br&gt;
choose Native.&lt;/p&gt;
&lt;h3&gt;
  
  
  Black Ops
&lt;/h3&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fj49itat7zuh07bi0hpax.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fj49itat7zuh07bi0hpax.webp" alt="Black Ops page" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Controller ON&lt;/strong&gt; sets the game's hidden &lt;code&gt;gpad_enabled&lt;/code&gt; in the Singleplayer/&lt;br&gt;
Zombies and Multiplayer configs (and Plutonium's copies). Optional aim&lt;br&gt;
slowdown (campaign/Zombies only, experimental, never Multiplayer).&lt;/p&gt;
&lt;h3&gt;
  
  
  CoD4x (CoD4 multiplayer)
&lt;/h3&gt;

&lt;p&gt;CoD4x comes with its own installer, and how the Steam copy behaves with it is&lt;br&gt;
unclear (see &lt;a href="https://github.com/The-Dorkknight/handheld-ops/blob/main/RESEARCH.md" rel="noopener noreferrer"&gt;RESEARCH.md&lt;/a&gt;), so the app gives the steps and opens&lt;br&gt;
the official release page instead of installing it.&lt;/p&gt;
&lt;h3&gt;
  
  
  Controller in IW3SP-MOD (CoD4 campaign)
&lt;/h3&gt;

&lt;p&gt;IW3SP-MOD's gamepad code comes from IW4x: it is switched on by&lt;br&gt;
&lt;code&gt;gpad_enabled&lt;/code&gt; (Handheld Ops adds &lt;code&gt;+set gpad_enabled 1&lt;/code&gt; at every start) and&lt;br&gt;
reads &lt;strong&gt;Xbox-style (XInput)&lt;/strong&gt; pads. If yours is a PlayStation, Switch or other&lt;br&gt;
pad, start the game &lt;strong&gt;from Steam&lt;/strong&gt; with Steam Input on (after Write to Steam),&lt;br&gt;
so Steam presents it as an Xbox pad. Starting &lt;code&gt;iw3sp_mod.exe&lt;/code&gt; outside Steam&lt;br&gt;
(an old shortcut, Explorer) gets no Steam Input. Fallback: in Steam, Games &amp;gt;&lt;br&gt;
Add a Non-Steam Game &amp;gt; &lt;code&gt;iw3sp_mod.exe&lt;/code&gt;, and play that entry.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;CoD4: stick works in menus but not in the game:&lt;/strong&gt; IW3SP-MOD stops reading&lt;br&gt;
the sticks for moving and looking whenever the mouse moves even a little&lt;br&gt;
(or a keyboard key is pressed). &lt;strong&gt;Mouse in the game&lt;/strong&gt; on the Controller card&lt;br&gt;
is set to Off for CoD4 by default (&lt;code&gt;+set in_mouse 0&lt;/code&gt;), so the pad stays in&lt;br&gt;
charge; the pad drives the menus. Tap &lt;strong&gt;Write to Steam&lt;/strong&gt; once. Use Steam&lt;br&gt;
Input's &lt;strong&gt;Xbox controller&lt;/strong&gt; layout.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;CoD4: left trigger latches aim on until pressed again:&lt;/strong&gt; that is the&lt;br&gt;
toggle-aim bind. Turn &lt;strong&gt;Controller override&lt;/strong&gt; on (Controller card), keep&lt;br&gt;
&lt;strong&gt;Aim: Hold to aim&lt;/strong&gt;, tap &lt;strong&gt;Write to Steam&lt;/strong&gt;. In the game console (&lt;code&gt;~&lt;/code&gt;) look&lt;br&gt;
for "Handheld Ops controller file loaded".&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Close Steam:&lt;/strong&gt; next to Write to Steam. Steam is asked to close cleanly and&lt;br&gt;
ended after 30 seconds if needed. Use it in Desktop Mode (in Gaming Mode&lt;br&gt;
Steam is the whole screen, so the app refuses).&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Controller override (CoD4 IW3SP-MOD, MW2 IW4x):&lt;/strong&gt; tick &lt;strong&gt;Controller&lt;br&gt;
override&lt;/strong&gt; on the game's page and Handheld Ops writes its own controller file&lt;br&gt;
(&lt;code&gt;handheld_pad.cfg&lt;/code&gt;, in the game's &lt;code&gt;main&lt;/code&gt; folder) and loads it at every start,&lt;br&gt;
so you set the bindings here and never touch the mod. Defaults are IW4x's&lt;br&gt;
own: hold to aim, normal sticks. Pick from the dropdowns: Button layout&lt;br&gt;
(Default, Tactical, Lefty, Nomad, Default Alt), Aim (hold or toggle), Sticks&lt;br&gt;
(Default = left moves, right looks; Southpaw swaps them), Look speed, and&lt;br&gt;
any single button. &lt;strong&gt;Put the defaults back&lt;/strong&gt; undoes it; &lt;strong&gt;I edit the&lt;br&gt;
file myself&lt;/strong&gt; keeps your hand-edited file. After tapping, press &lt;strong&gt;Write to&lt;br&gt;
Steam&lt;/strong&gt; once, then start from Steam. Proof it loaded: open the game console&lt;br&gt;
with &lt;code&gt;~&lt;/code&gt;, it prints "Handheld Ops controller file loaded". &lt;strong&gt;Controller&lt;br&gt;
report&lt;/strong&gt; lists what the game saved and what Handheld Ops will start with&lt;br&gt;
(also saved as a text file to send).&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Left trigger latches aim on until pressed again / look feels heavy:&lt;/strong&gt; the&lt;br&gt;
&lt;strong&gt;Controller sticks&lt;/strong&gt; card shows what the left trigger is bound to. &lt;strong&gt;Hold to&lt;br&gt;
aim fix&lt;/strong&gt; (on by default) changes a toggle bind to hold. &lt;strong&gt;Look speed&lt;/strong&gt; steps&lt;br&gt;
x1, x1.5, x2, x3. Tap &lt;strong&gt;Write to Steam&lt;/strong&gt;, then start from Steam.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Right stick works in menus but not in the game:&lt;/strong&gt; first set Steam Input to&lt;br&gt;
&lt;strong&gt;Xbox controller&lt;/strong&gt; (CoD4 &amp;gt; Controller settings &amp;gt; Browse configs / choose the&lt;br&gt;
Xbox controller layout). That is what fixed it on the Ally; the default&lt;br&gt;
"controller" layout did not. If it is still dead: the view turns by the&lt;br&gt;
stick times the game's turn speed settings, and 0 means no turning. The&lt;br&gt;
&lt;strong&gt;Controller sticks&lt;/strong&gt; card shows the saved values and &lt;strong&gt;Turn speed fix&lt;/strong&gt; (on by&lt;br&gt;
default) sets standard ones at every start. Tap &lt;strong&gt;Write to Steam&lt;/strong&gt;, then start&lt;br&gt;
from Steam. Still nothing: tell me what the card shows.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Right stick does nothing (left stick and buttons work):&lt;/strong&gt; the sticks are&lt;br&gt;
tied to moving and looking by &lt;code&gt;bindaxis&lt;/code&gt; lines the game keeps in its profile&lt;br&gt;
config (&lt;code&gt;players/profiles/&amp;lt;name&amp;gt;/config.cfg&lt;/code&gt;). A config without right-stick&lt;br&gt;
lines leaves the right stick dead. The &lt;strong&gt;Controller sticks&lt;/strong&gt; card on the CoD4&lt;br&gt;
page shows whether the right stick is bound, and &lt;strong&gt;Right stick fix&lt;/strong&gt; (on by&lt;br&gt;
default) adds IW4x's default lines for a stick with no binding at all (a&lt;br&gt;
custom layout such as southpaw is kept); switched &lt;strong&gt;off&lt;/strong&gt; it takes out&lt;br&gt;
exactly the lines it added. It runs on Write to Steam, Start, and the button&lt;br&gt;
itself:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;bindaxis A_LSTICK_X VA_SIDE MAP_SQUARED
bindaxis A_LSTICK_Y VA_FORWARD MAP_SQUARED
bindaxis A_RSTICK_X VA_YAW MAP_LINEAR
bindaxis A_RSTICK_Y VA_PITCH MAP_LINEAR
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;0.4.2 also put these four on the start command line (&lt;code&gt;+bindaxis ...&lt;/code&gt;), and&lt;br&gt;
with 0.4.2 the controller stopped working altogether in IW3SP-MOD. 0.4.3 takes&lt;br&gt;
them off the command line. &lt;strong&gt;Show last start log&lt;/strong&gt; (Windows) shows the exact&lt;br&gt;
command the game was started with.&lt;/p&gt;

&lt;h3&gt;
  
  
  Display: borderless window (Windows)
&lt;/h3&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fadjqr3eto8fw1ju9mgfh.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fadjqr3eto8fw1ju9mgfh.webp" alt="Display mode" width="800" height="101"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Switching away (Alt+Tab) from these DirectX 9 games in exclusive fullscreen&lt;br&gt;
can end in "DirectX encountered an unrecoverable error". On Windows the&lt;br&gt;
default for IW4x is a &lt;strong&gt;borderless window&lt;/strong&gt; at the locked resolution&lt;br&gt;
(&lt;code&gt;+set r_fullscreen 0 +set r_noborder 1 +set vid_xpos 0 +set vid_ypos 0&lt;/code&gt;),&lt;br&gt;
which switches away cleanly. Tap &lt;strong&gt;Display&lt;/strong&gt; to go back to fullscreen.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;IW3SP-MOD (CoD4 campaign) starts in fullscreen by default (0.4.2+).&lt;/strong&gt;&lt;br&gt;
Started borderless it opened as a 1280x720 window in the top-left corner&lt;br&gt;
whatever the locked resolution was. Borderless can still be picked for it&lt;br&gt;
(Display button) to test. If Alt+Tab crashes CoD4 in fullscreen, avoid&lt;br&gt;
switching away in game for now. Stock games keep fullscreen (they have no&lt;br&gt;
borderless mode).&lt;/p&gt;

&lt;h3&gt;
  
  
  Safety
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;MW2 campaign / Special Ops: play offline only.&lt;/strong&gt; IW-Pad warns that the
Steam build's singleplayer still has remote-code-execution holes. The app
shows this before installing.&lt;/li&gt;
&lt;li&gt;IW4x is about &lt;strong&gt;800 MB&lt;/strong&gt;; the app checks free space first.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The look (Linux)
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fya63gcy9ra7c6q2me2c3.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fya63gcy9ra7c6q2me2c3.webp" alt="Look tab" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Pick a graphics tier on each game's page (see Graphics tiers above).&lt;/li&gt;
&lt;li&gt;Press &lt;strong&gt;Write to Steam&lt;/strong&gt; so the Play button uses that game's vkBasalt file.&lt;/li&gt;
&lt;li&gt;In game, &lt;strong&gt;Home&lt;/strong&gt; switches the effects off and on.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;On Windows, use the normal &lt;a href="https://reshade.me/" rel="noopener noreferrer"&gt;ReShade&lt;/a&gt; installer (all three&lt;br&gt;
games are DirectX 9).&lt;/p&gt;

&lt;h2&gt;
  
  
  On Windows
&lt;/h2&gt;

&lt;p&gt;Handheld Ops runs on Windows too (tested so far only by pretending, on Linux;&lt;br&gt;
see Known limitations). The differences:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Steam's Play button starts the mod&lt;/strong&gt; through a small Handheld Ops
launcher (&lt;code&gt;"pythonw.exe" "...\steam-launch.pyw" %command%&lt;/code&gt;, no console
window). It starts the mod for the right mode (e.g. CoD4 Singleplayer -&amp;gt;
IW3SP-MOD) and the stock game otherwise (CoD4 Multiplayer). Because Steam
started it, &lt;strong&gt;Steam Input&lt;/strong&gt; can turn any controller into an Xbox pad, which
is what IW3SP-MOD and IW4x read. The desktop / Start menu shortcuts start
the mod directly (fine for Xbox-style pads).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The resolution lock still works from Steam:&lt;/strong&gt; "Write to Steam" (Steam
closed) sets each game's launch option to &lt;code&gt;+set r_mode 1920x1080 +set
r_fullscreen 1&lt;/code&gt;, which Steam adds to the game's command line. The Start
buttons and shortcuts pass the same, and the config files are set too.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Look:&lt;/strong&gt; vkBasalt is Linux-only, so the Look tab installs ReShade from
reshade.me with just the essentials (see Graphics tiers).&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3zn0j6mgx5qnelbal6xf.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3zn0j6mgx5qnelbal6xf.webp" alt="Look tab on Windows" width="800" height="188"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Touchscreen
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fzzy1hrtlucsobc8ym439.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fzzy1hrtlucsobc8ym439.webp" alt="Help tab" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Steam Gaming Mode:&lt;/strong&gt; a tap is a left mouse click where you touched.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;If taps do nothing:&lt;/strong&gt; hold the Armoury Crate button for Mouse Mode.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Desktop Mode:&lt;/strong&gt; the touchscreen is a normal mouse.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Command line
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;HandheldOps-Linux.sh --list
HandheldOps-Linux.sh --install iw3sp      # also: iwpad, iw4x, bo1-gamepad
HandheldOps-Linux.sh --remove iw4x
HandheldOps-Linux.sh --install bo1-gamepad --aim-slowdown
HandheldOps-Linux.sh --get-shaders --look sweetfx --look-for mw2
HandheldOps-Linux.sh --screen mw2=1920x1080       # or 2560x1440, 3840x2160, off
HandheldOps-Linux.sh --steam cod4 --steam mw2   # resolution + launch options (Steam closed)
HandheldOps-Linux.sh --tier mw2=quality       # off, battery, balanced, quality, ultra
HandheldOps-Linux.sh --put-back bo1            # undo removed conflicts
HandheldOps-Windows.bat --display mw2=borderless    # or fullscreen
HandheldOps-Linux.sh --stick-fix cod4=off         # take the stick lines out (on = add)
HandheldOps-Linux.sh --look-speed 1.25        # Black Ops gamepad look speed (experimental)
HandheldOps-Linux.sh --show-options
HandheldOps-Windows.bat --reshade mw2          # Windows: official ReShade setup on MW2
HandheldOps-Windows.bat --shortcuts cod4       # Windows: shortcuts for installed mods
HandheldOps-Linux.sh --game-dir "/path/to/Call of Duty 4" --list
HandheldOps-Linux.sh --ui web               # the web page (also: gui, dialogs, text)
HandheldOps-Linux.sh --version
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  What exists and what doesn't
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Part&lt;/th&gt;
&lt;th&gt;State&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Finds the three games in every Steam library (SD card too)&lt;/td&gt;
&lt;td&gt;Works on test data&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Black Ops controller ON/OFF, backups, undo&lt;/td&gt;
&lt;td&gt;Works on test data&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;IW-Pad install / update / remove&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Real download and install done&lt;/strong&gt; (simulated MW2 folder)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;IW4x install / update / remove from its &lt;code&gt;update.json&lt;/code&gt; manifests&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Real 787 MB install, no-change update and full removal done&lt;/strong&gt; (simulated MW2 folder)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;IW3SP-MOD install / update / remove&lt;/td&gt;
&lt;td&gt;Works against a stand-in server; the real host (gitea.com) is blocked from the build machine&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;CoD4x&lt;/td&gt;
&lt;td&gt;Steps + official page only&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Writing Steam launch options&lt;/td&gt;
&lt;td&gt;Works on a test copy of Steam's file; exe swap checked in bash&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Resolution lock (configs + launch option)&lt;/td&gt;
&lt;td&gt;Works on test configs; arguments checked in bash to reach the game after the exe swap; &lt;strong&gt;not yet seen in the real games&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Look (vkBasalt) presets&lt;/td&gt;
&lt;td&gt;Loaded by real vkBasalt on software Vulkan&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Graphics tiers (Linux)&lt;/td&gt;
&lt;td&gt;All four vkBasalt files loaded by real vkBasalt with the real SweetFX shaders; values read in; game-setting changes tested on Black Ops-style configs&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Graphics tiers (Windows ReShade preset)&lt;/td&gt;
&lt;td&gt;Preset and ReShade.ini format taken from ReShade's source; tested against sample files; &lt;strong&gt;not yet seen in ReShade itself&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Windows Steam launcher (mod via Play, Steam Input)&lt;/td&gt;
&lt;td&gt;The launcher script run end to end with stand-in .exe files: SP starts the mod with its settings, MP stays stock; &lt;strong&gt;Steam Input through it not yet seen on Windows&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Borderless window (IW4x / IW3SP-MOD)&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;r_noborder&lt;/code&gt; confirmed in IW4x's source; &lt;strong&gt;not yet seen in IW4x&lt;/strong&gt;. IW3SP-MOD: tried on Windows, opened at 1280x720 in a corner, so it defaults to fullscreen&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Controller override file (&lt;code&gt;handheld_pad.cfg&lt;/code&gt; + &lt;code&gt;+exec&lt;/code&gt;)&lt;/td&gt;
&lt;td&gt;Written, rewritten and kept when hand-edited, and the launch option only when ticked, all tested; the file name and &lt;code&gt;main&lt;/code&gt; folder are what CoD4 normally searches, &lt;strong&gt;not yet seen loading in the game&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;IW3SP-MOD turn speed (in-game look)&lt;/td&gt;
&lt;td&gt;Values are the usual console ones; added as &lt;code&gt;+set&lt;/code&gt;, skipped when the config has a value above zero; &lt;strong&gt;not yet seen in the game&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;IW3SP-MOD stick bindings (right stick)&lt;/td&gt;
&lt;td&gt;Names taken from IW4x's source and its &lt;code&gt;thumbstick_default.cfg&lt;/code&gt;; config add/take-out tested on sample files. 0.4.2 (lines in the config &lt;strong&gt;and&lt;/strong&gt; on the command line) left the controller dead on Windows; 0.4.3 uses the config only, &lt;strong&gt;not yet seen in the game&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Removing conflicts (autoexec-style files, ReShade overrides, old launch options) and Put back&lt;/td&gt;
&lt;td&gt;Tested on sample files; byte-for-byte put back&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Existing ReShade kept, exact restore&lt;/td&gt;
&lt;td&gt;Tested with a full existing install (DLL, own preset, effect package): nothing overwritten, ReShade.ini restored byte for byte&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;ReShade essentials install (headless setup + 10 files)&lt;/td&gt;
&lt;td&gt;Headless behaviour read from ReShade's setup source; the 10 files downloaded for real; SweetFX SMAA compiled through vkBasalt; &lt;strong&gt;headless setup not yet run on Windows&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Aim feel: vsync off / look speed&lt;/td&gt;
&lt;td&gt;Written; effect on the feel &lt;strong&gt;not yet tested&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;SweetFX shader download&lt;/td&gt;
&lt;td&gt;Real download with git done; the zip route (no git) works against a stand-in server&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Windows: resolution launch option, Start buttons, shortcuts, ReShade setup&lt;/td&gt;
&lt;td&gt;Logic tested on Linux by pretending to be Windows; &lt;strong&gt;not yet run on Windows&lt;/strong&gt; (reshade.me and GitHub zip links are blocked from the build machine)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Tested on a real ROG Ally / with the real games&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;Not yet&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;HD textures, new lighting&lt;/td&gt;
&lt;td&gt;Back burner (see RESEARCH.md)&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  More
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;a href="https://github.com/The-Dorkknight/handheld-ops/blob/main/HOW_IT_WORKS.md" rel="noopener noreferrer"&gt;HOW_IT_WORKS.md&lt;/a&gt;: exactly what is read, changed and backed up&lt;/li&gt;
&lt;li&gt;
&lt;a href="https://github.com/The-Dorkknight/handheld-ops/blob/main/RESEARCH.md" rel="noopener noreferrer"&gt;RESEARCH.md&lt;/a&gt;: all research notes, sources and decisions&lt;/li&gt;
&lt;li&gt;
&lt;a href="https://github.com/The-Dorkknight/handheld-ops/blob/main/CHANGELOG.md" rel="noopener noreferrer"&gt;CHANGELOG.md&lt;/a&gt;, &lt;a href="https://github.com/The-Dorkknight/handheld-ops/blob/main/CONTRIBUTING.md" rel="noopener noreferrer"&gt;CONTRIBUTING.md&lt;/a&gt;,
&lt;a href="https://github.com/The-Dorkknight/handheld-ops/blob/main/SECURITY.md" rel="noopener noreferrer"&gt;SECURITY.md&lt;/a&gt;, &lt;a href="https://github.com/The-Dorkknight/handheld-ops/blob/main/THIRD_PARTY.md" rel="noopener noreferrer"&gt;THIRD_PARTY.md&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;&lt;strong&gt;Source code:&lt;/strong&gt; &lt;a href="https://github.com/The-Dorkknight/handheld-ops" rel="noopener noreferrer"&gt;The-Dorkknight/handheld-ops on GitHub&lt;/a&gt;&lt;/p&gt;

</description>
      <category>gamedev</category>
      <category>opensource</category>
      <category>software</category>
    </item>
    <item>
      <title>Scantastic</title>
      <dc:creator>Matthew Armstrong</dc:creator>
      <pubDate>Sat, 14 Nov 2026 12:16:01 +0000</pubDate>
      <link>https://dev.to/open-madscience/scantastic-29bb</link>
      <guid>https://dev.to/open-madscience/scantastic-29bb</guid>
      <description>&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;WORK IN PROGRESS: MANY FEATURES MAY BE BROKEN.&lt;/strong&gt;&lt;br&gt;
Version 0.11.0-wip. Source-available and free for &lt;strong&gt;non-commercial&lt;/strong&gt; use&lt;br&gt;
(&lt;a href="https://github.com/The-Dorkknight/scantastic/blob/main/LICENSE.md" rel="noopener noreferrer"&gt;PolyForm Noncommercial 1.0.0&lt;/a&gt;). No warranty.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;&lt;strong&gt;Scantastic&lt;/strong&gt; (until 0.7.0 called &lt;em&gt;RealSense 3D Scanner&lt;/em&gt;) is a desktop app that turns an&lt;br&gt;
Intel RealSense depth camera into a tabletop 3D scanner: live preview → scan while you turn the object → build a mesh → save&lt;br&gt;
&lt;code&gt;.stl&lt;/code&gt; / &lt;code&gt;.obj&lt;/code&gt; / &lt;code&gt;.ply&lt;/code&gt;. It can also record the raw camera streams and clean them&lt;br&gt;
up in several passes before the mesh is built.&lt;/p&gt;

&lt;p&gt;Works with the &lt;strong&gt;SR305 / SR300&lt;/strong&gt; (coded light) and &lt;strong&gt;D400-series&lt;/strong&gt; cameras such as&lt;br&gt;
the &lt;strong&gt;D415 inside the Prowise MOVE 3in1&lt;/strong&gt;. The camera is detected automatically.&lt;br&gt;
A &lt;strong&gt;Kinect v1 or v2&lt;/strong&gt; can be used too, as an experimental option that has not been&lt;br&gt;
tried on real hardware yet (&lt;a href="https://github.com/The-Dorkknight/scantastic/blob/main/KINECT.md" rel="noopener noreferrer"&gt;KINECT.md&lt;/a&gt;).&lt;br&gt;
It is designed for modest hardware (a 7th-gen i3 with no GPU is the target);&lt;br&gt;
everything has a CPU path.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fc1j2nrcy81cx40ydyqpa.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fc1j2nrcy81cx40ydyqpa.webp" alt="Main window" width="800" height="637"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;h2&gt;
  
  
  What has actually been tested
&lt;/h2&gt;

&lt;p&gt;Being straight about this, because it is a work in progress:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Area&lt;/th&gt;
&lt;th&gt;Status&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Whole app on a real camera&lt;/td&gt;
&lt;td&gt;0.2.0 has been run on real hardware (i5 U-series laptop, 8 GB). It scans and runs light, but the scans lack definition. 0.5.1 has been run on a D415 (sensor check and a face scan). The rest of 0.2.1 to 0.11.0 has &lt;strong&gt;not&lt;/strong&gt; been run on a camera yet.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Stream clean-up pipeline (&lt;code&gt;stream_pipeline.py&lt;/code&gt;)&lt;/td&gt;
&lt;td&gt;Unit-tested on &lt;strong&gt;synthetic&lt;/strong&gt; scenes (8 tests, run in CI). Not yet tried on real footage.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Sensor check breakdown and surface-detail options (new in 0.2.1)&lt;/td&gt;
&lt;td&gt;Unit-tested on &lt;strong&gt;synthetic&lt;/strong&gt; walls and scans (17 tests with the 3D-view ones; they need Open3D, so CI skips them). Not yet tried on a real camera.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;3D view colours, backgrounds, remembered settings, cancelling a build (0.2.2, merged from an older fork)&lt;/td&gt;
&lt;td&gt;Exercised headless on synthetic scans, plus 4 unit tests. Not yet tried on a real camera.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Kinect v1 / v2 (new in 0.3.0, optional modules)&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Untested on real hardware and against real drivers.&lt;/strong&gt; 25 unit tests with stand-in drivers and a headless run of the app with a stand-in Kinect.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Retro TV look (new in 0.3.0, optional module)&lt;/td&gt;
&lt;td&gt;Exercised headless on Linux, plus 5 unit tests. Not checked on Windows or macOS. &lt;code&gt;--plain&lt;/code&gt; switches it off.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Cutting on the model (new in 0.4.0)&lt;/td&gt;
&lt;td&gt;25 unit tests on made-up shapes, and a headless run of the app that draws, drags, applies and undoes cuts on a noisy made-up scan of a ball on a table. &lt;strong&gt;Not yet tried on a real scan, or by a real hand on a real mouse.&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Double-click launchers (new in 0.4.1)&lt;/td&gt;
&lt;td&gt;The Linux launcher: 9 tests with a stand-in Python, and started for real under a headless display. The Windows launcher has &lt;strong&gt;not been run&lt;/strong&gt; on Windows. The Mac launcher was run by the owner on an M5 Mac with 0.9.0: set-up ran, then Open3D 0.20 failed to load (Homebrew libusb). Fixed in 0.9.1, which is &lt;strong&gt;not yet tried on a Mac&lt;/strong&gt;.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Live picture by itself, camera on/off, Rotate (new in 0.5.0)&lt;/td&gt;
&lt;td&gt;13 tests, 6 of them running the whole window with a stand-in camera and a stand-in "what is plugged in" answer (plugging in, turning off and on, turning off while starting, unplugging, a camera that will not start, turning the picture). &lt;strong&gt;Not yet tried with a real camera being plugged in and out.&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;D400 settings (0.5.1, 0.5.2)&lt;/td&gt;
&lt;td&gt;18 tests with a stand-in RealSense driver. 0.5.1 tried on a real D415: much better than 0.5.0, still soft. &lt;strong&gt;0.5.2 not yet tried.&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Handheld and turntable stitching, picture features, marker ring, scan target (new in 0.9.0)&lt;/td&gt;
&lt;td&gt;Measured on &lt;strong&gt;made-up&lt;/strong&gt; scans with known answers: turntable 0 of 40 failed, typical worst frame 0.6 mm; handheld 0 of 20 held-out and 3 of 20 tuning scans failed (0.8.1 failed every one). 21 tests. &lt;strong&gt;Not yet tried on a real camera, a real turntable, a printed ring or a printed target.&lt;/strong&gt; See &lt;a href="https://github.com/The-Dorkknight/scantastic/blob/main/docs/research/stitching-0.9.0-results.md" rel="noopener noreferrer"&gt;the results&lt;/a&gt;.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Live tracking, scan styles, record-now-stitch-later, captures off the camera thread (new in 0.10.0)&lt;/td&gt;
&lt;td&gt;Made-up video-rate scans: live tracking 0 of 20 failed (tuning and held-out); record-now-stitch-later 2 of 20 held-out scans had one stray frame 6–8 mm out (of about 50 kept); the rest had a typical worst frame of 0.9 mm. 18 tests, including end-to-end window runs with a stand-in camera. &lt;strong&gt;Not yet tried on a real camera or a real 7th-gen i3.&lt;/strong&gt; See &lt;a href="https://github.com/The-Dorkknight/scantastic/blob/main/docs/research/live-tracking-0.10.0.md" rel="noopener noreferrer"&gt;the write-up&lt;/a&gt;.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Speed profiles, alignments side by side, graphics card (new in 0.11.0)&lt;/td&gt;
&lt;td&gt;Held-out made-up handheld scans: the same answer as 0.10.0 on all 20 (1 of 20 fails, as 0.10.0 does when run again; its earlier "0 of 20" was not from its final code). The graphics-card method, run on the processor: also 1 of 20, worst frames within 0.04 mm. Final polish about a quarter quicker; side by side only about 10% quicker on the 2-core test machine. 23 tests. &lt;strong&gt;No real graphics card, no PC with more than 2 cores, and Windows and macOS not tried.&lt;/strong&gt; See &lt;a href="https://github.com/The-Dorkknight/scantastic/blob/main/docs/research/speed-and-gpu-0.11.0.md" rel="noopener noreferrer"&gt;the write-up&lt;/a&gt;.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;SR305 unchanged by the D400 work (0.8.1)&lt;/td&gt;
&lt;td&gt;Compared with 0.5.0 on the same inputs: camera set-up, window settings and mesh match, the last within run-to-run noise. 3 tests keep it so.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Ready-made Intel driver, private Python 3.11, driver reuse (new in 0.8.0)&lt;/td&gt;
&lt;td&gt;A real first run from nothing, here, with only Python 3.12 on the path: Python 3.11 downloaded, Intel's driver installed and hash-checked, app started. The suite passes on that Python (Tk 9). &lt;strong&gt;Not yet tried with a camera plugged in, or on your laptop.&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Advanced-mode switch, firmware note, review fixes (new in 0.7.1)&lt;/td&gt;
&lt;td&gt;20 tests with stand-in drivers, cameras and meshes. The advanced-mode switch matches librealsense 2.54.2's source, but has &lt;strong&gt;not been tried on a real D415&lt;/strong&gt;.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Magenta mask overlay, no-depth-only mask, fill small holes (new in 0.6.1)&lt;/td&gt;
&lt;td&gt;10 tests on made-up walls, a stand-in camera and a synthetic mesh with holes. &lt;strong&gt;Not yet tried on a real camera.&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Settings per camera, Reset, camera badge (new in 0.6.0)&lt;/td&gt;
&lt;td&gt;8 tests, 4 running the whole window with stand-in cameras that are swapped. &lt;strong&gt;Not yet tried with two real cameras.&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Optional ErrCode hook (new in 0.4.1)&lt;/td&gt;
&lt;td&gt;Tried with stand-ins only, not with the real plugin.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Laser power option (new in 0.2.1)&lt;/td&gt;
&lt;td&gt;Checked against a fake sensor only. What value 1 does on SR305 firmware is unconfirmed.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Recording infrared; opening &lt;code&gt;.bag&lt;/code&gt; files; the optional AI mask; the GitHub workflows&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Untested&lt;/strong&gt; with real data.&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;All screenshots here come from synthetic test data, not a real object. The Kinect one&lt;br&gt;
uses a stand-in driver, not a real Kinect.&lt;br&gt;
If something breaks on your camera, please &lt;a href="//../../issues/new/choose"&gt;open an issue&lt;/a&gt;.&lt;/p&gt;
&lt;h2&gt;
  
  
  Kinect (0.3.0, experimental)
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fkd5ia809bpi58b8phu05.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fkd5ia809bpi58b8phu05.webp" alt="The app showing a stand-in Kinect" width="800" height="637"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Camera type&lt;/strong&gt; at the top of the Camera panel switches between RealSense and Kinect.&lt;br&gt;
Kinect support is untested on real hardware, its drivers have to be installed by hand, and&lt;br&gt;
the sensor check, depth confidence and laser settings do not apply to it. The Kinect runs&lt;br&gt;
in its own process, so a driver crash cannot close the app; it is restarted automatically&lt;br&gt;
if it crashes, hangs or stops sending pictures.&lt;br&gt;
Everything is in &lt;a href="https://github.com/The-Dorkknight/scantastic/blob/main/KINECT.md" rel="noopener noreferrer"&gt;KINECT.md&lt;/a&gt;.&lt;/p&gt;
&lt;h2&gt;
  
  
  The look (0.3.0)
&lt;/h2&gt;

&lt;p&gt;The window is dressed as a wood-panelled television: walnut cabinet, black bezel, a&lt;br&gt;
rounded screen, and static on the screen while the camera is off. The static stops the&lt;br&gt;
moment the first camera picture arrives, and a lamp under the screen shows whether the&lt;br&gt;
camera is on standby or live.&lt;/p&gt;

&lt;p&gt;Everything is drawn by the app; there are no picture files. The static is a few pictures&lt;br&gt;
made once and swapped on a timer. In a headless test the whole app used about 4% of one&lt;br&gt;
processor core while showing it.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Look&lt;/strong&gt; (bottom of the right-hand panel) chooses &lt;em&gt;Retro TV&lt;/em&gt;, &lt;em&gt;Retro TV, still screen&lt;/em&gt; (no&lt;br&gt;
movement) or &lt;em&gt;Plain&lt;/em&gt;. It applies the next time the app starts. &lt;code&gt;--plain&lt;/code&gt; on the command&lt;br&gt;
line forces the plain look, and the app falls back to it by itself if the skin fails.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F64j2qc023tk3dqf6magu.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F64j2qc023tk3dqf6magu.webp" alt="Plain look" width="800" height="677"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;h2&gt;
  
  
  Using it
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Plug the camera in. The live picture appears by itself (the app looks for a camera every
few seconds while the live view is showing). Point it at the subject; the live distance
guide shows the sweet spot.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;START SCAN&lt;/strong&gt; (green). In &lt;em&gt;Handheld&lt;/em&gt; mode, move the camera slowly round the object; in
&lt;em&gt;Turntable&lt;/em&gt; mode, keep the camera still and turn the object (below). &lt;strong&gt;STOP&lt;/strong&gt; when done.
&lt;em&gt;Capture Frame&lt;/em&gt; takes single frames instead.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;BUILD MESH&lt;/strong&gt;. Choose &lt;em&gt;Smooth surface&lt;/em&gt; (fuses depth images, recommended) or
&lt;em&gt;Watertight&lt;/em&gt; (Poisson; fills unscanned gaps so it can be printed). While it builds,
the same button turns into &lt;strong&gt;CANCEL&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Save mesh…&lt;/strong&gt; as STL/OBJ/PLY, in millimetres or metres.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Other controls: near/far range, depth confidence, laser power, average-when-still,&lt;br&gt;
keep-only-the-subject, a CAD-style view cube, and &lt;strong&gt;cutting&lt;/strong&gt; (next section) to remove&lt;br&gt;
the table or split overlapping scans. While scanning, the STOP&lt;br&gt;
button shows the elapsed time and the number of frames. &lt;em&gt;Clear captured frames&lt;/em&gt;&lt;br&gt;
asks first if it would throw away a mesh you have not saved.&lt;/p&gt;
&lt;h2&gt;
  
  
  Handheld or turntable (new in 0.9.0)
&lt;/h2&gt;

&lt;p&gt;Pick the &lt;strong&gt;Scan mode&lt;/strong&gt; in the Scan panel. Frames are cut out differently in each mode, so&lt;br&gt;
changing it clears the frames captured so far (it asks first).&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Mode&lt;/th&gt;
&lt;th&gt;You move&lt;/th&gt;
&lt;th&gt;Best for&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Handheld&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;the camera, slowly round the object&lt;/td&gt;
&lt;td&gt;big or fixed things, faces&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Turntable&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;the object, on a turntable or lazy Susan, a little between captures. The camera stays on a stand and must not move.&lt;/td&gt;
&lt;td&gt;small objects; the most reliable&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;strong&gt;Turntable mode in four steps:&lt;/strong&gt;&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Put the camera on a stand, 30–50 cm from the turntable, looking down at it a little.&lt;/li&gt;
&lt;li&gt;Optional, but best: with nothing on the turntable, press &lt;strong&gt;Record empty turntable&lt;/strong&gt;.
Everything that is there now (table, wall, the turntable itself) is left out of the scan.
Without it, the table and the turntable's top are found and removed instead.&lt;/li&gt;
&lt;li&gt;For a &lt;strong&gt;plain or round object&lt;/strong&gt; (a mug, a smooth figure, anything with little pattern),
press &lt;strong&gt;Save marker ring to print (PDF)&lt;/strong&gt;. Print it at 100% (the 50 mm bar must measure
50 mm), lay it flat on the turntable and put the object in the middle. The markers say
exactly how far the turntable turned.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;START SCAN&lt;/strong&gt; and turn the object slowly through a full turn, or Capture Frame every
10–20 degrees.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;How it lines frames up: every frame is a turn about the same axis, so instead of six&lt;br&gt;
unknowns per frame there is one. A step that doesn't fit the axis is re-checked. A full&lt;br&gt;
turn is checked against the start, and the axis and every turn are then fine-tuned&lt;br&gt;
together against all the frames that see the same side. The build report says what it&lt;br&gt;
found, for example "Turntable: one axis fits the steps (0.4 mm typical); 24 frames over&lt;br&gt;
345 degrees; full-turn check from 1 closure(s)". If the steps don't fit one axis (the&lt;br&gt;
object slid, or the camera was knocked) it says so and aligns as a handheld scan instead.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Picture features&lt;/strong&gt; (tick box, on by default): every step between frames is also worked&lt;br&gt;
out from the colour picture, using ORB corners on a greyscale copy, lifted to 3D with the&lt;br&gt;
depth. A shape match that the picture clearly contradicts is overruled. This is what stops&lt;br&gt;
a round object being glued on at the wrong turn. Untick it to compare.&lt;/p&gt;
&lt;h3&gt;
  
  
  Slow PCs, live tracking, and what to expect (new in 0.10.0)
&lt;/h3&gt;

&lt;p&gt;The &lt;strong&gt;Scan style&lt;/strong&gt; box (Scan panel) chooses when the heavy work happens.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Scan style&lt;/th&gt;
&lt;th&gt;While you scan&lt;/th&gt;
&lt;th&gt;After STOP&lt;/th&gt;
&lt;th&gt;For&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Stitch while scanning&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Every capture is lined up as it comes. Handheld, the camera is also followed on the live picture (up to 10 frames a second), and a red banner says &lt;strong&gt;TRACKING LOST - move back slowly&lt;/strong&gt; if it loses its way.&lt;/td&gt;
&lt;td&gt;Build Mesh&lt;/td&gt;
&lt;td&gt;a reasonably quick PC&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Record now, stitch afterwards&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Only the camera's frames are saved to disk (about 10 a second), which even a slow PC keeps up with.&lt;/td&gt;
&lt;td&gt;Stitching starts by itself and shows how long is left. Then Build Mesh.&lt;/td&gt;
&lt;td&gt;slower PCs, such as a &lt;strong&gt;7th-gen Core i3&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The first time it starts, the app times one tracking step on your PC (about a third of&lt;br&gt;
a second). With 4 processor threads or fewer, or a slow step, it picks &lt;em&gt;Record now,&lt;br&gt;
stitch afterwards&lt;/em&gt;. You can change it at any time when not scanning.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;On a 7th-gen i3, be ready for slow.&lt;/strong&gt; It won't freeze or crash, but it takes a while.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Stitching runs as a separate program with a memory limit.&lt;/strong&gt; If it runs out of memory,
only that program stops, and it starts again on every 2nd frame (twice at most). The
recording stays safe on disk whatever happens. A recording cut short by a crash keeps
all but its last few frames.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Expect minutes, not seconds.&lt;/strong&gt; On a 2-core test machine, about as fast as a 7th-gen
i3, a 40-second handheld recording (about 120 frames) took about half a minute to
clean and track. Then lining up the 48 frames it kept took 2–4 minutes, and Build Mesh
takes longer again. A turntable scan is much quicker to line up. These are measured
on made-up scans, not timed on a real i3.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;While stitching&lt;/strong&gt;, the status bar shows the progress and the minutes left. You can
leave it running.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Whichever style you choose, captures are now worked out away from the camera's own&lt;br&gt;
thread. If the PC falls behind, a capture is skipped with a message rather than freezing&lt;br&gt;
the live picture.&lt;/p&gt;
&lt;h3&gt;
  
  
  Speed profiles and the graphics card (new in 0.11.0)
&lt;/h3&gt;

&lt;p&gt;The &lt;strong&gt;Speed and graphics card&lt;/strong&gt; box (right-hand panel) is taken from STL2STEP. It sets how&lt;br&gt;
hard Build Mesh and stitching work the computer.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Profile&lt;/th&gt;
&lt;th&gt;Alignments at once&lt;/th&gt;
&lt;th&gt;Priority&lt;/th&gt;
&lt;th&gt;Graphics card&lt;/th&gt;
&lt;th&gt;
&lt;em&gt;Auto&lt;/em&gt; picks it for&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Potato&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;1&lt;/td&gt;
&lt;td&gt;lowest: the PC stays usable&lt;/td&gt;
&lt;td&gt;no&lt;/td&gt;
&lt;td&gt;under 4 cores or 12 GB, such as a &lt;strong&gt;7th-gen i3&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Moderate&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;half the cores&lt;/td&gt;
&lt;td&gt;below normal&lt;/td&gt;
&lt;td&gt;no&lt;/td&gt;
&lt;td&gt;most laptops&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Extreme&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;every core&lt;/td&gt;
&lt;td&gt;normal&lt;/td&gt;
&lt;td&gt;yes, where it passed its test&lt;/td&gt;
&lt;td&gt;8+ cores and 16 GB&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0qplhamqjn94c4ex10xr.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0qplhamqjn94c4ex10xr.webp" alt="Speed and graphics card box" width="800" height="637"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Be realistic about what it gains.&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Lining the frames up is over 90% of Build Mesh&lt;/strong&gt;, and that is Open3D's ICP alignment.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;On a 7th-gen i3, Potato makes nothing faster.&lt;/strong&gt; It keeps the computer usable while it
works.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Alignments side by side&lt;/strong&gt; give the same results. On the 2-core test machine they were
only about 10% faster; a PC with more cores should gain more, but &lt;strong&gt;that hasn't been
measured&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The final polish&lt;/strong&gt; is now about a quarter quicker, with the same results.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;Graphics cards:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Nvidia on Linux:&lt;/strong&gt; the standard Open3D already contains Nvidia (CUDA) support, so
nothing extra is installed. Press &lt;strong&gt;Test the card&lt;/strong&gt;; the card is only used if a real
alignment and depth fusion on it give the right answers. Otherwise the processor does the
work, as before.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Nvidia on Windows:&lt;/strong&gt; not used. Open3D's Windows graphics-card build is an experimental
preview that isn't on the normal download site.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Apple M1–M5 Macs:&lt;/strong&gt; not used. Open3D can't use Apple's graphics chip; Extreme uses all
the processor cores instead.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;AMD and Intel graphics:&lt;/strong&gt; not used. Open3D supports Nvidia cards only.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;No real graphics card has been tried.&lt;/strong&gt; The card's code was run on the processor
instead; see &lt;a href="https://github.com/The-Dorkknight/scantastic/blob/main/docs/research/speed-and-gpu-0.11.0.md" rel="noopener noreferrer"&gt;the write-up&lt;/a&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;code&gt;python3 scantastic.py --speed-check&lt;/code&gt; prints the same information and tests the card.&lt;/p&gt;
&lt;h3&gt;
  
  
  Scan target and compare (new in 0.9.0)
&lt;/h3&gt;

&lt;p&gt;&lt;strong&gt;Save printable target (STL)&lt;/strong&gt; writes a 70 x 70 mm test block, shown below. Each corner&lt;br&gt;
has one feature:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;steps 2, 4, 6 and 8 mm high;&lt;/li&gt;
&lt;li&gt;a dome 30 mm across;&lt;/li&gt;
&lt;li&gt;grooves 1, 2 and 3 mm wide;&lt;/li&gt;
&lt;li&gt;a ramp.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;An L-shaped notch on one side means it only fits one way round. Print it in a matt&lt;br&gt;
colour, scan it on the turntable and build the mesh. Then &lt;strong&gt;Compare mesh with scan target&lt;/strong&gt;&lt;br&gt;
reports:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;the typical error and the error that 9 points in 10 stay within, in mm;&lt;/li&gt;
&lt;li&gt;how much of the top was covered;&lt;/li&gt;
&lt;li&gt;each step height as scanned.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The same can be run outside the app:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python scan_target.py target.stl          &lt;span class="c"&gt;# the STL&lt;/span&gt;
python scan_target.py --ring ring.pdf     &lt;span class="c"&gt;# the marker ring&lt;/span&gt;
python scan_target.py --compare scan.stl  &lt;span class="c"&gt;# compare a scan with the target&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fye9ktphf7fil7wo911xz.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fye9ktphf7fil7wo911xz.png" alt="Scan target" width="800" height="578"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ff9grcrah12jnnw5eg5pe.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ff9grcrah12jnnw5eg5pe.webp" alt="Turntable mode" width="800" height="637"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  The camera on and off, and turning the picture (new in 0.5.0)
&lt;/h3&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Control&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Turn camera off / Turn camera on&lt;/strong&gt; (Camera panel)&lt;/td&gt;
&lt;td&gt;Off stops the camera and keeps it off, also the next time the app starts, until you turn it on. The line under the button says what the camera is doing: live, waiting for one to be plugged in, or off.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Rotate&lt;/strong&gt; (above the picture)&lt;/td&gt;
&lt;td&gt;Turns the live picture a quarter turn clockwise each press, for a camera mounted on its side or upside down. The 3D model's default view and the sensor check's pictures turn with it. The scan data itself is not changed. Remembered.&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Unplug the camera and the app notices (at once on Linux, within about 8 seconds&lt;br&gt;
elsewhere), says so, and starts the picture again when the camera is back. A camera that is&lt;br&gt;
plugged in but will not start is reported once in the status line, with no dialog; press&lt;br&gt;
&lt;em&gt;Capture Frame&lt;/em&gt; to see the full reason. START SCAN and Capture Frame still start the camera&lt;br&gt;
themselves, and turn it on if it was off.&lt;/p&gt;

&lt;p&gt;A Kinect is looked for in the same way on Linux. On Windows and macOS there is no quick way&lt;br&gt;
to look for one, so a Kinect starts when you press START SCAN or Capture Frame.&lt;/p&gt;

&lt;h2&gt;
  
  
  Settings for each camera, and which camera is running (new in 0.6.0)
&lt;/h2&gt;

&lt;p&gt;Each kind of camera has its own settings. When a camera starts, the app puts that&lt;br&gt;
camera's settings in the window, and anything you change is remembered for that camera&lt;br&gt;
only. So a D415 and an SR305 can be swapped without retuning. &lt;strong&gt;Reset ... settings&lt;/strong&gt; in&lt;br&gt;
the Camera panel puts the starting values back for the camera in use. The line under it&lt;br&gt;
lists what you have changed.&lt;/p&gt;

&lt;p&gt;The coloured badge at the top of the Camera panel says which camera is running: blue for&lt;br&gt;
an SR305 / SR300, orange for a D400, purple for a Kinect, grey for none. The model name is&lt;br&gt;
also shown in that colour on the live picture and in the window title.&lt;/p&gt;

&lt;p&gt;Starting values (best guesses; the reasons are in DESIGN_NOTES section 3):&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Setting&lt;/th&gt;
&lt;th&gt;SR305 / SR300&lt;/th&gt;
&lt;th&gt;D400 (D415)&lt;/th&gt;
&lt;th&gt;Kinect&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Near / Far&lt;/td&gt;
&lt;td&gt;0.20 / 1.00 m&lt;/td&gt;
&lt;td&gt;0.25 / 1.00 m&lt;/td&gt;
&lt;td&gt;0.5 / 3.0 m&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Depth confidence&lt;/td&gt;
&lt;td&gt;Standard&lt;/td&gt;
&lt;td&gt;(no such setting)&lt;/td&gt;
&lt;td&gt;(no such setting)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Laser power&lt;/td&gt;
&lt;td&gt;Maximum&lt;/td&gt;
&lt;td&gt;Maximum&lt;/td&gt;
&lt;td&gt;(no such setting)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Quality&lt;/td&gt;
&lt;td&gt;60&lt;/td&gt;
&lt;td&gt;60&lt;/td&gt;
&lt;td&gt;60&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Surface detail&lt;/td&gt;
&lt;td&gt;Standard&lt;/td&gt;
&lt;td&gt;Fine (1.0 mm grid)&lt;/td&gt;
&lt;td&gt;Standard&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Narrow band / 3+ frames&lt;/td&gt;
&lt;td&gt;off / off&lt;/td&gt;
&lt;td&gt;off / off&lt;/td&gt;
&lt;td&gt;off / off&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Final smoothing passes&lt;/td&gt;
&lt;td&gt;5&lt;/td&gt;
&lt;td&gt;2&lt;/td&gt;
&lt;td&gt;5&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Fill holes smaller than&lt;/td&gt;
&lt;td&gt;off&lt;/td&gt;
&lt;td&gt;8 mm&lt;/td&gt;
&lt;td&gt;off&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Sensor-check mask: only no-depth patches&lt;/td&gt;
&lt;td&gt;off&lt;/td&gt;
&lt;td&gt;on&lt;/td&gt;
&lt;td&gt;off&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Only keep the subject / Average when still&lt;/td&gt;
&lt;td&gt;on / on&lt;/td&gt;
&lt;td&gt;on / on&lt;/td&gt;
&lt;td&gt;on / on&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  Black spots in the same place on the screen (0.6.1)
&lt;/h2&gt;

&lt;p&gt;A black spot that stays put whatever you scan is one of two things, and the live depth&lt;br&gt;
picture now tells them apart:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Magenta:&lt;/strong&gt; the sensor-check mask. You ran the sensor check and the app ignores those
pixels on purpose. Untick &lt;strong&gt;Ignore damaged sensor areas&lt;/strong&gt; to see what the camera gives
there, or tick &lt;strong&gt;...only patches with no depth&lt;/strong&gt; to keep the merely noisy parts
(run the sensor check again once, to save that smaller mask).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Black:&lt;/strong&gt; the camera gives no depth there (a scratch, mark or smear on the window,
or a dark or shiny patch). Clean the window with a blower and a dry, soft cloth (no liquids).&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Either way, &lt;strong&gt;move the object (or yourself) between frames&lt;/strong&gt;. A spot at a fixed place on&lt;br&gt;
the screen falls on a different part of a turning object in the next frame, so the&lt;br&gt;
other frames fill it in. And &lt;strong&gt;Fill holes smaller than ... mm&lt;/strong&gt; (Quality panel) closes&lt;br&gt;
what is left in the finished mesh with flat patches. It starts at 8 mm for a D400.&lt;/p&gt;

&lt;h2&gt;
  
  
  D400 advanced mode and firmware (0.7.1)
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Advanced mode.&lt;/strong&gt; A D400 has a setting stored in the camera called "advanced mode".&lt;br&gt;
Intel's driver only offers the High Accuracy preset and 0.1 mm depth steps while it is on,&lt;br&gt;
and the ripple correction needs it too. The first time a D400 with it off starts, the app&lt;br&gt;
asks whether it may switch it on. The camera restarts once, a few seconds, and it stays&lt;br&gt;
on. The tick box "Let the app switch on a D400's advanced mode" in the Camera panel&lt;br&gt;
changes your answer. Intel RealSense Viewer can switch it off again.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;D400 firmware.&lt;/strong&gt; The app uses Intel's driver 2.54.2. It has to: 2.55 and later dropped the&lt;br&gt;
SR305. That driver is fine for a D415, because the depth is worked out inside the&lt;br&gt;
camera, but Intel recommends D400 firmware 5.15.0.2 or later with it. The status line&lt;br&gt;
says when a camera's firmware is older. The D415 tested here reports 5.11.10, from 2019.&lt;br&gt;
It works. Updating is optional, done with Intel's tools rather than this app:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Intel RealSense Viewer, under More, Update Firmware;&lt;/li&gt;
&lt;li&gt;or &lt;code&gt;rs-fw-update&lt;/code&gt;.
Read the D400 firmware release notes first. Keep the camera plugged in and don't
interrupt the update.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  D400 cameras such as the D415 (0.5.1)
&lt;/h2&gt;

&lt;p&gt;Before 0.5.1 a D400 shared the SR305's settings, and scans looked coated in a thick soft&lt;br&gt;
skin. Now a D400 gets its own:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;depth at 848x480 instead of 640x480 (sharper; still works from about 30 cm);&lt;/li&gt;
&lt;li&gt;0.1 mm depth steps instead of 1 mm;&lt;/li&gt;
&lt;li&gt;Intel's ripple correction (A-factor 0.08), if the camera's advanced mode is on. If
it is off, the log says so. Intel RealSense Viewer can switch it on once and it stays on;&lt;/li&gt;
&lt;li&gt;a wider trim along object outlines, where stereo depth spills past the real edge;&lt;/li&gt;
&lt;li&gt;(0.5.2) the colour laid onto the depth, so the depth is never resampled;&lt;/li&gt;
&lt;li&gt;(0.5.2) 2 smoothing passes to start with instead of 5; (0.6.0) its own remembered settings.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The status line and the sensor check report say whether the ripple correction is on.&lt;/p&gt;

&lt;p&gt;Scan at the close end of the guide, about 35 to 45 cm. If a white or shiny object still&lt;br&gt;
looks blobby, try Laser &lt;strong&gt;Reduced&lt;/strong&gt;. A sensor-check mask made before 0.5.1 is ignored on a&lt;br&gt;
D400 (wrong resolution), so run the check again.&lt;/p&gt;

&lt;h2&gt;
  
  
  Cutting: point at where the cut goes (new in 0.4.0)
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3ncnc9vw0cwzj8r6no50.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3ncnc9vw0cwzj8r6no50.webp" alt="A cut set on the model" width="800" height="637"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The cut is set on the model itself, in the 3D view. The part that will be cut off turns&lt;br&gt;
red before anything is changed.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Press &lt;strong&gt;Draw a cut&lt;/strong&gt; (in the 3D view, or &lt;em&gt;Draw the cut on the model&lt;/em&gt; in the Edit
panel) and drag a line across the model. The cut follows the line and goes straight
into the screen. Hold Shift for a level, upright or 15-degree line. Esc cancels.&lt;/li&gt;
&lt;li&gt;Adjust it if needed: drag the &lt;strong&gt;yellow dot&lt;/strong&gt; to move the cut, drag the &lt;strong&gt;red arrow&lt;/strong&gt; to
turn it. The arrow points at the part that goes. Or press &lt;strong&gt;Redraw&lt;/strong&gt; and draw the line
again.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Flip side&lt;/strong&gt; if the wrong part is red. &lt;strong&gt;Apply cut&lt;/strong&gt; when it looks right. &lt;strong&gt;Undo last
edit&lt;/strong&gt; brings back both the mesh and the cut, so you can nudge it and apply again.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Three shortcuts, in the Edit panel:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Button&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Find the table / flat base&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Finds the biggest flat area of the model and puts the cut just above it, so the table goes and the subject stays. It is a suggestion: check the red part first.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Top, Bottom, Left, Right, Near, Far&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Cuts off that side of the model &lt;em&gt;as you see it now&lt;/em&gt;. Turn the view first (the view cube snaps to a face) and these give exactly level or upright cuts.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;How much&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Slides the cut through the model, from nothing cut off to everything.&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;"Cap the cut" (on by default) closes the cut with a flat face, which makes a solid you can&lt;br&gt;
print; untick it to simply trim the surface. A scan seen from one side only is filled in&lt;br&gt;
straight back to a flat back, just behind the deepest point of the part you keep.&lt;/p&gt;

&lt;p&gt;This replaces the earlier orientation list, position slider and tilt joystick. A line drawn&lt;br&gt;
on the model removes whichever part is the thinner slice; if you meant the other, Flip side.&lt;/p&gt;

&lt;h2&gt;
  
  
  Looking at the result: colours and background (0.2.2)
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffdv4l6f44t361f8vj5ds.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffdv4l6f44t361f8vj5ds.webp" alt="Height colours in the 3D view" width="800" height="637"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Above the 3D view, &lt;strong&gt;Colours&lt;/strong&gt; chooses how the mesh is painted:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Colours&lt;/th&gt;
&lt;th&gt;Shows&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Scan colours&lt;/td&gt;
&lt;td&gt;The colours the camera saw.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Plain&lt;/td&gt;
&lt;td&gt;Grey, shaded only. Best for judging the shape without the texture distracting.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Height (low - high)&lt;/td&gt;
&lt;td&gt;Blue at the bottom to red at the top, with the range in millimetres and the highest and lowest points marked.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Relief (back - front)&lt;/td&gt;
&lt;td&gt;The same map, front to back. Small bumps and dents stand out.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Coverage (rescan hints)&lt;/td&gt;
&lt;td&gt;How many frames saw each part, so you know where to scan again.&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;strong&gt;Background&lt;/strong&gt; switches the 3D view between dark, grey, light, white, navy or a colour&lt;br&gt;
you pick. Both choices are remembered in &lt;code&gt;settings.json&lt;/code&gt; next to the script.&lt;/p&gt;

&lt;h2&gt;
  
  
  Sensor check: what kind of fault is it? (new in 0.2.1)
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ff3bbq0517yzjo7wut95q.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ff3bbq0517yzjo7wut95q.webp" alt="Sensor check result" width="800" height="726"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Sensor check (point at a wall)…&lt;/strong&gt; looks at a plain, matt wall for about two&lt;br&gt;
seconds and finds the parts of the view that misbehave, so scans can ignore them.&lt;br&gt;
Since 0.2.1 it also says what kind of fault it found, because the cures differ:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Flicker (noise)&lt;/strong&gt;: the reading jumps about from frame to frame. Left picture.
It cannot be subtracted; those pixels can only be trusted less.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Steady offset&lt;/strong&gt;: the reading is wrong by the same amount every frame. Right
picture (red = reads too far, blue = too near). An offset can in principle be subtracted.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Missing depth&lt;/strong&gt;: purple in both pictures.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;It also reports how curved the camera sees the flat wall (calibration error, or a&lt;br&gt;
wall that is not flat) and whether a suspicious ring is only smooth calibration&lt;br&gt;
error. Every run saves two pictures, the two maps (&lt;code&gt;.npz&lt;/code&gt;) and a text report in&lt;br&gt;
&lt;code&gt;sensor_check/&lt;/code&gt;, named by wall distance and laser setting, so runs can be compared.&lt;/p&gt;

&lt;p&gt;Useful comparisons, each one a fresh run of the check:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Change&lt;/th&gt;
&lt;th&gt;If the pattern changes&lt;/th&gt;
&lt;th&gt;If it stays the same&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Laser power Maximum → Reduced&lt;/td&gt;
&lt;td&gt;Glare or saturation&lt;/td&gt;
&lt;td&gt;Calibration or the window&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Wall at about 30, 40 and 60 cm&lt;/td&gt;
&lt;td&gt;Glare or saturation&lt;/td&gt;
&lt;td&gt;Calibration or optics&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Camera turned 10–20° off square&lt;/td&gt;
&lt;td&gt;Glare off the wall&lt;/td&gt;
&lt;td&gt;The camera&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Camera upside down or slid sideways&lt;/td&gt;
&lt;td&gt;Stays with the wall: the wall is not flat&lt;/td&gt;
&lt;td&gt;Stays with the image: the camera&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Which pixels are ignored is decided exactly as in 0.2.0. One thing the breakdown&lt;br&gt;
shows that the ignore-mask misses: a large area with a steady offset (a third of&lt;br&gt;
the view reading 4 mm off, in the tests) is &lt;strong&gt;not&lt;/strong&gt; ignored, because the old rule&lt;br&gt;
compares each pixel with the spread across the whole view. The report says so when it happens.&lt;/p&gt;

&lt;h2&gt;
  
  
  Surface detail options (new in 0.2.1)
&lt;/h2&gt;

&lt;p&gt;Under &lt;strong&gt;Quality&lt;/strong&gt;, for the &lt;em&gt;Smooth surface&lt;/em&gt; method. The defaults give exactly the&lt;br&gt;
surface earlier versions gave.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Option&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Surface detail&lt;/td&gt;
&lt;td&gt;Grid size for fusing. &lt;em&gt;Standard&lt;/em&gt; follows the Quality slider (2.5 mm at Low, about 1.5 mm in the middle, 1 mm at High). &lt;em&gt;Fine / Finer / Finest&lt;/em&gt; fix it at 1.0 / 0.75 / 0.5 mm.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Narrow fusion band&lt;/td&gt;
&lt;td&gt;Sharper, but may leave holes where depth is noisy.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Only surface seen in 3+ frames&lt;/td&gt;
&lt;td&gt;Less noise, more holes. Needs 6 or more frames.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Final smoothing passes&lt;/td&gt;
&lt;td&gt;0 = sharpest, 5 = as before.&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Measured on a synthetic test object (a ball with 1 mm ripples, 40 frames, 1.5 mm&lt;br&gt;
depth noise, subject isolated), with the camera positions known exactly:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Setting&lt;/th&gt;
&lt;th&gt;6 mm ripples kept&lt;/th&gt;
&lt;th&gt;3 mm ripples kept&lt;/th&gt;
&lt;th&gt;Fusing time&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Standard, Quality in the middle (as before)&lt;/td&gt;
&lt;td&gt;64%&lt;/td&gt;
&lt;td&gt;48%&lt;/td&gt;
&lt;td&gt;0.7 s&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Standard, smoothing 0&lt;/td&gt;
&lt;td&gt;70%&lt;/td&gt;
&lt;td&gt;66%&lt;/td&gt;
&lt;td&gt;0.7 s&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Fine 1.0 mm, smoothing 0&lt;/td&gt;
&lt;td&gt;73%&lt;/td&gt;
&lt;td&gt;71%&lt;/td&gt;
&lt;td&gt;1.1 s&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Fine 1.0 mm, narrow band, smoothing 0&lt;/td&gt;
&lt;td&gt;77%&lt;/td&gt;
&lt;td&gt;73%&lt;/td&gt;
&lt;td&gt;1.1 s&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Finest 0.5 mm, smoothing 0&lt;/td&gt;
&lt;td&gt;76%&lt;/td&gt;
&lt;td&gt;75%&lt;/td&gt;
&lt;td&gt;5.0 s&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;So these options help, modestly. &lt;strong&gt;How accurately the frames are lined up matters&lt;br&gt;
far more.&lt;/strong&gt; The same test with small errors added to every camera position:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Error per frame&lt;/th&gt;
&lt;th&gt;6 mm ripples kept (Fine, smoothing 0)&lt;/th&gt;
&lt;th&gt;3 mm ripples kept&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;none&lt;/td&gt;
&lt;td&gt;73%&lt;/td&gt;
&lt;td&gt;71%&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;0.1° and 0.3 mm&lt;/td&gt;
&lt;td&gt;71%&lt;/td&gt;
&lt;td&gt;63%&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;0.25° and 0.5 mm&lt;/td&gt;
&lt;td&gt;65%&lt;/td&gt;
&lt;td&gt;46%&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;0.5° and 1 mm&lt;/td&gt;
&lt;td&gt;48%&lt;/td&gt;
&lt;td&gt;7%&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Fine grids need more memory. If a scan would need too much, the app uses the&lt;br&gt;
finest grid that fits and says so. Scanning only the subject ("Only keep the&lt;br&gt;
subject in the middle") keeps the volume small.&lt;/p&gt;

&lt;h2&gt;
  
  
  Raw stream and clean-up
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fbvkbcx2fw6iutu3ccdya.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fbvkbcx2fw6iutu3ccdya.webp" alt="Clean-up dialog" width="546" height="472"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Tick &lt;strong&gt;Record every camera frame&lt;/strong&gt; before scanning to keep the raw colour and depth&lt;br&gt;
(and optionally infrared) on disk. Afterwards, &lt;strong&gt;Open recording and clean up…&lt;/strong&gt;&lt;br&gt;
replays a recording folder (or a RealSense &lt;code&gt;.bag&lt;/code&gt;) through these passes, each optional:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Depth gate&lt;/strong&gt;: drop anything outside the near/far range.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Remove the table&lt;/strong&gt;: finds the flat supporting surface (RANSAC) and removes it and everything below.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Edge-aware smoothing&lt;/strong&gt;: smooths depth noise but stops at edges the colour picture shows.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Track camera movement from the picture&lt;/strong&gt;: matches features in infrared first, then colour, and lifts them to 3D using the depth. The resulting poses guide the mesh alignment: they seed the ICP, rescue frames ICP can't verify, and veto an ICP result that contradicts confident tracking.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Compare neighbouring frames&lt;/strong&gt;: re-projects neighbours into each frame, drops depth that disagrees (noise) and fills small holes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Small AI background mask&lt;/strong&gt; (optional, off by default): needs &lt;code&gt;pip install onnxruntime&lt;/code&gt; into &lt;code&gt;rs_env&lt;/code&gt; and &lt;code&gt;models/u2netp.onnx&lt;/code&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Then it loads the result as normal scan frames, ready for BUILD MESH. How each pass&lt;br&gt;
works, and where it can go wrong, is in &lt;a href="https://github.com/The-Dorkknight/scantastic/blob/main/HOW_IT_WORKS.md" rel="noopener noreferrer"&gt;HOW_IT_WORKS.md&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2d2qm791178a26r8wvda.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2d2qm791178a26r8wvda.webp" alt="Quality and detail options" width="800" height="637"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Notes, research and tools
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Where&lt;/th&gt;
&lt;th&gt;What&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/The-Dorkknight/scantastic/blob/main/DESIGN_NOTES.md" rel="noopener noreferrer"&gt;DESIGN_NOTES.md&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;What was looked into, what was measured, what was decided and why, what is untested, what to build next. Start here.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/The-Dorkknight/scantastic/blob/main/docs/research/README.md" rel="noopener noreferrer"&gt;docs/research&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;The full research report on soft scans and damaged sensors (68 sources), the working notes behind it, a printable PDF in plain English, and the Kinect driver bug report.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/The-Dorkknight/scantastic/blob/main/HOW_IT_WORKS.md" rel="noopener noreferrer"&gt;HOW_IT_WORKS.md&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;What the app does, step by step, written from the code.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/The-Dorkknight/scantastic/blob/main/PROJECT_NOTES.md" rel="noopener noreferrer"&gt;PROJECT_NOTES.md&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Short hand-over notes: where things stand and what is open.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/The-Dorkknight/scantastic/tree/main/tools" rel="noopener noreferrer"&gt;tools&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;detail_test.py&lt;/code&gt; repeats the surface-detail measurements; &lt;code&gt;make_screenshots.py&lt;/code&gt; retakes every picture here from made-up data; &lt;code&gt;make_research_pdf.py&lt;/code&gt; rebuilds the PDF; &lt;code&gt;bench/&lt;/code&gt; is the stitching benchmark (&lt;code&gt;python tools/bench/bench.py --mode rotary --seeds 0-1&lt;/code&gt;).&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/The-Dorkknight/scantastic/blob/main/scan_target.py" rel="noopener noreferrer"&gt;scan_target.py&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;The printable scan target, its compare report, and the turntable marker ring.&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;




&lt;p&gt;&lt;strong&gt;Source code:&lt;/strong&gt; &lt;a href="https://github.com/The-Dorkknight/scantastic" rel="noopener noreferrer"&gt;The-Dorkknight/scantastic on GitHub&lt;/a&gt;&lt;/p&gt;

</description>
      <category>3dscanning</category>
    </item>
    <item>
      <title>Blip</title>
      <dc:creator>Matthew Armstrong</dc:creator>
      <pubDate>Fri, 13 Nov 2026 11:53:21 +0000</pubDate>
      <link>https://dev.to/open-madscience/blip-p77</link>
      <guid>https://dev.to/open-madscience/blip-p77</guid>
      <description>&lt;p&gt;&lt;strong&gt;WORK IN PROGRESS: MANY FEATURES MAY BE BROKEN&lt;/strong&gt; — proof of concept, v0.2.2-wip&lt;/p&gt;

&lt;h3&gt;
  
  
  Turn a tiny radar chip and an old phone into a people-spotting radar 📡
&lt;/h3&gt;

&lt;p&gt;Ever wanted to know who is walking about, and how far away they are, without a camera? Blip does it with a tiny radar chip and a spare phone. The chip sends out invisible radio waves and listens for the echoes. When someone moves, the echo changes, and Blip works out where they are.&lt;/p&gt;

&lt;p&gt;The phone is the screen. Blip makes its own Wi-Fi network; you join it, open a web page, and a glowing green radar scope appears, like something from a submarine film. Up to three people show up as dots with their distance, direction and speed. No app to install, no internet needed, no cloud, no camera.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;How it works, in three steps&lt;/strong&gt;&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Plug a little ESP32 board into a radar chip (the Ai-Thinker &lt;strong&gt;RD-03D&lt;/strong&gt;) and power it from USB.&lt;/li&gt;
&lt;li&gt;Join the Wi-Fi network called &lt;em&gt;Blip-Radar&lt;/em&gt; on any old phone, tablet or laptop.&lt;/li&gt;
&lt;li&gt;Open the page and watch the dots move. Walk about and see yourself on the scope!&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;&lt;strong&gt;Good things to know&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;It only sees things that &lt;strong&gt;move&lt;/strong&gt;. Someone sitting very still fades from the scope.&lt;/li&gt;
&lt;li&gt;It also bounces off big metal things such as fridges, so a small room can look busy.&lt;/li&gt;
&lt;li&gt;A tiny optional screen (an OLED) can sit on the board and show the same picture. Blip finds it by itself when it starts.&lt;/li&gt;
&lt;li&gt;You can switch between tracking the one nearest person or up to three.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;It is a slimmed-down take on &lt;a href="https://github.com/Stevee87/Arduino-ESP32-Radarproject" rel="noopener noreferrer"&gt;Stevee87's Arduino-ESP32-Radarproject&lt;/a&gt;, which pairs the same radar with a XIAO ESP32-S3 &lt;em&gt;and&lt;/em&gt; an Arduino GIGA R1 + display shield. The drawing is the easy part; a phone already does it.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Required Notice: Copyright Matthew Armstrong (&lt;a href="https://github.com/The-Dorkknight" rel="noopener noreferrer"&gt;https://github.com/The-Dorkknight&lt;/a&gt;).&lt;br&gt;
Licensed under the &lt;a href="https://github.com/The-Dorkknight/blip-radar/blob/main/LICENSE.md" rel="noopener noreferrer"&gt;PolyForm Noncommercial License 1.0.0&lt;/a&gt; — source-available, free for non-commercial use. Provided as-is with &lt;strong&gt;no warranty and no liability&lt;/strong&gt;. &lt;strong&gt;Not a medical device.&lt;/strong&gt; Radar can see motion through thin walls: respect other people's privacy and the 24 GHz rules where you live.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Project started 2026-11-05 10:24.&lt;/p&gt;

&lt;p&gt;📘 &lt;strong&gt;The whole manual and research notes as one PDF:&lt;/strong&gt; &lt;a href="https://github.com/The-Dorkknight/blip-radar/blob/main/docs/Blip-manual.pdf" rel="noopener noreferrer"&gt;docs/Blip-manual.pdf&lt;/a&gt; (also attached to every release).&lt;/p&gt;

&lt;p&gt;
&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fblip-radar%2Fmain%2Fdocs%2Fimages%2Fui-scope.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fblip-radar%2Fmain%2Fdocs%2Fimages%2Fui-scope.webp" alt="Blip radar scope on a phone" width="640" height="1384"&gt;&lt;/a&gt;
&lt;/p&gt;

&lt;p&gt;Phone screenshot from the simulator (tools/sim_radar.py).&lt;/p&gt;

&lt;h2&gt;
  
  
  Bonus experiment: the Stetho add-on (parked)
&lt;/h2&gt;

&lt;p&gt;The idea: use a second, very cheap radar-style module (a £3 CDM324 or HB100 Doppler sensor) as a &lt;strong&gt;stethoscope that listens to a heartbeat from a distance&lt;/strong&gt;. A modded MAX4466 microphone board boosts the signal, and the ESP32 cleans up the sound (ducks knocks, lifts heart, lung and throat sounds, cuts hiss) and plays it live to the phone, to 3.5 mm earphones, or to a small speaker.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Status: abandoned, but kept as an optional install.&lt;/strong&gt; The radar and phone scope are the project; the stethoscope is parked. The firmware, page and tests are still in the repo and still build, and you can switch it on in Settings, but nobody is working on it. Why, in the author's words: &lt;em&gt;"Needing to build an ultra-low-noise, ultra-high-gain, fast-slew-rate instrumentation amplifier is just too much work for a 'fun' 30-minute project, but I encourage anyone gifted enough with analogue, or mad enough, to carry on my research 😉"&lt;/em&gt; Findings and test recordings are in &lt;a href="https://github.com/The-Dorkknight/blip-radar/blob/main/docs/STETHO_SIGNAL_NOTES.md" rel="noopener noreferrer"&gt;docs/STETHO_SIGNAL_NOTES.md&lt;/a&gt;.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;
&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fblip-radar%2Fmain%2Fdocs%2Fimages%2Fui-stetho.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fblip-radar%2Fmain%2Fdocs%2Fimages%2Fui-stetho.webp" alt="Stetho screen" width="640" height="1384"&gt;&lt;/a&gt;
&lt;/p&gt;

&lt;p&gt;Simulator screenshot. The real thing never got past the noise floor.&lt;/p&gt;

&lt;h2&gt;
  
  
  What exists vs what doesn't
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Part&lt;/th&gt;
&lt;th&gt;State&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;RD-03D frame parser (sign bits, resync, ACK tolerance)&lt;/td&gt;
&lt;td&gt;Done; unit-tested natively in C++ and Python (&lt;code&gt;tests/&lt;/code&gt;)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Single/multi-target mode command&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Fixed in v0.2.1&lt;/strong&gt;: now wrapped in enable/end-config like the original project (earlier versions probably did nothing)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;"Smart" target filter (own track ids, smoothing, hides flickers, wall reflections, fans)&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Experimental, opt-in.&lt;/strong&gt; Native tests on synthetic scenes pass; not tried in a real room&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Auto-sleep + optional power switches for the RD-03D and the CDM324&lt;/td&gt;
&lt;td&gt;Written, compiles; not tried on hardware&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Phone web UI: scope, targets table, tracker ping, settings, about&lt;/td&gt;
&lt;td&gt;Done; tested in Chromium against the simulator&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Flasher page the launcher opens (Quick flash of bundled firmware, or build from source; serial output; error tips)&lt;/td&gt;
&lt;td&gt;Done; tested end to end with stand-ins for esptool and PlatformIO (&lt;code&gt;tests/test_flasher.py&lt;/code&gt;); the bundled images and the esptool command were checked with the real esptool (&lt;code&gt;image_info&lt;/code&gt;: valid); &lt;strong&gt;not yet used on a real board&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Wi-Fi hotspot + captive portal + WebSocket/HTTP API&lt;/td&gt;
&lt;td&gt;Running on the author's bench (Nov 2026): the page loads and shows live targets&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;ESP32-C3 Super Mini and Waveshare ESP32-S3-Pico builds&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Compile cleanly&lt;/strong&gt; (pioarduino, Arduino-ESP32 3.3.12); not yet flashed&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;OLED mirror (0.91" 128×32 / 0.96" 128×64 SSD1306)&lt;/td&gt;
&lt;td&gt;Works on the author's bench; found automatically at boot. The previews below are rendered by the firmware's own drawing code on a PC&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Stetho audio clean-up on the ESP32 (bump gate, heart/lungs/voice presets, noise suppression, AGC)&lt;/td&gt;
&lt;td&gt;Done in fixed point; native test on a synthetic scene passes (knocks −38 dB vs heart sounds, hiss −11 dB)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Live low-latency audio to the phone, chest-motion trace, heart/breath estimates&lt;/td&gt;
&lt;td&gt;Done; tested in Chromium against the simulator (54–83 ms delay there)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Earphone (delta-sigma) and I2S amp outputs on the node&lt;/td&gt;
&lt;td&gt;Written, compiles; &lt;strong&gt;not heard yet&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;CDM324 + modded MAX4466 front end&lt;/td&gt;
&lt;td&gt;Built and tried on a bench (Nov 2026): static fixed by supply caps, but the heartbeat sits at the noise floor even at 2 cm. &lt;strong&gt;Abandoned&lt;/strong&gt;: it needs a custom ultra-low-noise, high-gain instrumentation amplifier (and ideally an I/Q module)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Making the RD-03D itself into a Doppler stethoscope&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Not possible through its UART&lt;/strong&gt; — see &lt;a href="https://github.com/The-Dorkknight/blip-radar/blob/main/DESIGN_NOTES.md" rel="noopener noreferrer"&gt;DESIGN_NOTES.md&lt;/a&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;RD-03D tracking in a real room&lt;/td&gt;
&lt;td&gt;Tracks up to 3 targets, and also big metal objects as you walk towards them; noisy in a small house, outdoor test still to do&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Clutter options (minimum range, edge limit, handheld mode, strict ghost test)&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Experimental.&lt;/strong&gt; Native tests on synthetic scenes pass; simulator street walk works; not tried on a real walk&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;LD2450 custom firmware tab in the flasher&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Experimental, brick risk.&lt;/strong&gt; Tested with stand-in tools (&lt;code&gt;tests/test_cfw_flasher.py&lt;/code&gt;); never run against a real radar by Blip's author&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Enclosure&lt;/td&gt;
&lt;td&gt;None yet&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  Bill of materials
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Qty&lt;/th&gt;
&lt;th&gt;Part&lt;/th&gt;
&lt;th&gt;Notes&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;1&lt;/td&gt;
&lt;td&gt;ESP32-C3 Super Mini &lt;strong&gt;or&lt;/strong&gt; Waveshare ESP32-S3-Pico&lt;/td&gt;
&lt;td&gt;Either; both envs in &lt;code&gt;firmware/platformio.ini&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;1&lt;/td&gt;
&lt;td&gt;Ai-Thinker RD-03D — or an HLK-LD2450 (see &lt;em&gt;Radar options&lt;/em&gt;)&lt;/td&gt;
&lt;td&gt;5 V, ≥200 mA, UART 256000 baud, 3.3 V logic&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;1&lt;/td&gt;
&lt;td&gt;Old phone / tablet / laptop with a browser&lt;/td&gt;
&lt;td&gt;The display&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;0–1&lt;/td&gt;
&lt;td&gt;SSD1306 I2C OLED, 0.91" 128×32 or 0.96" 128×64&lt;/td&gt;
&lt;td&gt;Optional second display&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;0–1&lt;/td&gt;
&lt;td&gt;CDM324 (24 GHz) or HB100 (10.5 GHz) CW Doppler module&lt;/td&gt;
&lt;td&gt;Optional "Stetho"; HB100 if you want to run it alongside the RD-03D (see below)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;0–2&lt;/td&gt;
&lt;td&gt;MAX4466 electret-mic amp board (Adafruit 1063 / GY-MAX4466)&lt;/td&gt;
&lt;td&gt;The Stetho amplifier, modded: &lt;a href="https://github.com/The-Dorkknight/blip-radar/blob/main/hardware/max4466-mod.md" rel="noopener noreferrer"&gt;hardware/max4466-mod.md&lt;/a&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;0–1&lt;/td&gt;
&lt;td&gt;470 Ω, 100 nF, 220 µF, 3.5 mm socket&lt;/td&gt;
&lt;td&gt;Earphone output, &lt;a href="https://github.com/The-Dorkknight/blip-radar/blob/main/hardware/audio-out.md" rel="noopener noreferrer"&gt;hardware/audio-out.md&lt;/a&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;0–1&lt;/td&gt;
&lt;td&gt;MAX98357A I2S amp board + 4–8 Ω speaker&lt;/td&gt;
&lt;td&gt;Speaker output&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;—&lt;/td&gt;
&lt;td&gt;USB-C cable / 5 V supply, wire&lt;/td&gt;
&lt;td&gt;The C3's 5V pin can feed the radar for bench tests&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  Wiring
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fgnhufwdp4ngjwtnbhgla.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fgnhufwdp4ngjwtnbhgla.webp" alt="Wiring, ESP32-C3 Super Mini" width="800" height="585"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F1vd9edfhv4qsua1zucu4.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F1vd9edfhv4qsua1zucu4.webp" alt="Wiring, Waveshare ESP32-S3-Pico" width="800" height="585"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Signal&lt;/th&gt;
&lt;th&gt;C3 Super Mini&lt;/th&gt;
&lt;th&gt;S3-Pico&lt;/th&gt;
&lt;th&gt;Note&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Radar TX → ESP RX&lt;/td&gt;
&lt;td&gt;GPIO20&lt;/td&gt;
&lt;td&gt;GPIO12&lt;/td&gt;
&lt;td&gt;UART1, 256000 8N1&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Radar RX ← ESP TX&lt;/td&gt;
&lt;td&gt;GPIO21&lt;/td&gt;
&lt;td&gt;GPIO11&lt;/td&gt;
&lt;td&gt;only used to send the single/multi-target command&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;OLED SDA / SCL&lt;/td&gt;
&lt;td&gt;GPIO6 / GPIO7&lt;/td&gt;
&lt;td&gt;GPIO17 / GPIO18&lt;/td&gt;
&lt;td&gt;I2C 0x3C&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Stetho: MAX4466 OUT&lt;/td&gt;
&lt;td&gt;GPIO1 (ADC1)&lt;/td&gt;
&lt;td&gt;GPIO7 (ADC1)&lt;/td&gt;
&lt;td&gt;MAX4466 powered from 3V3&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Earphones (delta-sigma)&lt;/td&gt;
&lt;td&gt;GPIO10&lt;/td&gt;
&lt;td&gt;GPIO16&lt;/td&gt;
&lt;td&gt;via 470 Ω / 100 nF / 220 µF&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;I2S amp BCLK / LRC / DIN&lt;/td&gt;
&lt;td&gt;GPIO3 / 4 / 5&lt;/td&gt;
&lt;td&gt;GPIO13 / 14 / 15&lt;/td&gt;
&lt;td&gt;MAX98357A&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Radar 5V / GND&lt;/td&gt;
&lt;td&gt;5V / GND&lt;/td&gt;
&lt;td&gt;VBUS / GND&lt;/td&gt;
&lt;td&gt;never 3V3&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Pin numbers live in one place, &lt;code&gt;firmware/platformio.ini&lt;/code&gt;; the diagrams are drawn from it.&lt;/p&gt;

&lt;h3&gt;
  
  
  Stetho front end: CDM324 → modded MAX4466
&lt;/h3&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fuhmo9yot85bnnbz6x6ed.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fuhmo9yot85bnnbz6x6ed.webp" alt="CDM324 into a modded MAX4466 board" width="799" height="538"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;⚠ DC warning.&lt;/strong&gt; When you desolder the mic capsule and wire the radar's IF pin onto its pads, &lt;strong&gt;also remove the mic bias resistor&lt;/strong&gt;, or put a 10–47 µF DC-blocking cap in series with the IF lead. Otherwise ~1.5 mA of DC from VCC is pushed into the CDM324's mixer diode. Power the MAX4466 from &lt;strong&gt;3V3&lt;/strong&gt;, never 5 V. Full steps: &lt;a href="https://github.com/The-Dorkknight/blip-radar/blob/main/hardware/max4466-mod.md" rel="noopener noreferrer"&gt;hardware/max4466-mod.md&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;RD-03D + CDM324 at the same time?&lt;/strong&gt; They share the 24 GHz band. The RD-03D's sweep passes right over the CDM324's frequency, so expect a buzz in the Stetho audio and possibly a noisier RD-03D. Use an &lt;strong&gt;HB100&lt;/strong&gt; (10.5 GHz) for the Stetho, or fit a 5 V switch on the radar and choose &lt;em&gt;Radar off while listening&lt;/em&gt; (&lt;code&gt;PIN_RADAR_PWR&lt;/code&gt;). Details in &lt;a href="https://github.com/The-Dorkknight/blip-radar/blob/main/DESIGN_NOTES.md#115-running-the-rd-03d-and-a-cdm324-together" rel="noopener noreferrer"&gt;DESIGN_NOTES.md §11.5&lt;/a&gt;.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  One launcher: flash it, or try it first (double-click)
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Unzip, then double-click &lt;strong&gt;&lt;code&gt;Blip-Linux.sh&lt;/code&gt;&lt;/strong&gt; (choose &lt;em&gt;Run&lt;/em&gt;), &lt;strong&gt;&lt;code&gt;Blip-Windows.bat&lt;/code&gt;&lt;/strong&gt; or &lt;strong&gt;&lt;code&gt;Blip-Mac.command&lt;/code&gt;&lt;/strong&gt; (first time on a Mac: right-click → Open).&lt;/li&gt;
&lt;li&gt;Your browser opens the &lt;strong&gt;Blip Flasher&lt;/strong&gt; at &lt;code&gt;http://127.0.0.1:8080/&lt;/code&gt; (the red page). Only Python 3 is needed to open it.

&lt;ul&gt;
&lt;li&gt;Want to look around before you have hardware? The small &lt;strong&gt;Radar simulator ›&lt;/strong&gt; button (top right) opens the normal Blip page at &lt;code&gt;/sim&lt;/code&gt;, with three pretend people wandering about and a synthetic Stetho signal; its &lt;strong&gt;⚡ Flasher&lt;/strong&gt; button brings you back. Start the launcher with &lt;code&gt;--lan&lt;/code&gt; to open the simulator from your phone too (the terminal prints the address).&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;To flash a real board:&lt;/strong&gt; plug the board in with a USB &lt;em&gt;data&lt;/em&gt; cable, pick the board (C3 Super Mini / S3-Pico), the radar (RD-03D / LD2450) and how to flash, pick the port (or leave Auto-detect), press &lt;strong&gt;⚡ Flash firmware&lt;/strong&gt;.

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Quick flash (default, recommended):&lt;/strong&gt; writes the firmware that ships in &lt;code&gt;firmware/prebuilt/&lt;/code&gt;. Nothing is hidden: the page shows the file, version, build date, size and full SHA-256, checks the file against &lt;code&gt;manifest.json&lt;/code&gt; before writing, says whether it was built from exactly the source in your folder, and shows the exact esptool command. The first run installs only esptool into &lt;code&gt;.venv&lt;/code&gt; (about 40 MB). No Git, no compiler.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Build from source:&lt;/strong&gt; compiles your folder with PlatformIO so your own changes go in. The first build downloads the ESP32 toolchain (a few GB; needs the internet and &lt;strong&gt;Git&lt;/strong&gt;; 5–15 minutes), later builds are quick.&lt;/li&gt;
&lt;li&gt;The steps light up as it goes (Quick: Firmware → Flash tool → Upload → Boot; Source: Prepare → PlatformIO → Build → Upload → Boot) with a progress bar for the upload and the full log underneath (&lt;em&gt;Copy&lt;/em&gt; puts it on the clipboard for a bug report).&lt;/li&gt;
&lt;li&gt;If something fails, a tip explains the usual cause: serial-port permission on Linux, the board not in flashing mode (hold BOOT, tap RESET), no board or several boards found, no internet on the first run, Git missing, a port held by another program, your source newer than the bundled firmware.&lt;/li&gt;
&lt;li&gt;Afterwards it shows the board's serial output (no terminal needed). &lt;em&gt;Cancel&lt;/em&gt; stops any step.&lt;/li&gt;
&lt;li&gt;Rather use another tool? Any ESP32 flasher can write &lt;code&gt;firmware/prebuilt/blip-&amp;lt;env&amp;gt;-full.bin&lt;/code&gt; at address &lt;code&gt;0x0&lt;/code&gt; (they are also attached to each GitHub release).&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;Wire the radar as above and power up. The LED on the C3 blinks slowly until frames arrive, then fast.&lt;/li&gt;
&lt;li&gt;On the phone: join Wi-Fi &lt;strong&gt;&lt;code&gt;Blip-Radar&lt;/code&gt;&lt;/strong&gt;, password &lt;strong&gt;&lt;code&gt;blipblip&lt;/code&gt;&lt;/strong&gt;. The captive portal should open the scope; if not, browse to &lt;code&gt;http://192.168.4.1/&lt;/code&gt;. It is the same green page as the simulator.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fa72p680wiwbbhyw2i9g6.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fa72p680wiwbbhyw2i9g6.webp" alt="Blip Flasher" width="760" height="925"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The flasher only answers this computer (&lt;code&gt;127.0.0.1&lt;/code&gt; / &lt;code&gt;localhost&lt;/code&gt;): starting a flash needs a per-run token only the page has, and the board, radar and port are checked against fixed lists before anything runs. &lt;code&gt;--no-flash&lt;/code&gt; hides it entirely.&lt;/p&gt;

&lt;p&gt;Prefer the terminal? &lt;code&gt;cd firmware &amp;amp;&amp;amp; pio run -e c3-supermini -t upload&lt;/code&gt; (envs: &lt;code&gt;c3-supermini&lt;/code&gt;, &lt;code&gt;s3-pico&lt;/code&gt;, &lt;code&gt;c3-supermini-ld2450&lt;/code&gt;, &lt;code&gt;s3-pico-ld2450&lt;/code&gt;). After editing &lt;code&gt;web/index.html&lt;/code&gt; run &lt;code&gt;python3 tools/embed_web.py&lt;/code&gt; (the Flash tab does this for you).&lt;/p&gt;

&lt;h2&gt;
  
  
  Using it
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Screen&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fyiz5erqv62yymybz9l5q.webp" alt="Scope" width="640" height="1384"&gt;&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Scope&lt;/strong&gt; — ±60° sector, a ring every metre to 8 m (solid and labelled at 2, 4, 6, 8 m; faint at the odd metres), one coloured blip per tracked person, labelled with distance, then whether it is moving closer or away (cm/s), then a dim #number that is only an ID. Trails fade over 4 s. &lt;em&gt;Tracker ping&lt;/em&gt; is the original project's buzzer, done in the phone's speaker: faster and higher as the nearest person approaches. &lt;em&gt;Mirror L/R&lt;/em&gt; flips the picture if you mount the radar the other way round. &lt;em&gt;Standing / 🚶 Handheld&lt;/em&gt; switches handheld mode (below). &lt;em&gt;Show filtered&lt;/em&gt; draws what Blip is hiding as dashed rings, with the reason: ghost / clutter (Smart filter), near / edge / scenery (clutter options).&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F6a1f6xycte1akf5io6k8.webp" alt="Targets" width="640" height="1384"&gt;&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Targets&lt;/strong&gt; — the numbers behind the blips plus the raw JSON, handy when checking orientation.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F5ulrw3sdigt0q0jkbnzg.webp" alt="Stetho" width="640" height="1384"&gt;&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Stetho&lt;/strong&gt; — the optional CW Doppler channel. Presets: &lt;strong&gt;Heart&lt;/strong&gt; (S1/S2 heart sounds, 20–150 Hz), &lt;strong&gt;Lungs&lt;/strong&gt; (breath hiss), &lt;strong&gt;Voice&lt;/strong&gt; (throat/larynx buzz), &lt;strong&gt;Wide&lt;/strong&gt;, &lt;strong&gt;Raw&lt;/strong&gt; (plain Doppler warble). &lt;em&gt;Listen on phone&lt;/em&gt; plays it live; &lt;em&gt;Delay&lt;/em&gt; shows how far behind real time you are. Knocks are ducked and limited (red in the trace). &lt;strong&gt;Beat lock&lt;/strong&gt; quietens the gaps between beats once a rhythm is found; &lt;strong&gt;Record 10 s → WAV&lt;/strong&gt; saves a recording for &lt;code&gt;tools/analyze_wav.py&lt;/code&gt;; &lt;em&gt;Experiments&lt;/em&gt; has ADC averaging and the burst blanker (see &lt;a href="https://github.com/The-Dorkknight/blip-radar/blob/main/docs/STETHO_SIGNAL_NOTES.md" rel="noopener noreferrer"&gt;docs/STETHO_SIGNAL_NOTES.md&lt;/a&gt;). &lt;strong&gt;Not a medical device.&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fl6lph46qmv1r0e9yivon.webp" alt="Chest motion" width="640" height="1384"&gt;&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Chest motion&lt;/strong&gt; view — the 100 Hz motion trace with heart-rate and breathing rates from the pattern trackers (a &lt;code&gt;?&lt;/code&gt; means only roughly steady); &lt;em&gt;Beat thumps&lt;/em&gt; adds a synthetic thump on each detected beat.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fn88ayzk2c1mtbublzpys.webp" alt="Clean-up" width="640" height="1384"&gt;&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Clean-up on the node&lt;/strong&gt; — noise suppression, bump rejection, AGC or manual gain; &lt;strong&gt;Sound out of the node&lt;/strong&gt; — earphones or I2S amp, with its own volume.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0knl691vvysdkvqeiqww.webp" alt="Settings" width="640" height="1384"&gt;&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Settings&lt;/strong&gt; — single/multi target mode (also on the Radar page as &lt;em&gt;Track 1 / Track 3&lt;/em&gt;), &lt;strong&gt;Target filter&lt;/strong&gt; (Raw / Smart, experimental) with &lt;em&gt;Forget learned clutter&lt;/em&gt;, the &lt;strong&gt;clutter options&lt;/strong&gt; (below), &lt;strong&gt;Auto-sleep&lt;/strong&gt;, OLED (Auto-detect, or a fixed size), Stetho on/off, radar-off-while-listening, hotspot name and password (reboots).&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffo4sf74prk5l130vhzqb.webp" alt="About" width="640" height="1384"&gt;&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;About&lt;/strong&gt; — version, build date, board, pin map and counters.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fa72p680wiwbbhyw2i9g6.webp" alt="Flash" width="760" height="925"&gt;&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Flash&lt;/strong&gt; (simulator on a PC only) — pick board, radar, Quick flash or Build from source, and port; see exactly what will be written; watch the board boot.&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;OLED mirror. Each person is a circle with their distance inside: big digit = metres, small raised digit = tenths, so "1⁹" = 1.9 m. The nearest person is solid. There's no status row, so the scope fills the panel. ◆ top-left = Smart filter on; Wi-Fi icon + number top-right = phones connected; the IP shows bottom-left only until a phone joins. These are exact renders of the firmware's drawing code, not photos:&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fim2y5ta3t33am8mcbzck.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fim2y5ta3t33am8mcbzck.webp" alt="OLED 128x64" width="660" height="340"&gt;&lt;/a&gt; &lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3jx0ka0mitptafcmkgai.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3jx0ka0mitptafcmkgai.webp" alt="OLED 128x64, no phone yet" width="660" height="340"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fgaophfcxsyex9pl74wvy.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fgaophfcxsyex9pl74wvy.webp" alt="OLED 128x32" width="660" height="180"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Clutter options (Settings, added after the first street test)
&lt;/h3&gt;

&lt;p&gt;All off by default except &lt;em&gt;Drop frozen targets&lt;/em&gt;, all work in Raw and Smart mode, all &lt;strong&gt;experimental&lt;/strong&gt;:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Option&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;th&gt;When to use it&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Ignore anything closer than&lt;/strong&gt; 0.1–1.0 m&lt;/td&gt;
&lt;td&gt;drops contacts nearer than that&lt;/td&gt;
&lt;td&gt;a persistent contact at 0.2–0.3 m (your own hand, wires near the antennas); 0.5 m is the module's own blind zone anyway&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Ignore the edges&lt;/strong&gt; ±65° → ±30°&lt;/td&gt;
&lt;td&gt;drops contacts far off to the side&lt;/td&gt;
&lt;td&gt;edge-of-view contacts are more often reflections&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Drop frozen targets&lt;/strong&gt; after 5–60 s (default 10 s, on)&lt;/td&gt;
&lt;td&gt;drops a contact reported at exactly the same x and y to the millimetre for that long; back as soon as it moves&lt;/td&gt;
&lt;td&gt;the module locked onto a phantom that stays after everyone has gone&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Handheld mode&lt;/strong&gt; (also on the Radar page)&lt;/td&gt;
&lt;td&gt;while you carry it, works out your walking speed from the contacts and hides the ones closing in at exactly that speed: lamp posts, parked cars, walls. People moving on their own stay. Stops the Smart filter learning clutter.&lt;/td&gt;
&lt;td&gt;walking about with it outdoors. Needs at least two fixed things in view to lock on; a lone lamp post is only hidden for 1.5 s after that&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Strict ghost test&lt;/strong&gt; (Smart only)&lt;/td&gt;
&lt;td&gt;also hides a contact whose range has moved in step with a nearer one for 3 s, at any bearing, with a similar speed&lt;/td&gt;
&lt;td&gt;reflections off side walls or cars. Can now and then hide a second person walking exactly like the first (about 1 frame in 10 in the tests)&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Tested on synthetic scenes only (&lt;code&gt;tests/test_gate.cpp&lt;/code&gt;, &lt;code&gt;tests/test_tracker.cpp&lt;/code&gt;), not on a real walk yet. The simulator has a &lt;strong&gt;street walk&lt;/strong&gt; button in Settings to try handheld mode. Background in &lt;a href="https://github.com/The-Dorkknight/blip-radar/blob/main/docs/RADAR_TRACKING_NOTES.md" rel="noopener noreferrer"&gt;docs/RADAR_TRACKING_NOTES.md&lt;/a&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  HLK-LD2450 custom firmware (experimental, brick risk)
&lt;/h3&gt;

&lt;p&gt;The flasher's third tab, &lt;strong&gt;☢ LD2450 custom firmware&lt;/strong&gt;, runs &lt;a href="https://github.com/earlynerd/HLK-LD2450_FW" rel="noopener noreferrer"&gt;earlynerd/HLK-LD2450_FW&lt;/a&gt; (MIT), third-party firmware that turns the LD2450 from "3 targets" into a raw range-bin stream: a Doppler spectrum in every range bin, and slow-time phase fine enough that its author shows breathing and heartbeats at about 0.8 m. &lt;strong&gt;It writes to the radar itself, can brick it, and Blip's author has not tried it yet. Use a spare module.&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Before: update the LD2450 to stock &lt;strong&gt;V2.14&lt;/strong&gt; with Hi-Link's &lt;strong&gt;HLKRadarTool&lt;/strong&gt; phone app (Bluetooth), and set its UART to &lt;strong&gt;256000&lt;/strong&gt; baud in the app. Bluetooth is only for this step; the firmware itself goes over the 4-pin plug.&lt;/li&gt;
&lt;li&gt;Wire the 4-pin plug to a &lt;strong&gt;3.3 V USB-serial adapter&lt;/strong&gt; (5V, GND, radar TX → adapter RX, radar RX → adapter TX), not to the ESP32.&lt;/li&gt;
&lt;li&gt;Then: &lt;strong&gt;Probe&lt;/strong&gt; (writes nothing) → &lt;strong&gt;Install&lt;/strong&gt; (type BRICK) → &lt;strong&gt;Open its live viewer&lt;/strong&gt;. &lt;strong&gt;Restore stock&lt;/strong&gt; puts Hi-Link V2.14 back (it then runs at 9600 baud; set 256000 in the app again).&lt;/li&gt;
&lt;li&gt;The data comes out of the radar's &lt;strong&gt;own USB pins&lt;/strong&gt; (D+, D−, GND test pads on the board), not the 4-pin plug, so you need a USB cable soldered to those pads to see it. While the custom firmware is on, Blip's ESP32 can't use that radar.&lt;/li&gt;
&lt;li&gt;First use downloads the project at a pinned commit (about 50 MB) into &lt;code&gt;.cache/&lt;/code&gt;; both images are checked against pinned SHA-256 values before anything runs.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  How it works
&lt;/h2&gt;

&lt;p&gt;Short version: the ESP32 parses 30-byte frames from the radar, keeps the latest three targets, and pushes them as JSON over a WebSocket 20 times a second to whoever is on the hotspot; the browser draws the scope on a canvas. The long version, including the frame format, is in &lt;a href="https://github.com/The-Dorkknight/blip-radar/blob/main/HOW_IT_WORKS.md" rel="noopener noreferrer"&gt;HOW_IT_WORKS.md&lt;/a&gt;. The research — radar physics, why the RD-03D can't be a stethoscope and what can, pin choices — is in &lt;a href="https://github.com/The-Dorkknight/blip-radar/blob/main/DESIGN_NOTES.md" rel="noopener noreferrer"&gt;DESIGN_NOTES.md&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Radar options
&lt;/h2&gt;

&lt;p&gt;Blip was written for the &lt;strong&gt;Ai-Thinker RD-03D&lt;/strong&gt; (1.25 mm 4-pin socket, 8 m, up to 3 targets). The &lt;strong&gt;HLK-LD2450&lt;/strong&gt; (Hi-Link, 24 GHz, about 6 m, up to 3 targets with X/Y position) is supported too:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;It sends the same 30-byte frame (&lt;code&gt;AA FF 03 00&lt;/code&gt;, three 8-byte slots, &lt;code&gt;55 CC&lt;/code&gt;), so the same parser reads it.&lt;/li&gt;
&lt;li&gt;It has no RD-03D configuration commands, so Blip skips the mode command. &lt;em&gt;Single (nearest)&lt;/em&gt; is picked in firmware instead.&lt;/li&gt;
&lt;li&gt;Use the &lt;code&gt;c3-supermini-ld2450&lt;/code&gt; or &lt;code&gt;s3-pico-ld2450&lt;/code&gt; build environment (or add &lt;code&gt;-DRADAR_LD2450=1&lt;/code&gt;). &lt;code&gt;/api/status&lt;/code&gt; reports &lt;code&gt;"model":"ld2450"&lt;/code&gt; while it is running.&lt;/li&gt;
&lt;li&gt;A 4-pin plug (5V, GND, TX, RX; listings disagree on the pitch, so check yours). Check the labels on the module: its &lt;strong&gt;TX&lt;/strong&gt; goes to the board's &lt;strong&gt;RX&lt;/strong&gt; pin, and its &lt;strong&gt;RX&lt;/strong&gt; to the board's &lt;strong&gt;TX&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;Weaker than the RD-03D on people who sit very still (the vendor notes this beyond about 2 m), and range is about 6 m.&lt;/li&gt;
&lt;li&gt;Still 24 GHz, so it clashes with a CDM324 in the same way. Use an HB100 for Stetho, or the radar power switch.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Not tested on hardware yet.&lt;/strong&gt; Check the angle direction on the Scope first (&lt;em&gt;Mirror L/R&lt;/em&gt; in Settings if left and right are swapped).&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Repository map
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;firmware/        PlatformIO project (src/main.cpp, include/rd03d.h, gate.h, tracker.h, audio_dsp.h, stetho.h, oled.h, oled_draw.h, web_ui.h)
firmware/prebuilt/  bundled whole-chip images for the four builds + manifest.json (checksums, version, source hash)
web/index.html   the phone UI (single file; embedded gzipped into the firmware)
web/flasher.html   the flasher (served by the PC tool only; never in the firmware)
web/flash_home.html  the red home page the launcher opens, with the flasher inlined
tools/           sim_radar.py, flasher.py, serial_tail.py, make_prebuilt.py, make_pdf.py, embed_web.py, draw_illustrations.py, rd03d_py.py, tracker_py.py, gate_py.py, analyze_wav.py
tests/           native C++ tests (parser, audio DSP, tracker, OLED render), Python parser + tracker cross-check, flasher (with fake_pio.py), launcher test
hardware/        wiring, MAX4466 mod (with the DC warning), earphone / I2S amp outputs
examples/        synthetic RD-03D frames for replay; dsp_before.wav / dsp_after.wav
docs/            GitHub Pages site + images + Blip-manual.pdf (built by tools/make_pdf.py)
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;






&lt;p&gt;&lt;strong&gt;Source code:&lt;/strong&gt; &lt;a href="https://github.com/The-Dorkknight/blip-radar" rel="noopener noreferrer"&gt;The-Dorkknight/blip-radar on GitHub&lt;/a&gt;&lt;/p&gt;

</description>
    </item>
    <item>
      <title>Bat Tree-Bar Proe</title>
      <dc:creator>Matthew Armstrong</dc:creator>
      <pubDate>Wed, 11 Nov 2026 16:34:28 +0000</pubDate>
      <link>https://dev.to/open-madscience/bat-tree-bar-proe-1im8</link>
      <guid>https://dev.to/open-madscience/bat-tree-bar-proe-1im8</guid>
      <description>&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F69001zbvpcblxfiyln23.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F69001zbvpcblxfiyln23.webp" alt="The Bat Tree-Bar Saloon"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;WORK IN PROGRESS: MANY FEATURES MAY BE BROKEN&lt;/strong&gt; · version &lt;code&gt;0.2.0-wip&lt;/code&gt; · source-available · non-commercial&lt;/p&gt;

&lt;p&gt;Battery percentage and time left &lt;strong&gt;at a glance&lt;/strong&gt;, with a click for the details: charge/discharge current, power, voltage and battery health. In the spirit of BatteryBar, built to run on &lt;strong&gt;Windows 11 and Linux&lt;/strong&gt; (and, experimentally, macOS).&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Licence and no warranty.&lt;/strong&gt; Source-available under the &lt;a href="https://github.com/The-Dorkknight/bat-tree-bar-proe/blob/main/LICENSE.md" rel="noopener noreferrer"&gt;PolyForm Noncommercial License 1.0.0&lt;/a&gt;: free to use, change and share for non-commercial purposes; commercial use needs separate permission. Provided as is, with no warranty and no liability. This is not legal advice. Bat Tree-Bar Proe only &lt;em&gt;reads&lt;/em&gt; your battery; it never changes a setting.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;📘 The whole manual and research notes as one PDF: &lt;a href="https://github.com/The-Dorkknight/bat-tree-bar-proe/blob/main/docs/BatTreeBarProe-manual.pdf" rel="noopener noreferrer"&gt;docs/BatTreeBarProe-manual.pdf&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/bat-tree-bar-proe/blob/main/CHANGELOG.md" rel="noopener noreferrer"&gt;Changelog&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/bat-tree-bar-proe/blob/main/docs/HISTORY.md" rel="noopener noreferrer"&gt;Build history&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/bat-tree-bar-proe/blob/main/HOW_IT_WORKS.md" rel="noopener noreferrer"&gt;How it works&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/bat-tree-bar-proe/blob/main/DESIGN_NOTES.md" rel="noopener noreferrer"&gt;Design notes&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  A tribute to BatteryBar
&lt;/h2&gt;

&lt;p&gt;The name is a pun on &lt;strong&gt;BatteryBar Pro&lt;/strong&gt;, the Windows taskbar battery meter made by &lt;strong&gt;Osiris Development&lt;/strong&gt; (&lt;a href="http://osirisdevelopment.com/BatteryBar/" rel="noopener noreferrer"&gt;osirisdevelopment.com/BatteryBar&lt;/a&gt;). It showed the battery level and the time left or time to full right on the taskbar, with a Pro version adding graphs, warnings and more. That is the idea this project follows, rebuilt from scratch for Windows 11 and Linux (the author reports the original does not run there). This project is &lt;strong&gt;not affiliated with Osiris Development&lt;/strong&gt;, shares no code with BatteryBar and is not a substitute for it; if BatteryBar still works for you, please go and use the original.&lt;/p&gt;

&lt;p&gt;Project started 2026-11-11 11:01. Renamed from "ChargeGlance" the same day, before any public release.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fpbggygc32tyzf0hlo7gb.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fpbggygc32tyzf0hlo7gb.webp" alt="The bar while charging"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What it looks like
&lt;/h2&gt;

&lt;p&gt;A tiny always-on-top bar shows the battery icon, the percentage and the time left (or time until full). &lt;strong&gt;Click it&lt;/strong&gt; and the details open. Drag it anywhere; it remembers where you left it. Right-click for the menu.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;On battery&lt;/th&gt;
&lt;th&gt;Charging&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fes34ne9o8zqeej1i9n70.webp" alt="popup on battery"&gt;&lt;/td&gt;
&lt;td&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fi5ci04bfhuy16na78o3k.webp" alt="popup charging"&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;em&gt;These screenshots use the built-in **simulated&lt;/em&gt;* battery (&lt;code&gt;--demo&lt;/code&gt;), because they were taken in a sandbox with no battery. The layout is the real one.*&lt;/p&gt;

&lt;h2&gt;
  
  
  Using it
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;You do&lt;/th&gt;
&lt;th&gt;You get&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Look at the bar&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;73%  4h 24m&lt;/code&gt; on battery; &lt;code&gt;64%  34m&lt;/code&gt; (blue icon with a bolt) while charging; &lt;code&gt;100%  full&lt;/code&gt; when full; &lt;code&gt;80%  on AC&lt;/code&gt; when plugged in but not charging&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Click the bar&lt;/td&gt;
&lt;td&gt;Popup: charge or discharge current, power, voltage, energy now/full, &lt;strong&gt;health&lt;/strong&gt; (full-charge capacity vs new), cycle count, temperature, mains status, a 30-minute power graph. Click again or press Esc to close&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Drag the bar&lt;/td&gt;
&lt;td&gt;Moves it; position is saved&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Right-click the bar&lt;/td&gt;
&lt;td&gt;Details, Start at login, Always on top, Allow over the taskbar (experimental), Move to bottom right, Quit&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The popup says in brackets where the time estimate came from: &lt;code&gt;rate&lt;/code&gt; (energy left divided by smoothed power), &lt;code&gt;trend&lt;/code&gt; (percent slope over the last few minutes) or &lt;code&gt;os&lt;/code&gt; (the number your operating system reports).&lt;/p&gt;

&lt;h3&gt;
  
  
  Command line
&lt;/h3&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 src/main.py --version
python3 src/main.py --once     &lt;span class="c"&gt;# one reading as JSON: paste this in bug reports&lt;/span&gt;
python3 src/main.py --text     &lt;span class="c"&gt;# no window, one line every 5 s&lt;/span&gt;
python3 src/main.py --demo     &lt;span class="c"&gt;# simulated battery (also: --demo charging, --demo full)&lt;/span&gt;
python3 src/main.py --autostart on&lt;span class="o"&gt;|&lt;/span&gt;off&lt;span class="o"&gt;|&lt;/span&gt;status   &lt;span class="c"&gt;# start at login&lt;/span&gt;
python3 src/main.py --force    &lt;span class="c"&gt;# start even if it thinks a copy is running&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Settings and a small log live in &lt;code&gt;~/.config/bat-tree-bar-proe/&lt;/code&gt; (Linux), &lt;code&gt;%APPDATA%\Bat Tree-Bar Proe\&lt;/code&gt; (Windows) or &lt;code&gt;~/Library/Application Support/Bat Tree-Bar Proe/&lt;/code&gt; (Mac).&lt;/p&gt;

&lt;h2&gt;
  
  
  What works, what does not
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;Linux&lt;/th&gt;
&lt;th&gt;Windows 11&lt;/th&gt;
&lt;th&gt;macOS&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Code written&lt;/td&gt;
&lt;td&gt;yes&lt;/td&gt;
&lt;td&gt;yes&lt;/td&gt;
&lt;td&gt;yes&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Reading logic unit-tested&lt;/td&gt;
&lt;td&gt;yes, with fake battery folders&lt;/td&gt;
&lt;td&gt;parser only, on a hand-written sample&lt;/td&gt;
&lt;td&gt;parser only, on a hand-written sample&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Run on a real battery&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;not yet&lt;/strong&gt; (built in a sandbox with no battery)&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;not yet&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;not yet&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Window runs&lt;/td&gt;
&lt;td&gt;yes (Xvfb, simulated battery)&lt;/td&gt;
&lt;td&gt;not run&lt;/td&gt;
&lt;td&gt;not run&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Graphical installer / uninstaller&lt;/td&gt;
&lt;td&gt;yes (Xvfb, into a temp folder)&lt;/td&gt;
&lt;td&gt;code written, not run (Start Menu shortcuts, Run key, Apps list entry)&lt;/td&gt;
&lt;td&gt;code written, not run (LaunchAgent, app folder)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Fuzz and edge-case tests (garbage readings, suspend, clock jumps, corrupt settings, HiDPI)&lt;/td&gt;
&lt;td&gt;yes&lt;/td&gt;
&lt;td&gt;parser only&lt;/td&gt;
&lt;td&gt;parser only&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  Project layout
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;src/main.py&lt;/code&gt; window and command line · &lt;code&gt;src/battery_source.py&lt;/code&gt; Linux, Windows, macOS readers · &lt;code&gt;src/estimator.py&lt;/code&gt; time estimates · &lt;code&gt;tests/&lt;/code&gt; unit tests · &lt;code&gt;tools/&lt;/code&gt; screenshot rig, PDF and history scripts.&lt;/p&gt;

&lt;p&gt;Run the tests: &lt;code&gt;python3 -m unittest discover -s tests -v&lt;/code&gt;&lt;/p&gt;

&lt;p&gt;Created by Matthew Armstrong (&lt;a href="https://github.com/The-Dorkknight" rel="noopener noreferrer"&gt;The-Dorkknight&lt;/a&gt;).&lt;/p&gt;




&lt;p&gt;&lt;strong&gt;Source code:&lt;/strong&gt; &lt;a href="https://github.com/The-Dorkknight/bat-tree-bar-proe" rel="noopener noreferrer"&gt;The-Dorkknight/bat-tree-bar-proe on GitHub&lt;/a&gt;&lt;/p&gt;

</description>
      <category>linux</category>
      <category>opensource</category>
      <category>tools</category>
    </item>
    <item>
      <title>MultiPlate for Cura</title>
      <dc:creator>Matthew Armstrong</dc:creator>
      <pubDate>Tue, 10 Nov 2026 23:26:47 +0000</pubDate>
      <link>https://dev.to/open-madscience/multiplate-for-cura-3b8d</link>
      <guid>https://dev.to/open-madscience/multiplate-for-cura-3b8d</guid>
      <description>&lt;p&gt;&lt;strong&gt;WORK IN PROGRESS: MANY FEATURES MAY BE BROKEN&lt;/strong&gt; (0.4.1-wip, proof of concept)&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Source-available · Non-commercial. Licensed under the &lt;a href="https://github.com/The-Dorkknight/cura-multiplate/blob/main/LICENSE.md" rel="noopener noreferrer"&gt;PolyForm Noncommercial License 1.0.0&lt;/a&gt;, no warranty, no liability. Not affiliated with UltiMaker or Bambu Lab.&lt;br&gt;
📘 The whole manual and research notes as one PDF: &lt;a href="https://github.com/The-Dorkknight/cura-multiplate/blob/main/docs/MultiPlate-manual.pdf" rel="noopener noreferrer"&gt;docs/MultiPlate-manual.pdf&lt;/a&gt;&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Bambu-Studio-style &lt;strong&gt;multi-plate projects for Cura&lt;/strong&gt;: a &lt;strong&gt;floating plate bar&lt;/strong&gt; (one small stacked-plates button that opens into the bar; drag it, flip it vertical) with a &lt;strong&gt;+&lt;/strong&gt; button and one button per plate that you can label (&lt;code&gt;P1 mechanism&lt;/code&gt;, &lt;code&gt;P2 body&lt;/code&gt;...), a &lt;strong&gt;slice queue&lt;/strong&gt;, per-plate &lt;strong&gt;sliced&lt;/strong&gt; flags, &lt;strong&gt;per-plate print settings&lt;/strong&gt; (the ⚙ button), &lt;strong&gt;autosave&lt;/strong&gt;, and a &lt;strong&gt;resume prompt&lt;/strong&gt; when Cura starts.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Project started 2026-11-10 12:55.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F81lkdi5mshnmr9l9u8k5.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F81lkdi5mshnmr9l9u8k5.webp" alt="The plate button, the open bar and the vertical bar" width="120" height="80"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Frh9qj5913v1m69vhfqq8.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Frh9qj5913v1m69vhfqq8.webp" alt="The plate bar" width="800" height="44"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F07to9wwozmxodk29njwd.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F07to9wwozmxodk29njwd.webp" alt="The vertical bar" width="330" height="330"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;The real &lt;code&gt;PlateBar.qml&lt;/code&gt;, drawn with a stand-in for Cura's theme (this picture was NOT taken inside Cura). Green dot = sliced, yellow = changed since slicing, orange = not sliced yet, red = failed, grey = empty. The tick box adds a plate to the queue.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Use it
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;You want to&lt;/th&gt;
&lt;th&gt;Do this&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Open / fold the bar&lt;/td&gt;
&lt;td&gt;The stacked-plates button (top left, under the printer name) opens the bar; &lt;strong&gt;–&lt;/strong&gt; folds it back. A pulsing green dot on the button means the queue is running&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Move the bar&lt;/td&gt;
&lt;td&gt;Drag the &lt;strong&gt;⠿&lt;/strong&gt; grip (or drag the folded button). It stays inside the window and remembers where you left it&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Vertical or horizontal&lt;/td&gt;
&lt;td&gt;Press &lt;strong&gt;⇅ / ⇄&lt;/strong&gt; in the bar&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Add a plate&lt;/td&gt;
&lt;td&gt;Press &lt;strong&gt;+&lt;/strong&gt; in the bar&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Switch plate&lt;/td&gt;
&lt;td&gt;Click its button. Only that plate's objects are in Cura; the others wait off to the side and come back when you click their button&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Label a plate&lt;/td&gt;
&lt;td&gt;Double-click it (or right-click &amp;gt; Label) and type e.g. &lt;code&gt;mechanism&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Put objects on a plate&lt;/td&gt;
&lt;td&gt;Select them, right-click the plate button &amp;gt; &lt;strong&gt;Move selected objects here&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Give a plate its own settings&lt;/td&gt;
&lt;td&gt;Press &lt;strong&gt;⚙&lt;/strong&gt;, pick a setting (e.g. Infill Density, Generate Support), type a value, &lt;strong&gt;Add&lt;/strong&gt;. It is written into every object on that plate; a badge like &lt;code&gt;⚙2&lt;/code&gt; shows on the plate button. Remove it with &lt;strong&gt;×&lt;/strong&gt; and the objects go back to the print profile&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Queue plates&lt;/td&gt;
&lt;td&gt;Tick the small box on each plate button&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Slice the queue&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Slice queue (N)&lt;/strong&gt;. Plates are sliced one after another, in plate order, with progress and elapsed time shown. Already-sliced plates are skipped&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Mark a plate sliced by hand&lt;/td&gt;
&lt;td&gt;Right-click &amp;gt; &lt;strong&gt;Mark as sliced&lt;/strong&gt; (useful if you sliced it elsewhere)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Reorder plates&lt;/td&gt;
&lt;td&gt;Right-click &amp;gt; Move plate left / right (the queue follows plate order)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Get the G-code&lt;/td&gt;
&lt;td&gt;The queue writes one file per plate into the autosave folder &lt;code&gt;MultiPlate/autosave/gcode/&lt;/code&gt; (e.g. &lt;code&gt;01-mechanism.gcode&lt;/code&gt;); Cura's normal &lt;strong&gt;Save&lt;/strong&gt; also works on the active plate&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Save a named project&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;⋯ &amp;gt; Save project as...&lt;/strong&gt; makes one &lt;code&gt;.multiplate&lt;/code&gt; file (plates, labels, objects, flags, G-code)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Come back after a crash or restart&lt;/td&gt;
&lt;td&gt;Autosave runs every minute when something changed. At startup, if your last session had two or more plates, Cura shows &lt;strong&gt;Resume / Start fresh&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;A plate turns &lt;strong&gt;yellow (changed since slicing)&lt;/strong&gt; automatically when you move or edit an object on it or change the print profile.&lt;/p&gt;

&lt;h2&gt;
  
  
  What exists and what does not
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Feature&lt;/th&gt;
&lt;th&gt;Status&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Floating plate bar (button, drag, flip, fold), labels, + button, switching&lt;/td&gt;
&lt;td&gt;Written in 0.3; drawn, clicked and dragged in a Qt test, not yet in real Cura (the 0.2 bar was seen working in real Cura on a screenshot)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Slice queue with sliced / changed / failed flags&lt;/td&gt;
&lt;td&gt;Written; tested against a stand-in for Cura's slicing back-end, not the real engine&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Per-plate G-code files&lt;/td&gt;
&lt;td&gt;Written; same caveat&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Autosave + resume prompt + &lt;code&gt;.multiplate&lt;/code&gt; projects&lt;/td&gt;
&lt;td&gt;Written; round trip tested with stand-in scene nodes&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Per-plate print settings (supports on plate 1, off on plate 2)&lt;/td&gt;
&lt;td&gt;Written in 0.2 on top of Cura's per-object settings; tested against stand-ins only. 300 of Cura's ~740 settings allow it (see Known limitations)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Sending the queue to a printer (Bambu LAN etc.)&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;Not built&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Per-plate preview thumbnails&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;Not built&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;




&lt;p&gt;&lt;strong&gt;Source code:&lt;/strong&gt; &lt;a href="https://github.com/The-Dorkknight/cura-multiplate" rel="noopener noreferrer"&gt;The-Dorkknight/cura-multiplate on GitHub&lt;/a&gt;&lt;/p&gt;

</description>
      <category>github</category>
      <category>software</category>
      <category>tools</category>
    </item>
    <item>
      <title>Hybrid Supports for UltiMaker Cura</title>
      <dc:creator>Matthew Armstrong</dc:creator>
      <pubDate>Tue, 10 Nov 2026 22:41:25 +0000</pubDate>
      <link>https://dev.to/open-madscience/hybrid-supports-for-ultimaker-cura-5a6n</link>
      <guid>https://dev.to/open-madscience/hybrid-supports-for-ultimaker-cura-5a6n</guid>
      <description>&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;WORK IN PROGRESS: MANY FEATURES MAY BE BROKEN&lt;/strong&gt; (proof of concept, &lt;code&gt;0.6.6-wip&lt;/code&gt;)&lt;/p&gt;

&lt;p&gt;Sliced inside Cura 5.13.0 (Linux AppImage) with the &lt;strong&gt;Elegoo Neptune 4&lt;/strong&gt; profile. &lt;strong&gt;Not yet test-printed.&lt;/strong&gt; Windows and macOS untested. Coming from 0.1.x? See What changed.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;&lt;strong&gt;Licence:&lt;/strong&gt; &lt;a href="https://github.com/The-Dorkknight/cura-hybrid-supports/blob/main/LICENSE.md" rel="noopener noreferrer"&gt;PolyForm Noncommercial 1.0.0&lt;/a&gt;. Source-available, free for personal and other non-commercial use; commercial use needs permission. Provided as-is, with no warranty and no liability. This is not legal advice.&lt;/p&gt;

&lt;p&gt;A Cura tool plugin that supports tall overhangs with &lt;strong&gt;one hollow pillar per overhang&lt;/strong&gt; instead of a forest of solid columns. Each pillar stands on a clear spot (the plate or the model), has &lt;strong&gt;base gussets&lt;/strong&gt; that are kept clear of the model, and ends in a &lt;strong&gt;hollow star of lobes&lt;/strong&gt;: a pseudo-cone whose arms flare out at 45° towards the corners of the overhang. Every layer is one closed outline, and the gaps between the lobes are short enough to bridge; extra lobes are added only where a bridge would be too long. A plain hollow cone with spokes inside is still there as an option.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fcura-hybrid-supports%2Fmain%2Fdocs%2Fimages%2Fcura-panel.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fcura-hybrid-supports%2Fmain%2Fdocs%2Fimages%2Fcura-panel.webp" alt="The Hybrid Supports panel in Cura 5.13 with the Neptune 4 profile"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fcura-hybrid-supports%2Fmain%2Fdocs%2Fimages%2Fcura-preview.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fcura-hybrid-supports%2Fmain%2Fdocs%2Fimages%2Fcura-preview.webp" alt="A box standing on end, sliced: one pillar with four-plus-one lobes under the inner ceiling, pillars under the ribs"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Cura 5.13.0 screenshots of a synthetic test part shaped like an "EN4 Rugged Box" (70 × 40 × 120 mm, standing on end, open on one side, with ribs, hinge knuckles and a latch lug), taken on a headless test machine with software rendering.&lt;/p&gt;

&lt;h2&gt;
  
  
  What a real slice looks like
&lt;/h2&gt;

&lt;p&gt;Five layers from the gcode Cura 5.13 produced for that box (Elegoo Neptune 4 profile, 0.2 mm, &lt;em&gt;Generate Support&lt;/em&gt; off, Support Wall Line Count 0):&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fcura-hybrid-supports%2Fmain%2Fdocs%2Fimages%2Fsliced-layers.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fcura-hybrid-supports%2Fmain%2Fdocs%2Fimages%2Fsliced-layers.webp" alt="Sliced layers: hollow rings low down, the lobes growing out, an X of hollow lobes under the ceiling"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Low down every support is a hollow ring. Higher up, hollow lobes grow out of the centre ring, and just under the ceiling they reach the four corners, with a fifth, shorter lobe towards the open front edge (the only edge not resting on a wall). Each lobe prints as a closed loop, even with Cura's default of 0 support walls. An automated check confirmed that every pillar prints on 100 % of its layers (the 0.1.x "tube in mid-air" problem).&lt;/p&gt;

&lt;h3&gt;
  
  
  How it compares (the box above, Cura 5.13 estimates, Neptune 4 profile)
&lt;/h3&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Support for the same box&lt;/th&gt;
&lt;th&gt;Print time&lt;/th&gt;
&lt;th&gt;Support filament&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Hybrid Supports, Lobes head&lt;/strong&gt; (default, support walls 0)&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;149 min&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;4.9 m&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Hybrid Supports, Cone head with cross&lt;/td&gt;
&lt;td&gt;151 min&lt;/td&gt;
&lt;td&gt;5.4 m&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Lobes with Support Wall Line Count 1&lt;/td&gt;
&lt;td&gt;168 min&lt;/td&gt;
&lt;td&gt;7.5 m&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Cone with Support Wall Line Count 1&lt;/td&gt;
&lt;td&gt;169 min&lt;/td&gt;
&lt;td&gt;7.8 m&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Cura tree support&lt;/td&gt;
&lt;td&gt;151 min&lt;/td&gt;
&lt;td&gt;6.5 m&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Cura normal support&lt;/td&gt;
&lt;td&gt;190 min&lt;/td&gt;
&lt;td&gt;22.3 m&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;These numbers are from 0.3, when the towers were Cura support meshes. Since 0.6.1 Lobes, Cone and Branch print as hollow walls (see below); they have not been re-measured on the box.&lt;/p&gt;

&lt;p&gt;On this part the Lobes head uses about 78 % less support filament than Cura's normal support and saves about 40 minutes. Against Cura's &lt;strong&gt;tree&lt;/strong&gt; support it is about 25 % less support filament and about the same time. Against the cone it saves about 10 % of the support filament and a minute. 1 support wall costs about 18 minutes and isn't needed for the lobes. These are Cura's own estimates for one synthetic part; nothing here has been printed, and tree support may well do better on other shapes.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fcura-hybrid-supports%2Fmain%2Fdocs%2Fimages%2Frender-lobes.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fcura-hybrid-supports%2Fmain%2Fdocs%2Fimages%2Frender-lobes.webp" alt="Geometry render of the same box with the Lobes head"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;📄 &lt;strong&gt;The research paper:&lt;/strong&gt; &lt;a href="https://github.com/The-Dorkknight/cura-hybrid-supports/blob/main/docs/paper/putting-plastic-on-a-pedestal.pdf" rel="noopener noreferrer"&gt;&lt;em&gt;Putting Plastic on a Pedestal: experimental methods of plastic waste reduction for 3D printing&lt;/em&gt;&lt;/a&gt; (a defensive publication describing these supports; physically untested; its own repository is &lt;a href="https://github.com/The-Dorkknight/putting-plastic-on-a-pedestal" rel="noopener noreferrer"&gt;putting-plastic-on-a-pedestal&lt;/a&gt;).&lt;/p&gt;

&lt;h3&gt;
  
  
  New in 0.6.4: big parts, measured
&lt;/h3&gt;

&lt;p&gt;Sliced on 200 x 200 mm parts with long stems (Cura 5.13, Neptune 4; not yet printed): Hybrid lobes use 84 % less support plastic than Cura's normal support on a 150 mm-tall part, and 24 to 26 % less than Cura's tree support (but 29 % more than tree on a 50 mm-tall part, so the saving is a tall-part effect). Full table in &lt;a href="https://github.com/The-Dorkknight/cura-hybrid-supports/blob/main/docs/RESEARCH.md" rel="noopener noreferrer"&gt;docs/RESEARCH.md&lt;/a&gt; section 16. Also fixes cone shells that flared past the edge of big overhangs.&lt;/p&gt;

&lt;h3&gt;
  
  
  New in 0.6.3: an interface cap on every support
&lt;/h3&gt;

&lt;p&gt;From the review of 0.6.2: the towers had no support interface, so the part's underside would have printed straight onto a few walls and drooped between them. Every Lobes, Cone, Branch, Solid and Conform support (and Tethers that fall back to lobes) now gets the same &lt;strong&gt;4-layer cap stack as the tethers&lt;/strong&gt;, 0.8 mm under the overhang:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;a lattice (3 mm spacing) bridging between the tower's walls, printed at &lt;em&gt;Bridge speed&lt;/em&gt; (20 mm/s) with the fan at 100 %,&lt;/li&gt;
&lt;li&gt;a second lattice the other way,&lt;/li&gt;
&lt;li&gt;a full layer of concentric rings at 50 %,&lt;/li&gt;
&lt;li&gt;a solid (100 %) layer, then the usual contact gap (0.2 mm) and the part.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;The towers now stop at the cap's underside, so the first lattice bonds straight onto them. The cap is its own small mesh (&lt;em&gt;HybridCap&lt;/em&gt;), printed as solid skin in Cura's preview; the lattice / rings / solid layers are written into the G-code when you &lt;strong&gt;Save to Disk&lt;/strong&gt; or send to the printer, as for the tethers (untick &lt;em&gt;Cap stack&lt;/em&gt; to keep plain solid layers). Nothing in your Cura profile is changed for it. A sloped or curved underside gets a staircase of flat caps, one per 0.8 mm of height; where a cap can't fit (an overhang less than about 1.4 mm above what's under it, a steep curved underside, or no room clear of the model) there is none, and the tower reaches the underside as before. On the six-part plate Cura's estimate went from 2 h 14 min to 2 h 21 min, about 0.2 m more plastic per part.&lt;/p&gt;

&lt;h3&gt;
  
  
  New in 0.6.2: wide rounded lobes with filleted joins
&lt;/h3&gt;

&lt;p&gt;From the review of the 0.6.1 top view: the thin walls met at the centre in sharp notches (stress raisers), and each was a 2-line slab whose two lines sit 0.4 mm apart, so it bends easily sideways when the nozzle drags across it. The Lobes support built to size is now &lt;strong&gt;one star of a few wide rounded lobes&lt;/strong&gt; (8 mm wide by default, &lt;code&gt;lobe_width&lt;/code&gt;), printed hollow with 2 walls (&lt;em&gt;Tube walls&lt;/em&gt;). The two walls of each lobe are about 8 mm apart, so it is far stiffer sideways, and the lobes blend into each other with rounded inner corners (at least 3 mm, &lt;code&gt;lobe_fillet&lt;/code&gt;) instead of notches. Being wider, fewer are needed: &lt;strong&gt;4 lobes&lt;/strong&gt; on the 40 mm channel and corner, where 0.6.1 had 6 walls. Each lobe still stands on the bed as far out as it must and flares at 45° to the overhang's edges.&lt;/p&gt;

&lt;p&gt;Branch and the Cone's ribs stay thin straight webs (that's what those heads are), but their centre now has a rounded hub, so they no longer meet in a sharp notch either.&lt;/p&gt;

&lt;p&gt;On the six-part plate (Cura 5.13, Neptune 4, measured as in 0.6.1): channel lobes 0.73 m (0.6.1: 0.65 m), corner lobes 0.65 m (0.57 m), so about 0.08 m more plastic each for the stiffer shape. The curved outlines take more, shorter moves (about 15,000 against 3,600 per part). Cura's estimate for the whole plate went from 2 h 6 min to 2 h 14 min, partly from Branch, which now prints its 3-line webs solid (2 walls) instead of as a loop with a gap.&lt;/p&gt;

&lt;h3&gt;
  
  
  New in 0.6.1: hollow walls, and one tower where 0.6.0 made four
&lt;/h3&gt;

&lt;p&gt;From the first look at 0.6.0 in Cura on the Neptune 4:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;The towers print as hollow walls, not zig-zag.&lt;/strong&gt; Cura fills a support mesh with the support pattern, and with the Neptune 4's &lt;em&gt;Support Wall Line Count&lt;/em&gt; of 0 that is one zig-zag line with no outline (Cura won't let a plugin change support walls or pattern for one mesh). 0.6.0 made the walls about 3 mm thick to stop them going missing, and they printed as a zig-zag inside a 3 mm strip. Now every Lobes, Cone and Branch support (and the standing webs) is added as a &lt;strong&gt;normal printed mesh&lt;/strong&gt; with its own settings: one wall round every outline, no infill, no top or bottom skin, printed at your support speed and flow. A 2-line wall prints as one closed loop, the tube as two rings, the lobes as closed loops. Tethers and the Solid/Conform heads stay Cura support meshes (they need the roof and the concentric pattern). In Cura's preview the towers show in the wall colour, not the support colour.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;One tower instead of 4, 8, 16 or 32.&lt;/strong&gt; When the overhang needed one support built to size, 0.6.0 asked for a clear round foot as wide as the reach (it hit the legs of a channel), and any support standing on a surface instead of the bare bed (a base plate under the part, a brim block, a floor of the model) saw that surface as "model above the wall", built nothing and split the overhang again and again. Now each wall of the criss-cross stands on the floor along its own line as far out as it must, and only a small foot round the centre has to be clear. On a channel or corner 8 to 25 mm high, 30 to 60 mm wide and 40 to 80 mm long, standing on a base plate, every head now makes 1 support (0.6.0: 1 to 32).&lt;/li&gt;
&lt;li&gt;On the six-part plate (Cura 5.13, Neptune 4): channel lobes 0.65 m of filament and 3,600 moves (0.6.0: 0.78 m, 8,500), channel branch 0.64 m and 3,000 (0.77 m, 8,500), corner lobes 0.57 m and 2,700 (0.66 m, 7,900); the cones are about the same (1.30 m, 21,700 moves). Measured from the G-code between the legs, so these include a little of the parts' own first layers.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  One criss-cross built to size (Lobes and Branch)
&lt;/h3&gt;

&lt;p&gt;When one normal-sized tower can't reach the whole overhang within the 45° flare (a 40 x 40 mm roof only 15 mm up, say), Lobes and Branch build one support to fit instead of several towers: thin walls (2 lines thick for Lobes, &lt;em&gt;branch lines&lt;/em&gt; for Branch) from one centre point with &lt;strong&gt;no hub&lt;/strong&gt;, each standing on the floor along its own line as far out as it must and then flaring out at 45° to the edge of the overhang. The first go to the corners (a criss-cross); more are added only where a bridge would be longer than &lt;em&gt;Max bridge span&lt;/em&gt;, counting edges that rest on a wall of the model as supported. On the 40 mm test channel that is 6 walls (four to the corners, two to the open ends) and 0.58 m of support, against 1.21 m for four lobes towers. Bigger overhangs still split. (&lt;code&gt;lobes_one_tower&lt;/code&gt; in the settings turns it off.) &lt;strong&gt;Cone&lt;/strong&gt; does the same with one thin cone (a 2-line wall from a circle on the bed, flaring to the overhang's edges) round an X of ribs to the corners: on the channel 0.92 m of support against 0.58 m for the criss-cross, and about 2.5 times the moves, so it may be dropped after the test print.&lt;/p&gt;

&lt;h3&gt;
  
  
  Experimental: Tethers
&lt;/h3&gt;

&lt;p&gt;Instead of a pillar from the floor, &lt;strong&gt;Tethers&lt;/strong&gt; strings strips of support between two walls of the model a short way under the overhang (10 mm by default), and puts a 4-layer cap on top (see below). The idea comes from failed supports: the loose "fuzz" catches on nearby walls and builds up into something solid.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Tethers:&lt;/strong&gt; straight strips from wall to wall. The first and last sit at the two ends of the overhang (0.2 mm in from the edge), the rest evenly between, covering about 20 % of the space. Each prints as a pair of straight lines, so its first layer is a proper bridge. Every other layer is shifted half a line width, so the layers only half overlap and snap apart easily. Cura can't do that shift from the shape (see &lt;a href="https://github.com/The-Dorkknight/cura-hybrid-supports/blob/main/docs/RESEARCH.md" rel="noopener noreferrer"&gt;docs/RESEARCH.md&lt;/a&gt; section 10), so the plugin shifts those lines in the G-code when you &lt;strong&gt;Save to Disk&lt;/strong&gt; or send to the printer. Cura's preview shows the strips without the shift.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Dropped after the first sliced test plate (0.6.0):&lt;/strong&gt; stepped tethers (no advantage over straight ones), the spiderweb (left large regions unsupported) and radial tethers (the fan plus a lobes tower for the far corner was the worst of both; one lobes tower built to size does the job). Their notes stay in docs/RESEARCH.md.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Slow first layers:&lt;/strong&gt; the first three tether layers (0.6.5; was two, in case the first are weak or slack) are treated as bridges in mid-air, and Cura prints support at the profile's support speed (165 mm/s on the Neptune 4). On save, the plugin slows those layers to &lt;em&gt;Bridge speed&lt;/em&gt; (20 mm/s by default; 0 turns this off) with the part fan at 100 %, then puts speed and fan back. This adds about a minute that Cura's time estimate doesn't show.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The cap&lt;/strong&gt; is kept clear of the walls by your normal X/Y distance. Only the strips touch the walls.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The cap stack&lt;/strong&gt; (0.8 mm, 4 layers, written into the G-code when you save, so Cura's preview shows ordinary roof lines instead): a lattice across the tethers (3 mm spacing, first layer at &lt;em&gt;Cap flow&lt;/em&gt;, 85 % by default, so the hot bead fuses with the thin tether instead of squashing it), a lattice along them, one full layer of concentric rings at 50 % fill, then a solid (100 %) diagonal layer, so the part's underside comes off a smooth, flat surface. Untick &lt;em&gt;Cap stack&lt;/em&gt; in the panel to go back to Cura's plain roof.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Test prints first:&lt;/strong&gt; &lt;code&gt;test-prints/&lt;/code&gt; has one Neptune 4 G-code with every head on a 40 mm channel and corner, with what to look for (&lt;a href="https://github.com/The-Dorkknight/cura-hybrid-supports/blob/main/test-prints/README.md" rel="noopener noreferrer"&gt;test-prints/README.md&lt;/a&gt;).&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fcura-hybrid-supports%2Fmain%2Fdocs%2Fimages%2Fsliced-tethers.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fcura-hybrid-supports%2Fmain%2Fdocs%2Fimages%2Fsliced-tethers.webp" alt="Sliced layers with tethers on the test channel: first layer, the next layer shifted, the strips, the roof cap"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;It only works where an overhang has walls on two opposite sides, no more than 110 mm apart (warning over 90 mm, falls back to lobes past 110). Anything else, such as the box's outside ribs, falls back to lobes automatically, and the status line says so.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;It changes some of Cura's own settings,&lt;/strong&gt; because Cura only lets you set them for the whole print, not per support:

&lt;ul&gt;
&lt;li&gt;
&lt;em&gt;Support X/Y Distance&lt;/em&gt; goes to 0. Cura keeps every support that far from the walls, so otherwise the strips stop short. The plugin's other supports and the cap keep their own clearance in their shape.&lt;/li&gt;
&lt;li&gt;
&lt;em&gt;Support Roof&lt;/em&gt; is turned on, 0.8 mm thick (the cap stack's four layers; 2 mm with the stack off).&lt;/li&gt;
&lt;li&gt;
&lt;em&gt;Support Pattern&lt;/em&gt; goes to concentric, so every strip prints as an outline of its own shape, and &lt;em&gt;Support Infill Line Directions&lt;/em&gt; is set along the strips.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Remove&lt;/strong&gt; puts them all back. (Undoing a Remove brings the meshes back but not the settings; press Generate again.)&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;The strips touch the walls, so expect small marks where they break off.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Not printed successfully yet.&lt;/strong&gt; The first test (stepped tethers on the old test box) showed the strips were too sparse and the roof touched the walls; both were fixed in 0.5.0, and the cap stack and slow-bridge step are new in 0.6.0. Whether the first mid-air layer holds at your printer's support speed is still the big unknown.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What changed since 0.1
&lt;/h2&gt;

&lt;p&gt;From the 0.1.x feedback on the EN4 Rugged Box:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;0.1.x problem&lt;/th&gt;
&lt;th&gt;Now&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Tubes printed only in places, in mid-air&lt;/td&gt;
&lt;td&gt;Cura profiles that drop small support areas (Elegoo's: 3 mm²) were deleting thin tube layers. Each support mesh now turns that off for itself only. Every pillar also checks it has a clear column down to a flat landing.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Big gussets stuck into the print&lt;/td&gt;
&lt;td&gt;Every gusset is shortened until it is clear of the model and still stands on something.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;No pillar in the middle of the box&lt;/td&gt;
&lt;td&gt;Pillar positions are searched for a clear column that can see the overhang.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Three pillars across one wide gap&lt;/td&gt;
&lt;td&gt;One pillar per overhang. It only splits when the cone can't reach (default over 250 mm).&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Slat branches were slow&lt;/td&gt;
&lt;td&gt;0.2 made the default head a hollow cone with a cross inside. 0.3 replaces that default with hollow lobes (your idea): cheaper, and every layer is one closed loop. Cone and Branch are still options.&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  How it works (short)
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Select a model and press &lt;strong&gt;H&lt;/strong&gt; (or click the Hybrid Supports tool on the left toolbar).&lt;/li&gt;
&lt;li&gt;Press &lt;strong&gt;Generate&lt;/strong&gt;. For every overhang the plugin:

&lt;ul&gt;
&lt;li&gt;finds one spot where a pillar can stand with a clear column down to the plate or the model,&lt;/li&gt;
&lt;li&gt;builds a hollow tube with gussets at its foot, trimmed to stay clear of the model,&lt;/li&gt;
&lt;li&gt;builds a hollow star of lobes that flare (at most 45°) from the tube out to the corners of the overhang, adding lobes until no bridge is longer than &lt;em&gt;Max bridge span&lt;/em&gt; (edges that rest on a wall of the model count as support),&lt;/li&gt;
&lt;li&gt;gives low overhangs (6 mm or less) thin standing webs instead,&lt;/li&gt;
&lt;li&gt;adds a thin &lt;strong&gt;support blocker&lt;/strong&gt; over the overhang so Cura's own automatic support doesn't fill the same area again.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;Slice as usual. Lobes, Cone and Branch supports are added as &lt;strong&gt;normal printed meshes&lt;/strong&gt; set to print as hollow walls (one wall, no infill, no top or bottom, at your support speed and flow); Tethers and the solid heads are Cura &lt;strong&gt;support meshes&lt;/strong&gt;. Either way they print even when &lt;em&gt;Generate Support&lt;/em&gt; is off.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;The full explanation is in &lt;a href="https://github.com/The-Dorkknight/cura-hybrid-supports/blob/main/HOW_IT_WORKS.md" rel="noopener noreferrer"&gt;HOW_IT_WORKS.md&lt;/a&gt;; the research behind it is in &lt;a href="https://github.com/The-Dorkknight/cura-hybrid-supports/blob/main/docs/RESEARCH.md" rel="noopener noreferrer"&gt;docs/RESEARCH.md&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Use
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Setting&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;th&gt;Default&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Head: Cone / Lobes / Tethers / Branch / Solid / Conform&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Tethers (experimental):&lt;/strong&gt; strips between walls just under the overhang, then a cap (see above). &lt;em&gt;Coverage&lt;/em&gt;, &lt;em&gt;Bridge speed&lt;/em&gt;, &lt;em&gt;Cap stack&lt;/em&gt; and &lt;em&gt;Cap flow&lt;/em&gt; are in the panel. &lt;strong&gt;Lobes:&lt;/strong&gt; hollow star-shaped pseudo-cone with lobes to the corners; one hub-less criss-cross built to size where a normal tower can't reach. &lt;strong&gt;Cone:&lt;/strong&gt; hollow inverted cone. &lt;strong&gt;Branch:&lt;/strong&gt; separate slat fins. &lt;strong&gt;Solid:&lt;/strong&gt; the 0.1 flat-topped head. &lt;strong&gt;Conform:&lt;/strong&gt; solid head whose top follows the underside&lt;/td&gt;
&lt;td&gt;Lobes&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Inside the cone (Cone head only): Cross / Rings / None&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Cross:&lt;/strong&gt; spokes from the pillar to the corners of the overhang, plus extra spokes where bridges would be too long. &lt;strong&gt;Rings:&lt;/strong&gt; concentric bands. &lt;strong&gt;None:&lt;/strong&gt; bare cone&lt;/td&gt;
&lt;td&gt;Cross&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Max bridge span&lt;/td&gt;
&lt;td&gt;Longest bridge you trust your printer with. Lobes adds lobes, Cross adds spokes and Rings adds bands until no bridge is longer (Branch: gap between fins)&lt;/td&gt;
&lt;td&gt;20 mm&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Gussets (0 = off)&lt;/td&gt;
&lt;td&gt;Number of fins at the pillar foot&lt;/td&gt;
&lt;td&gt;4&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Gusset height&lt;/td&gt;
&lt;td&gt;How far up the pillar the fins reach (max half the pillar height)&lt;/td&gt;
&lt;td&gt;6 mm&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Gusset length&lt;/td&gt;
&lt;td&gt;How far the fins reach out. Fins that would hit the model are shortened automatically&lt;/td&gt;
&lt;td&gt;5 mm&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Tube walls (lines)&lt;/td&gt;
&lt;td&gt;Thickness of the hollow tube, in support line widths&lt;/td&gt;
&lt;td&gt;2&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Contact Z gap&lt;/td&gt;
&lt;td&gt;Gap between the head top and the model&lt;/td&gt;
&lt;td&gt;0.2 mm&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Max tower radius&lt;/td&gt;
&lt;td&gt;Upper limit for the tube radius (it scales with how far the head has to reach). The lobe width follows it: 4–10 mm&lt;/td&gt;
&lt;td&gt;8 mm&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Split spans over&lt;/td&gt;
&lt;td&gt;Overhangs longer than this get more than one pillar&lt;/td&gt;
&lt;td&gt;250 mm&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Overhang angle&lt;/td&gt;
&lt;td&gt;Uses Cura's &lt;em&gt;Support Overhang Angle&lt;/em&gt; by default, or your own&lt;/td&gt;
&lt;td&gt;Cura's&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Block Cura auto-support over towers&lt;/td&gt;
&lt;td&gt;Adds the support blockers&lt;/td&gt;
&lt;td&gt;On&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Remove&lt;/strong&gt; deletes the generated meshes for the selected model (or for all models if nothing is selected).&lt;/li&gt;
&lt;li&gt;Pressing &lt;strong&gt;Generate&lt;/strong&gt; again replaces that model's previous supports.&lt;/li&gt;
&lt;li&gt;The generated meshes are children of the model, so they move with it. Re-generate after rotating or scaling.&lt;/li&gt;
&lt;li&gt;The status line tells you how many supports were made, which ones the model only lets the head reach partly, and when gussets were shortened.&lt;/li&gt;
&lt;li&gt;Since 0.6.1 the towers print as walls, so Cura's &lt;em&gt;Support Wall Line Count&lt;/em&gt; and &lt;em&gt;Support Pattern&lt;/em&gt; no longer matter for them (they still do for Tethers and the solid heads). To go back to support meshes, set &lt;code&gt;print_walls = False&lt;/code&gt; in &lt;code&gt;hybrid_core.py&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;Cura's &lt;em&gt;Support X/Y Distance&lt;/em&gt; is respected: the plugin builds its meshes that far (plus 0.1 mm) from the model. (Printed as walls, Cura doesn't add that gap itself, so the clearance is in the shape.)&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fcura-hybrid-supports%2Fmain%2Fdocs%2Fimages%2Frender-blob.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fcura-hybrid-supports%2Fmain%2Fdocs%2Fimages%2Frender-blob.webp" alt="Small curved overhangs on a tilted part get standing webs"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Gusset height trades stiffness for plastic and time. On a 50 mm test tower (0.1 geometry), going from no gussets to 24 mm added about 8 % to the support mesh volume.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fcura-hybrid-supports%2Fmain%2Fdocs%2Fimages%2Fgusset-heights.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fcura-hybrid-supports%2Fmain%2Fdocs%2Fimages%2Fgusset-heights.webp" alt="One tower at gusset heights 0, 4, 12 and 24 mm"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What exists and what doesn't
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Exists (0.6.6-wip)&lt;/th&gt;
&lt;th&gt;Doesn't exist yet&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;One pillar per overhang, with a clear-column landing check&lt;/td&gt;
&lt;td&gt;A test print on a real printer&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Hollow Lobes head (one hub-less criss-cross built to size where needed); experimental Tethers head with a cap stack and slow first bridge layers; hollow Cone head with Cross / Rings / None; Branch, Solid and Conform heads&lt;/td&gt;
&lt;td&gt;Paint-on support (Cura 5.14 adds this natively, see below)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Gussets trimmed to stay clear of the model&lt;/td&gt;
&lt;td&gt;Background thread (generation runs on the UI thread)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Standing webs for low overhangs&lt;/td&gt;
&lt;td&gt;Testing on Windows, macOS or Cura 5.14&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Per-mesh fix for profiles with a minimum support area (Elegoo)&lt;/td&gt;
&lt;td&gt;Multi-extruder support selection beyond reading Cura's support extruder&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Support blockers over each overhang&lt;/td&gt;
&lt;td&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Click-to-run installer (window + text mode), &lt;code&gt;.curapackage&lt;/code&gt; and plugin-zip builds&lt;/td&gt;
&lt;td&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Unit tests (no support inside the model in any head mode) and an automated run inside real Cura (click, generate, slice, per-pillar layer check)&lt;/td&gt;
&lt;td&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;




&lt;p&gt;&lt;strong&gt;Source code:&lt;/strong&gt; &lt;a href="https://github.com/The-Dorkknight/cura-hybrid-supports" rel="noopener noreferrer"&gt;The-Dorkknight/cura-hybrid-supports on GitHub&lt;/a&gt;&lt;/p&gt;

</description>
      <category>github</category>
      <category>opensource</category>
      <category>software</category>
      <category>tools</category>
    </item>
    <item>
      <title>Putting Plastic on a Pedestal: experimental methods of plastic waste reduction for 3D printing</title>
      <dc:creator>Matthew Armstrong</dc:creator>
      <pubDate>Tue, 10 Nov 2026 22:12:33 +0000</pubDate>
      <link>https://dev.to/open-madscience/putting-plastic-on-a-pedestal-experimental-methods-of-plastic-waste-reduction-for-3d-printing-j4g</link>
      <guid>https://dev.to/open-madscience/putting-plastic-on-a-pedestal-experimental-methods-of-plastic-waste-reduction-for-3d-printing-j4g</guid>
      <description>&lt;h3&gt;
  
  
  Experimental methods of plastic waste reduction for 3D printing
&lt;/h3&gt;

&lt;p&gt;&lt;em&gt;Hollow towers, fat lobes, thin caps and tethers strung between walls: a hobbyist's field guide and a defensive publication&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Matthew Armstrong&lt;/strong&gt; (&lt;a href="https://github.com/The-Dorkknight" rel="noopener noreferrer"&gt;The-Dorkknight&lt;/a&gt;) · Version 1.0 · First published 10 November 2026&lt;br&gt;
Text and figures: &lt;a href="https://github.com/The-Dorkknight/putting-plastic-on-a-pedestal/blob/main/LICENSE.md" rel="noopener noreferrer"&gt;CC BY-SA 4.0&lt;/a&gt; · Code: PolyForm Noncommercial 1.0.0 · The software: &lt;a href="https://github.com/The-Dorkknight/cura-hybrid-supports" rel="noopener noreferrer"&gt;cura-hybrid-supports&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;📘 &lt;strong&gt;&lt;a href="https://github.com/The-Dorkknight/putting-plastic-on-a-pedestal/blob/main/docs/putting-plastic-on-a-pedestal.pdf" rel="noopener noreferrer"&gt;The whole paper as one PDF&lt;/a&gt;&lt;/strong&gt;&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Physically untested.&lt;/strong&gt; Every number in this paper comes from geometry code, Cura's slicer, the G-code it writes, or a simulation. None of the supports described here, in the versions described here, has been printed and snapped off yet. I have nearly seventy unfinished projects on the go at once, and a job, a social life and the need to sleep keep interfering with finishing them. The one physical test that did happen (an early tether print in October, an older version) is reported in section 7, mostly because it went wrong in useful ways. Treat every saving below as a prediction, and treat section 11 as an invitation.&lt;/p&gt;
&lt;/blockquote&gt;




&lt;h2&gt;
  
  
  Abstract
&lt;/h2&gt;

&lt;p&gt;Support material is plastic you pay for, wait for and then throw in the bin. On a tall part with a wide roof, Cura's normal support can easily weigh more than the part. This paper describes a Cura plug-in, &lt;em&gt;Hybrid Supports&lt;/em&gt;, that replaces a block of support with a few &lt;strong&gt;pedestals&lt;/strong&gt;: hollow towers with a wide hollow head of rounded lobes, standing on gusseted feet, topped with a thin four-layer interface cap. It also describes &lt;strong&gt;tethers&lt;/strong&gt;, an alternative for roofs that sit between two walls: strips strung from wall to wall a little under the roof, with the same cap on top, and no tower at all.&lt;/p&gt;

&lt;p&gt;On three test parts 200 x 200 mm across, sliced in Cura 5.13 for an Elegoo Neptune 4, the lobed pedestals used &lt;strong&gt;64 % less support plastic than Cura's normal support with a 50 mm stem and 84 % less with a 150 mm stem&lt;/strong&gt;. Against Cura's tree support they used &lt;strong&gt;24 to 26 % less on the tall parts but 29 % more on the short one&lt;/strong&gt;, so the saving is mostly a tall-part effect. A tether span limit (a warning above 90 mm, a hard stop at 110 mm) was set from a physical sag simulation.&lt;/p&gt;

&lt;p&gt;The methods, the dead ends and every variation I could think of are written down here and placed in the public domain as prior art, so they stay free for everyone to use. Nothing has been printed yet (see the box above, and my calendar).&lt;/p&gt;




&lt;h2&gt;
  
  
  Contents
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Why publish this at all&lt;/li&gt;
&lt;li&gt;The problem: we print bins&lt;/li&gt;
&lt;li&gt;A five-minute primer&lt;/li&gt;
&lt;li&gt;The idea on one page&lt;/li&gt;
&lt;li&gt;How a pedestal is built&lt;/li&gt;
&lt;li&gt;Tethers: support with no floor&lt;/li&gt;
&lt;li&gt;Experiments: three big parts&lt;/li&gt;
&lt;li&gt;Dead ends, bugs and one real print&lt;/li&gt;
&lt;li&gt;Prior art&lt;/li&gt;
&lt;li&gt;Defensive disclosure&lt;/li&gt;
&lt;li&gt;Limitations and open questions&lt;/li&gt;
&lt;li&gt;A test plan for anyone with a printer and a free weekend&lt;/li&gt;
&lt;li&gt;How this was made&lt;/li&gt;
&lt;li&gt;Licence&lt;/li&gt;
&lt;li&gt;How to cite&lt;/li&gt;
&lt;li&gt;References&lt;/li&gt;
&lt;/ol&gt;




&lt;h2&gt;
  
  
  0. Why publish this at all
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;To keep it free.&lt;/strong&gt; A defensive publication puts an idea on the public record with a date on it, so nobody can patent it later and charge the rest of us for it. I don't want to own these supports. I want nobody to own them.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;To write down the dead ends.&lt;/strong&gt; Half of this project was trying things that didn't work. A failed idea that's written down saves the next person a weekend. A failed idea that isn't gets reinvented forever.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;To get it printed by someone.&lt;/strong&gt; See the box at the top. If you print any of this, good or bad, I'd love to know (section 11 says what to measure).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Because it's interesting.&lt;/strong&gt; Support is the least loved part of 3D printing, and it hides some lovely geometry.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  1. The problem: we print bins
&lt;/h2&gt;

&lt;p&gt;FDM printers build parts out of thin layers of melted plastic, and each layer needs something under it. Where a part has a roof, a ledge or anything that leans out too far, the slicer adds &lt;strong&gt;support&lt;/strong&gt;: a scaffold that is printed, used once, snapped off and thrown away.&lt;/p&gt;

&lt;p&gt;How much? Take the plainest roof there is: a U shape, 200 mm wide, with legs 10 mm thick and a 5 mm roof whose underside is 150 mm above the bed. Sliced in Cura 5.13 with the Elegoo Neptune 4 profile, the U itself takes about 6.5 hours to print. Cura's normal support under it is &lt;strong&gt;328.7 m of filament, about 980 g&lt;/strong&gt;: most of a 1 kg spool, and roughly another 17 hours of printing, to hold up a roof that weighs a fraction of that. Tree support does much better (68 m, 203 g), and that is the bar this paper has to clear.&lt;/p&gt;

&lt;p&gt;The aim is simple to say: &lt;strong&gt;put less plastic in the bin without letting the part droop.&lt;/strong&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  2. A five-minute primer
&lt;/h2&gt;

&lt;p&gt;If you already know your bridges from your interfaces, skip to section 3.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fch1vd37g0mxestuklm8v.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fch1vd37g0mxestuklm8v.png" alt="Figure 1" width="800" height="248"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Overhang.&lt;/strong&gt; Each layer can lean out a little past the one below, but past about 45 to 50 degrees from vertical there is nothing under the new plastic, and it droops. Cura's &lt;em&gt;Support Overhang Angle&lt;/em&gt; decides where that line is.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Bridge.&lt;/strong&gt; A strand of plastic stretched between two anchors in mid-air. Bridges work surprisingly well when they are short, fast-cooled and taut, and they sag when they are long.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Support.&lt;/strong&gt; The sacrificial scaffold. Cura's "normal" support fills the whole space under an overhang with a zig-zag; tree support grows branches that merge into a few trunks.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Interface.&lt;/strong&gt; The top few layers of a support, printed denser, so the part's first layer has something flat to sit on. It is separated from the part by a tiny air gap (0.2 mm here) so it peels off.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The trade-off is always the same. More support means a nicer underside and less risk, and more support also means more plastic, more time and more picking at the part with pliers afterwards.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fo63603hk8qf4rz2ygbud.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fo63603hk8qf4rz2ygbud.png" alt="Figure 2" width="800" height="296"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  3. The idea on one page
&lt;/h2&gt;

&lt;p&gt;A roof only needs holding up &lt;strong&gt;at its underside&lt;/strong&gt;. Everything between the bed and the underside only has to carry that load down and not wobble while the nozzle drags past it. Solid or zig-zag support spends most of its plastic on that in-between part.&lt;/p&gt;

&lt;p&gt;So:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Hollow towers.&lt;/strong&gt; Carry the load down a tube with two walls and nothing inside. A tube is stiff for its weight (this is why bicycles aren't made of solid rods).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Wide hollow heads.&lt;/strong&gt; Flare out at 45 degrees (the steepest a printer can manage without support of its own) in a few fat rounded lobes, so one tower reaches a lot of roof.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A thin cap.&lt;/strong&gt; Lay a four-layer interface across the tops of the lobes: a lattice to bridge between them, a second lattice across it, 50 % rings and a 100 % top layer. The part sits on that.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Feet.&lt;/strong&gt; Gussets and a foot at the bottom, so a tall thin tower doesn't fall over.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Tethers, where there are walls.&lt;/strong&gt; If a roof sits between two walls of the part, skip the towers altogether: string strips from wall to wall a little under the roof and put the cap on those.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;That's the pedestal. The rest of this paper is the detail, which is where the work was.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fdkxiud34haykdzqrcpui.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fdkxiud34haykdzqrcpui.png" alt="Figure 3" width="800" height="661"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  4. How a pedestal is built
&lt;/h2&gt;

&lt;p&gt;Hybrid Supports is a Cura plug-in (Python, Cura 5.x; tested in Cura 5.13 on Linux). You select a part, choose a head shape and press &lt;em&gt;Generate&lt;/em&gt;. The plug-in adds ordinary meshes to the scene: towers, caps, and a &lt;em&gt;support blocker&lt;/em&gt; over each overhang so that Cura doesn't add its own support on top. It then fixes up a few layers in the G-code when the file is saved.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fbtamqva7mbz2qxlefbqy.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fbtamqva7mbz2qxlefbqy.png" alt="Figure 4" width="800" height="299"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  4.1 Finding what needs holding up
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Overhangs.&lt;/strong&gt; A triangle of the part needs support when it faces down more steeply than the support angle (&lt;code&gt;-n_z &amp;gt; cos(angle)&lt;/code&gt;, the same meaning as Cura's setting; the plug-in uses Cura's own angle by default).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Regions.&lt;/strong&gt; Neighbouring overhang samples are grouped into regions. Each region gets &lt;strong&gt;one support if one can reach it&lt;/strong&gt;. More than one appears only when one can't (section 4.4). Early versions split everything into fixed 30 mm chunks and made a forest of pillars; the user (me) objected.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Landing.&lt;/strong&gt; For every sample, a ray is cast straight down to find what's below: the bed, or another part of the model. A tower may only stand on a clear, nearly flat landing (flatness within 1 mm), and the column above it is checked to be clear of the model all the way up. One early bug had a tower start in mid-air because its landing ray hit a rib.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  4.2 Hollow towers that really are hollow
&lt;/h3&gt;

&lt;p&gt;Cura is strict about support meshes. It fills them with the support pattern (zig-zag on most profiles), and the settings that would make them print as outlines (support wall count, pattern and density) &lt;strong&gt;can't be set per mesh&lt;/strong&gt; in Cura 5.13. So the first versions printed what looked like towers in the preview as zig-zag scribble inside a thin strip: no proper walls at all.&lt;/p&gt;

&lt;p&gt;The fix (0.6.1) was to stop calling them support. The towers are added as &lt;strong&gt;normal meshes&lt;/strong&gt; with their own per-mesh settings: walls only, no infill, no top or bottom skin, the support line width, the support speed and flow. Cura prints them as closed wall loops, which is exactly what a hollow tube needs. The trunk is two lines thick. Its radius is a quarter of the distance it has to reach, kept between 2 and 8 mm. Four gussets (6 mm high, 5 mm long) and a foot keep it upright.&lt;/p&gt;

&lt;h3&gt;
  
  
  4.3 The heads: lobes, cone and branch
&lt;/h3&gt;

&lt;p&gt;The &lt;strong&gt;head&lt;/strong&gt; is where a narrow trunk becomes a wide support.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Lobes (the default).&lt;/strong&gt; A star of hollow capsules, each &lt;strong&gt;8 mm wide&lt;/strong&gt;, printed with 2 walls. Every lobe flares out at 45 degrees towards a corner or an edge of the roof, and new lobes are added only where a bridge between two of them would be longer than the maximum bridge span (20 mm by default). Edges that rest on a wall of the model count as supported already.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cone.&lt;/strong&gt; A thin hollow cone (a 2-line wall, 64 sides) flaring at 45 degrees from a circle on the bed, with an X of ribs inside.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Branch.&lt;/strong&gt; Straight 3-line webs from a rounded hub, like the spokes of a wheel.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Why are the lobes so fat? Version 0.6.1 used six thin webs meeting at a point, and in review I pointed out two things: the sharp inner corners are stress concentrators, and a 0.8 mm web is floppy sideways when the nozzle drags past it. Bending stiffness across a web grows with the cube of its thickness. A hollow lobe whose two walls sit about 8 mm apart behaves like a box section, roughly (8 / 0.8)² = 100 times stiffer across than a 0.8 mm slab for the same wall material per unit length (ignoring the ends). So 0.6.2 went to fewer, wider lobes.&lt;/p&gt;

&lt;p&gt;The joins are rounded with a &lt;strong&gt;smooth maximum&lt;/strong&gt;. Each lobe is a capsule, which in polar form around the trunk is a radius for every direction. The star's outline in each direction is a polynomial smooth maximum of the lobes' radii, with a 3 mm width. That rounds the inner corners without moving the lobes' straight sides. The outline is sampled in 128 directions, about 1 mm per step at the tips, which keeps the number of nozzle moves down.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fnjqhi9okp4yxeohvf9g7.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fnjqhi9okp4yxeohvf9g7.png" alt="Figure 5" width="800" height="275"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  4.4 Built to size, and splitting when it can't reach
&lt;/h3&gt;

&lt;p&gt;A 45 degree flare can only reach out as far as it is tall. When a normal tower can't reach the whole region, the plug-in first tries a &lt;strong&gt;support built to size&lt;/strong&gt;: one star (or cone) as wide as the region needs, each arm with its own foot along its line, as far as the floor beside it is clear and level. If that support reaches at least &lt;strong&gt;85 %&lt;/strong&gt; of the region, it is kept. If not, the region is split in two and each half tries again (up to five times).&lt;/p&gt;

&lt;p&gt;On the 200 mm parts this produced &lt;strong&gt;two to four towers&lt;/strong&gt;, reaching 91 to 98 % of the roof. The last few percent are corner tips, left to the cap alone.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fcx3z5rmesxqkyarg6tpo.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fcx3z5rmesxqkyarg6tpo.png" alt="Figure 6" width="800" height="385"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Figure 6 is the clearest picture of the whole idea. Each panel is a real layer of real G-code, seen from above. At 60 mm up, Cura's normal support is a solid hatch across the full 200 x 200 mm. Tree support is about forty separate round branches. The pedestal is four small rings.&lt;/p&gt;

&lt;h3&gt;
  
  
  4.5 The interface cap
&lt;/h3&gt;

&lt;p&gt;Since the towers print as walls, Cura's support interface no longer applies to them, and without an interface a part's first layer would bridge everything between the lobe walls and droop. So every pedestal (and every tether set) gets a &lt;strong&gt;cap&lt;/strong&gt;: a separate normal mesh of four layers, 0.8 mm in all, sitting directly on the tower tops (two normal meshes stack with no gap), with the usual 0.2 mm air gap above it to the part.&lt;/p&gt;

&lt;p&gt;Cura can only print a cap one way (as skin), so when the file is saved the plug-in &lt;strong&gt;rewrites the cap's layers in the G-code&lt;/strong&gt;, finding them by Cura's &lt;code&gt;;MESH:&lt;/code&gt; comments:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Layer&lt;/th&gt;
&lt;th&gt;What it is&lt;/th&gt;
&lt;th&gt;Why&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;1&lt;/td&gt;
&lt;td&gt;Lattice, 3 mm pitch, slowed to 20 mm/s&lt;/td&gt;
&lt;td&gt;bridges across the lobe tops; slow and cool so it doesn't sag&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;2&lt;/td&gt;
&lt;td&gt;Lattice, crossing the first&lt;/td&gt;
&lt;td&gt;turns lines into a grid&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;3&lt;/td&gt;
&lt;td&gt;Concentric rings at 50 %&lt;/td&gt;
&lt;td&gt;a near-solid layer that's cheap and quick&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;4&lt;/td&gt;
&lt;td&gt;100 % solid top&lt;/td&gt;
&lt;td&gt;the flat surface the part actually sits on&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;On a flat roof the cap is one big sheet. On a slope it becomes a staircase of convex pieces, one per 0.8 mm height band. Round a dome it splits into 8 sectors. Each piece is cut back from the model with half-planes, across the direction of the nearest model surface.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F5az3judqk1guw3fl0hn7.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F5az3judqk1guw3fl0hn7.png" alt="Figure 7" width="799" height="187"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;A side effect of Figure 7: look at the last tower layer. At the top, the lobes of four towers meet and make a grid. That wasn't designed, it fell out of the rules.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7gw4tpskgwrwzs09h517.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7gw4tpskgwrwzs09h517.png" alt="Figure 8" width="799" height="226"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  5. Tethers: support with no floor
&lt;/h2&gt;

&lt;p&gt;Tethers came from something I noticed on failed prints. When a support collapses, the loose "fuzz" often catches on a nearby wall, and within about 10 mm it has built itself into something solid. So why not do that on purpose?&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;What it does.&lt;/strong&gt; For each overhang, the plug-in tries 12 directions and looks for two walls of the model facing each other. It keeps the direction with the shortest span where most strips find a wall at both ends. Then:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Strips are &lt;strong&gt;3 lines wide&lt;/strong&gt; (about 1.26 mm), spaced so they cover &lt;strong&gt;20 %&lt;/strong&gt; of the roof, evenly, with the first and last strip 0.2 mm in from the two ends of the overhang.&lt;/li&gt;
&lt;li&gt;They start &lt;strong&gt;10 mm below&lt;/strong&gt; the roof and go straight up to the cap.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Every other layer is shifted half a line width&lt;/strong&gt; (a G-code post-process, because Cura flattens any per-layer zig-zag built into the shape). Each line then sits half on the one below, so the strip is a stack of half-overlapping lines that snaps apart easily.&lt;/li&gt;
&lt;li&gt;The &lt;strong&gt;first three layers&lt;/strong&gt; are bridges in mid-air, so they are slowed to 20 mm/s with the part fan at 100 %. Cura doesn't apply its bridge settings to support, and the Neptune 4 profile otherwise prints support at 165 mm/s.&lt;/li&gt;
&lt;li&gt;The &lt;strong&gt;cap&lt;/strong&gt; is the same four-layer stack, with the first layer under-extruded to 85 % so a hot bead fuses with a thin strip instead of melting through it.&lt;/li&gt;
&lt;li&gt;Tethers refuse an overhang if more than 10 % of it would be farther than 1.5 strip pitches (at least 8 mm) from any strip (an L-shaped corner, for example). Lobes take over.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ftf6qreb73vej5j9j4sd0.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ftf6qreb73vej5j9j4sd0.png" alt="Figure 9" width="799" height="156"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  5.1 How far can a strip go?
&lt;/h3&gt;

&lt;p&gt;The obvious question, and the first version's answer (a 120 mm limit) turned out to have no source at all: it was a search distance in the first tether code, written when the test parts were 40 mm wide. When I asked where it came from, nobody (human or software) could say. So I went looking:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Published bridging guidance is all over the place: &lt;strong&gt;10 to 30 mm&lt;/strong&gt; in conservative guides, &lt;strong&gt;50 to 80 mm&lt;/strong&gt; "well tuned", and &lt;strong&gt;about 100 mm&lt;/strong&gt; in hobbyist reports with very good cooling (one forum post measured 96 mm with under 0.3 mm of sag). I found nothing at all for 180 mm.&lt;/li&gt;
&lt;li&gt;An AI chatbot confidently told me the Neptune 4 could bridge 100 to 150 mm and cited a review. The review gives no bridging distance whatsoever. Lesson noted.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;So I wrote my own &lt;strong&gt;sag simulation&lt;/strong&gt; (included in the plug-in's repository). A PLA strand leaves the nozzle, is drawn taut by the moving head, hangs like a string while it's hot, and freezes into shape as it cools past its glass transition, element by element, while the fan blows on it. One factor was fitted to published Ender 3 guidance. Everything else, including every Neptune 4 number, comes from the physics. Its findings:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Up to about &lt;strong&gt;60 mm&lt;/strong&gt;, both printers stay under 0.4 mm of sag at any sensible speed.&lt;/li&gt;
&lt;li&gt;Past about &lt;strong&gt;100 mm&lt;/strong&gt;, cooling stops being the limit. The frozen 0.32 mm strand simply bends under its own weight (sag grows with span to the fourth power), which is a ceiling for a 0.4 mm nozzle and 0.2 mm layers on any printer.&lt;/li&gt;
&lt;li&gt;Conservatively the Neptune 4 manages &lt;strong&gt;70 to 90 mm&lt;/strong&gt;. With rigid walls, where the strand's shrinkage pulls it taut after it lands, that rises to &lt;strong&gt;85 to 130 mm&lt;/strong&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;From that, the rule: &lt;strong&gt;up to 90 mm, tethers build. From 90 to 110 mm they build, with a warning. Over 110 mm, the plug-in refuses, says why ("the walls are 180 mm apart, over the 110 mm limit"), and builds lobes instead.&lt;/strong&gt; The warning, for the record, reads: &lt;em&gt;"the vengeful Flying Spaghetti Monster may appear suddenly for trying to violate the laws of physics."&lt;/em&gt; I consider this an accurate engineering statement.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fre2to2wni8kowvcedbq8.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fre2to2wni8kowvcedbq8.png" alt="Figure 10" width="800" height="281"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F299txrcknqjd22fppila.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F299txrcknqjd22fppila.png" alt="Figure 11" width="800" height="474"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  6. Experiments: three big parts
&lt;/h2&gt;

&lt;p&gt;Small test parts flatter everyone, because the fixed costs dominate. So I asked for big ones, with long stems.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Parts.&lt;/strong&gt; A U 200 x 200 x 200 mm (10 mm legs, 5 mm roof) with the roof's underside &lt;strong&gt;50 mm&lt;/strong&gt; and &lt;strong&gt;150 mm&lt;/strong&gt; above the bed, and a 200 x 200 mm &lt;strong&gt;table top on a single 30 x 30 mm stem&lt;/strong&gt;, 150 mm up.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Method.&lt;/strong&gt; Each part was sliced headless in Cura 5.13 with the stock Elegoo Neptune 4 profile (PLA, 0.4 mm nozzle, 0.2 mm layers): bare, with Cura's normal support, with Cura's tree support, and with the plug-in's lobes, cone and branch heads (Generate, then save, so the caps are written in). Support plastic is the job minus the bare part, counted move by move from the G-code. Time comes from a plain distance-over-feed-rate model of every move, because Cura's own estimate can't see the slowed cap and tether layers written after slicing, and so flatters the plug-in. Grams assume PLA at 1.24 g/cm³.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Part (bare print)&lt;/th&gt;
&lt;th&gt;Support&lt;/th&gt;
&lt;th&gt;Support plastic&lt;/th&gt;
&lt;th&gt;Support time&lt;/th&gt;
&lt;th&gt;Whole job&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;U, stem 50 mm (230 min)&lt;/td&gt;
&lt;td&gt;Cura normal&lt;/td&gt;
&lt;td&gt;109.8 m · 328 g&lt;/td&gt;
&lt;td&gt;342 min&lt;/td&gt;
&lt;td&gt;574 min&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;/td&gt;
&lt;td&gt;Cura tree&lt;/td&gt;
&lt;td&gt;30.5 m · 91 g&lt;/td&gt;
&lt;td&gt;89 min (+45 travel)&lt;/td&gt;
&lt;td&gt;359 min&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Hybrid lobes&lt;/strong&gt; (4 towers)&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;39.4 m · 118 g&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;137 min&lt;/td&gt;
&lt;td&gt;374 min&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;/td&gt;
&lt;td&gt;Hybrid cone (2 towers)&lt;/td&gt;
&lt;td&gt;27.0 m · 80 g&lt;/td&gt;
&lt;td&gt;99 min&lt;/td&gt;
&lt;td&gt;343 min&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;/td&gt;
&lt;td&gt;Hybrid branch (8 towers)&lt;/td&gt;
&lt;td&gt;83.2 m · 248 g&lt;/td&gt;
&lt;td&gt;270 min&lt;/td&gt;
&lt;td&gt;573 min&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;U, stem 150 mm (393 min)&lt;/td&gt;
&lt;td&gt;Cura normal&lt;/td&gt;
&lt;td&gt;328.7 m · 980 g&lt;/td&gt;
&lt;td&gt;1009 min&lt;/td&gt;
&lt;td&gt;1404 min&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;/td&gt;
&lt;td&gt;Cura tree&lt;/td&gt;
&lt;td&gt;67.9 m · 203 g&lt;/td&gt;
&lt;td&gt;202 min (+99 travel)&lt;/td&gt;
&lt;td&gt;666 min&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Hybrid lobes&lt;/strong&gt; (4 towers)&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;51.5 m · 154 g&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;173 min&lt;/td&gt;
&lt;td&gt;583 min&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;/td&gt;
&lt;td&gt;Hybrid cone (2 towers)&lt;/td&gt;
&lt;td&gt;87.1 m · 260 g&lt;/td&gt;
&lt;td&gt;282 min&lt;/td&gt;
&lt;td&gt;720 min&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;/td&gt;
&lt;td&gt;Hybrid branch (4 towers)&lt;/td&gt;
&lt;td&gt;98.6 m · 294 g&lt;/td&gt;
&lt;td&gt;317 min&lt;/td&gt;
&lt;td&gt;786 min&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Table top, stem 150 mm (206 min)&lt;/td&gt;
&lt;td&gt;Cura normal&lt;/td&gt;
&lt;td&gt;363.5 m · 1084 g&lt;/td&gt;
&lt;td&gt;1115 min&lt;/td&gt;
&lt;td&gt;1313 min&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;/td&gt;
&lt;td&gt;Cura tree&lt;/td&gt;
&lt;td&gt;76.7 m · 229 g&lt;/td&gt;
&lt;td&gt;228 min (+77 travel)&lt;/td&gt;
&lt;td&gt;496 min&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;Hybrid lobes&lt;/strong&gt; (4 towers)&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;56.8 m · 169 g&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;192 min&lt;/td&gt;
&lt;td&gt;400 min&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;/td&gt;
&lt;td&gt;Hybrid cone (2 towers)&lt;/td&gt;
&lt;td&gt;99.3 m · 296 g&lt;/td&gt;
&lt;td&gt;321 min&lt;/td&gt;
&lt;td&gt;570 min&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;/td&gt;
&lt;td&gt;Hybrid branch (4 towers)&lt;/td&gt;
&lt;td&gt;107.4 m · 320 g&lt;/td&gt;
&lt;td&gt;346 min&lt;/td&gt;
&lt;td&gt;617 min&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fd3x4fctg3ufrw547cxii.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fd3x4fctg3ufrw547cxii.png" alt="Figure 12" width="800" height="480"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fp3sq0mv0ehbqu4x2648r.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fp3sq0mv0ehbqu4x2648r.png" alt="Figure 13" width="800" height="477"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  6.1 What it says
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Against Cura's normal support,&lt;/strong&gt; lobes use 64 % less support plastic on the short U and &lt;strong&gt;84 %&lt;/strong&gt; less on both tall parts. The whole job is 35 %, 58 % and 70 % shorter.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Against tree support,&lt;/strong&gt; lobes use 24 % less on the tall U and 26 % less on the table top, and the whole job is 12 % and 19 % faster. On the short U they lose: 29 % &lt;strong&gt;more&lt;/strong&gt; plastic. At 50 mm the cap (a full 0.8 mm sheet under 35,000 mm² of roof, 5.2 m) is a seventh of the support, and a tree is already cheap.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Height is the whole story&lt;/strong&gt; (Figure 14). Tripling the stem from 50 to 150 mm triples Cura's normal support, more than doubles the tree, and adds only about 30 % to the pedestals. A hollow tube costs little per extra millimetre of height. A tree pays for every branch all the way down.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The cone&lt;/strong&gt; is the cheapest of everything on the short U (27.0 m, two towers), but its flared shell gets expensive when tall. It only wins on short, wide parts.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Branch&lt;/strong&gt; lost everywhere. Lots of thin webs is a lot of nozzle travel.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Tethers&lt;/strong&gt; refused all three parts, correctly (Figure 10): the U's legs are 180 mm apart and the table top has no walls. They fell back to lobes, so their hatched bars in Figures 12 and 13 are the lobes numbers. Where they &lt;em&gt;can&lt;/em&gt; work, they win (Figure 15). On an earlier benchmark with a 100 x 100 mm channel roof between two walls (sliced before the cap stack existed, with Cura's own time estimate), tethers used 6.8 m of support at 100 mm up and 6.9 m at 150 mm, against 7.2 and 7.6 m for lobes, 23.1 and 34.7 m for Cura's sparsest support (5 % zig-zag, no roof) and 28.0 and 41.3 m for tree support. That is about 5 to 10 % less than lobes and about 75 to 85 % less than tree support. Note the 100 mm span: under the new rules that tether would now build with a Flying Spaghetti Monster warning.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fpejs5i95ycx0eatja2zs.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fpejs5i95ycx0eatja2zs.png" alt="Figure 14" width="800" height="444"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F1ao17iytek31twul9y51.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F1ao17iytek31twul9y51.png" alt="Figure 15" width="800" height="330"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  6.2 What it doesn't say
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Whether any of it holds.&lt;/strong&gt; Four 8 mm lobes standing 150 mm tall may wobble. A 3 mm lattice bridging 15 to 20 mm between lobes may sag. If a pedestal falls over, the saving is zero and you've lost the part as well.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Time to the second.&lt;/strong&gt; The move model ignores acceleration, so real times will be longer for everyone, and Cura's own estimate tells a kinder story for the plug-in (lobes support 203, 291 and 310 min against tree 300, 504 and 555 min). I've used the stricter numbers.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Removal effort, surface finish and success rate.&lt;/strong&gt; All unmeasured.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Other printers, materials or tuned tree settings.&lt;/strong&gt; One printer profile, PLA only, stock tree settings.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  7. Dead ends, bugs and one real print
&lt;/h2&gt;

&lt;p&gt;In roughly the order I tripped over them:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;What happened&lt;/th&gt;
&lt;th&gt;Why&lt;/th&gt;
&lt;th&gt;What changed&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Towers printed only in places, some starting in mid-air&lt;/td&gt;
&lt;td&gt;Elegoo's profile deletes support areas under 3 mm², which includes a thin tube's whole layer; one landing ray hit a rib&lt;/td&gt;
&lt;td&gt;per-mesh minimum support area 0; every tower checks a clear column to a flat landing&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;A wide gap got three pillars&lt;/td&gt;
&lt;td&gt;regions were split at a fixed 30 mm&lt;/td&gt;
&lt;td&gt;one support per region unless it can't reach&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Single-line X ribs (the cheapest possible head)&lt;/td&gt;
&lt;td&gt;Cura shrinks support areas by half a line before printing, so a one-line rib vanishes entirely&lt;/td&gt;
&lt;td&gt;dropped; the narrowest clean rib is a two-line loop, which is exactly what a lobe wall already is&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;First real print&lt;/strong&gt; (October, an early tether test box)&lt;/td&gt;
&lt;td&gt;the tethers were too sparse, so much of the cap had nothing under it, and the roof lattice touched the walls&lt;/td&gt;
&lt;td&gt;tethers cover 20 %, evenly spaced; the cap is kept clear of the walls&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Tether lines came out as zig-zag springs&lt;/td&gt;
&lt;td&gt;the profile's 65 degree support line direction crossed the strips&lt;/td&gt;
&lt;td&gt;lines run along the strips; a straight line pulled taut bridges far better than a chain of diagonals&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Per-layer stagger built into the shape vanished&lt;/td&gt;
&lt;td&gt;Cura slices each layer at top and bottom and keeps what they share&lt;/td&gt;
&lt;td&gt;the stagger is written into the G-code after slicing&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Spiderweb supports, radial tether fans, stepped tethers&lt;/td&gt;
&lt;td&gt;in review, regions of roof with nothing under them, or no gain over straight tethers&lt;/td&gt;
&lt;td&gt;dropped (the spiderweb is in the research notes for anyone braver)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;"Zig-zag towers, and 4 towers when 1 would do" (my review of 0.6.0)&lt;/td&gt;
&lt;td&gt;Cura fills support meshes with zig-zag; a ray cast from inside the base plate made every wall think the model was right above it, so it split, and split again&lt;/td&gt;
&lt;td&gt;towers as walls-only normal meshes; rays cast from above the foot; per-arm feet&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Sharp notches, floppy webs (review of 0.6.1)&lt;/td&gt;
&lt;td&gt;thin webs meeting at a point&lt;/td&gt;
&lt;td&gt;fat rounded lobes with smooth joins (Figure 5)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;"They don't have any support interface, the part will droop" (review of 0.6.2)&lt;/td&gt;
&lt;td&gt;walls-only towers get no Cura interface&lt;/td&gt;
&lt;td&gt;the four-layer cap on every support&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Cone slices on the tall parts cut off after 4 layers&lt;/td&gt;
&lt;td&gt;the cone's reach was taken from the farthest point in a 5.6 degree sector, which on a 100 mm overhang lies well off the ray, so the shell flared 17 mm past the roof and off the bed&lt;/td&gt;
&lt;td&gt;reach measured along the ray&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;A 120 mm tether limit with no source&lt;/td&gt;
&lt;td&gt;a search distance that became a rule&lt;/td&gt;
&lt;td&gt;research plus the sag simulation: 90 mm warning, 110 mm hard limit&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  8. Prior art
&lt;/h2&gt;

&lt;p&gt;An informal search (web, papers, Google Patents, Cura plug-ins), not a professional one. If you know of anything closer, please tell me and I'll add it.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Work&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;th&gt;How it relates&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Dumas, Hergel, Lefebvre, &lt;em&gt;Bridging the Gap: Automated Steady Scaffoldings for 3D Printing&lt;/em&gt; (SIGGRAPH 2014)&lt;/td&gt;
&lt;td&gt;scaffolds of horizontal bridges held up by vertical pillars, optimised for low material and stability&lt;/td&gt;
&lt;td&gt;the closest academic idea to tethers (bridges between anchors) and to sparse pillars; there the bridges hang between printed pillars, not the part's own walls&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Vaněk, Galicia, Beneš, &lt;em&gt;Clever Support&lt;/em&gt; (Computer Graphics Forum, 2014)&lt;/td&gt;
&lt;td&gt;tree-like support grown from the contact points down&lt;/td&gt;
&lt;td&gt;the ancestor of today's tree supports&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Freire, Hornus, Perchy, Lefebvre, procedural bridges-and-pillars support (Eurographics short paper, 2022)&lt;/td&gt;
&lt;td&gt;pillars joined by bridges; on their test model the support length didn't drop, the gain was fewer contact points&lt;/td&gt;
&lt;td&gt;supports the observation that the win is in the shape, not just the count&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Autodesk, US 10,775,770 B2 (priority 2017, filed 2018)&lt;/td&gt;
&lt;td&gt;thin single-bead support walls offset from the part perimeter, support where bridged infill is too long, small Z gaps to ease removal&lt;/td&gt;
&lt;td&gt;thin-walled supports in general; I couldn't read its claims in full and draw no conclusion&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;US 2021/0283856 A1, "Automatic reusable support for 3D printing"&lt;/td&gt;
&lt;td&gt;not read (the page wouldn't load)&lt;/td&gt;
&lt;td&gt;listed for completeness&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Cura tree support, PrusaSlicer organic supports&lt;/td&gt;
&lt;td&gt;branching supports from the bed&lt;/td&gt;
&lt;td&gt;the main comparison in section 6&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Cura &lt;em&gt;lightning infill&lt;/em&gt;
&lt;/td&gt;
&lt;td&gt;branches grown from inside walls only where top surfaces need them&lt;/td&gt;
&lt;td&gt;the same instinct as tethers (anchor on the part, grow only where needed), but only inside a part&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Arc overhangs (Steven McCulloch, 2022) and wave overhangs (2026)&lt;/td&gt;
&lt;td&gt;print the roof's first layer as arcs or ripples growing from a supported edge, no support at all&lt;/td&gt;
&lt;td&gt;a different route to the same goal; slow (about 2 mm/s)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;em&gt;CustomSupportCylinder&lt;/em&gt; (Cura plug-in by 5@xes)&lt;/td&gt;
&lt;td&gt;user-placed cylinders, tubes and other shapes as support&lt;/td&gt;
&lt;td&gt;hand-placed towers; no automatic sizing, heads or caps&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Armstrong, &lt;em&gt;Fin-ally Supported&lt;/em&gt; (2026)&lt;/td&gt;
&lt;td&gt;my own shark-fin supports for long tilted faces&lt;/td&gt;
&lt;td&gt;a sister method, already published&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;As far as I could find, nothing describes the combination here: &lt;strong&gt;hollow towers printed as walls, sized from the overhang, with wide filleted lobe heads and per-arm feet, a four-layer interface cap rewritten into the G-code, and wall-to-wall tethers with staggered layers, slowed bridge layers and a simulated span limit.&lt;/strong&gt; Each piece on its own has relatives. I'm claiming nothing, and only writing it down.&lt;/p&gt;

&lt;h2&gt;
  
  
  9. Defensive disclosure
&lt;/h2&gt;

&lt;p&gt;The following methods, and the obvious variations of each, are hereby published and dedicated to the public. They are described well enough for a person skilled in FDM slicing to build them, and the working code is published alongside (section 13).&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;9.1 Pedestal supports&lt;/strong&gt;&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Generating support for an overhang region as one or more hollow towers printed as closed wall loops (walls only, no infill, no top or bottom skin), added to the slicer as normal model meshes with per-mesh settings rather than as support meshes, so that the slicer's global support pattern does not apply to them.&lt;/li&gt;
&lt;li&gt;Choosing one tower per overhang region and adding more only when one cannot reach it: first trying a single support "built to size" (a hub-less star or cone wide enough for the region, with each arm on its own foot along its line), keeping it when it reaches at least a set share of the region (85 % here, any value 50 to 100 %), and otherwise splitting the region and repeating, to a set depth.&lt;/li&gt;
&lt;li&gt;Heads made of hollow capsule-shaped lobes flaring at the printer's overhang limit (45 degrees here, any angle 30 to 60), each lobe much wider than its wall (8 mm here; any width 3 to 20 mm; 2 walls; 1 to 4 walls), with the number and direction of lobes chosen so that no bridge between lobes, or between a lobe and a supported model edge, exceeds a maximum bridge span.&lt;/li&gt;
&lt;li&gt;Joining lobes with a smooth maximum of their polar outline radii (polynomial, exponential or any other smooth-max or fillet), so inner corners are rounded by a set radius without moving the lobes' sides; sampling the outline in a set number of directions to bound the number of print moves.&lt;/li&gt;
&lt;li&gt;Hollow cone heads built to size, with internal ribs to the region's corners, with the cone's reach in each direction measured along that direction from overhang samples near it, not as the farthest point in an angular sector.&lt;/li&gt;
&lt;li&gt;Gussets and feet: triangular gussets round the trunk's base (any count, height and length), feet along each arm's line as far as the floor is clear and level, or a round foot as wide as is clear.&lt;/li&gt;
&lt;li&gt;A support blocker (anti-overhang volume) over each generated support's overhang, so the slicer adds no support of its own there.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;&lt;strong&gt;9.2 Interface caps&lt;/strong&gt;&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Placing on top of every generated support a separate model mesh of a few layers (0.8 mm here; any thickness 0.4 to 2 mm), stacked directly on the support with no Z gap and kept a contact gap (0.2 mm here) below the part, and rewriting that mesh's layers in the G-code after slicing, identified by the slicer's mesh-name comments, as an interface stack: one or more sparse lattice layers (any pitch; any angle), optionally slowed and/or under-extruded on the first; then a partly filled layer (rings at 50 % here, any pattern and density); then a fully solid top layer.&lt;/li&gt;
&lt;li&gt;Building caps on sloped or curved undersides as a staircase of convex pieces, one per height band, splitting ring-shaped bands into sectors, and cutting each piece back from the model with half-planes perpendicular to the direction of the nearest model surface.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;&lt;strong&gt;9.3 Tethers&lt;/strong&gt;&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Supporting an overhang that lies between two facing walls of the part itself with strips strung from wall to wall below it, starting a set distance below the overhang (10 mm here, any 2 to 30 mm), covering a set share of the overhang (20 % here, any 5 to 60 %), evenly spaced with the first and last strip at the overhang's ends, choosing among several directions the one with the shortest span for which most strips find walls at both ends; refusing when a set share of the overhang would be beyond a set distance from every strip.&lt;/li&gt;
&lt;li&gt;Shifting the strips' lines by a fraction of a line width (half here) on alternate layers by post-processing the G-code, so each layer rests partly on the one below and the strip separates easily; or shifting by any pattern of offsets per layer.&lt;/li&gt;
&lt;li&gt;Printing the first one or more strip layers (three here) as bridges: slowed, with increased part cooling, optionally changed flow; and printing the first cap layer over the strips under-extruded (85 % here).&lt;/li&gt;
&lt;li&gt;Limiting the wall-to-wall span with a warning threshold and a hard limit (90 and 110 mm here, any values), derived from a physical model of strand sag (drawing tension, hot-zone droop frozen in at the glass transition, tensioned-beam sag of the frozen strand, forced-convection cooling by the part fan), optionally per printer, speed, nozzle, layer height and material; and, past the hard limit, falling back to another support type automatically and telling the user why, with the measured span.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;&lt;strong&gt;9.4 Variations&lt;/strong&gt; that are also disclosed: any of the above in any slicer (Cura, PrusaSlicer, OrcaSlicer, Bambu Studio, Simplify3D or others) or in the printer firmware or host; written as slicer plug-ins, post-processing scripts, or native slicer features; in any FDM/FFF material; with tethers of other cross-sections (single lines, wider strips, I-beams) and towers of other cross-sections (square, polygonal, elliptical, multi-tube); with lobes of other shapes (teardrops, ellipses); with the lattice, ring and solid layers in other orders or counts; combined with arc or wave overhangs grown from the lobe tops or the strips; and with any of the numeric values above changed.&lt;/p&gt;

&lt;h2&gt;
  
  
  10. Limitations and open questions
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Nothing in this version has been printed.&lt;/strong&gt; I can't say this often enough.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Stability of tall thin towers&lt;/strong&gt; under nozzle drag, especially 150 mm tall on two-line walls.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cap sag&lt;/strong&gt; between lobes, and whether 85 % flow on the first cap layer helps or leaves gaps.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Tether sag&lt;/strong&gt; past 60 mm, and whether the "layers above catch up within about 10 mm" rule from my failed prints holds.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Removal:&lt;/strong&gt; whether pedestals come off in one piece, and what the 0.2 mm gap leaves on the underside.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Wall marks:&lt;/strong&gt; tethers touch the part's walls and will leave marks there.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Shared settings:&lt;/strong&gt; some Cura settings are per job, not per part, so tethers change the support X/Y distance, pattern and roof for the whole job, which affects other supports on the same plate.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;One printer profile, one material, one slicer.&lt;/strong&gt; The sim's speed and fan advice suggests the right numbers differ a lot between printers.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Prior-art search&lt;/strong&gt; was informal. Not legal advice.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  11. A test plan for anyone with a printer and a free weekend
&lt;/h2&gt;

&lt;p&gt;If you print any of these, please open an issue on the plug-in's repository with photos and, ideally, these numbers:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;The test plate&lt;/strong&gt; (&lt;code&gt;test-prints/&lt;/code&gt; in the plug-in repository): six small parts, one per support style, one job, about 2 h 20 min on a Neptune 4. Did each roof stay flat? Did the supports come off in one piece? Weigh the removed support.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A tether span ladder:&lt;/strong&gt; single strips across gaps of 20 to 120 mm in 10 mm steps. Measure mid-span sag with calipers, on the first layer and after five layers. This calibrates the sim.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The 200 mm U at 50 mm tall,&lt;/strong&gt; lobes against tree support. It's the cheapest of the big parts (about 6 hours) and the one where the plug-in currently loses on paper, so it's the fairest fight.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A tall tower wobble test:&lt;/strong&gt; one lobes support 150 mm tall. Watch whether it sways as the head prints.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  12. How this was made
&lt;/h2&gt;

&lt;p&gt;I designed this and made the decisions: the hollow tower, the tethers, the lobes getting fat, the cap recipe, the span limits, which ideas lived and which were dropped. An AI assistant (Anthropic's Claude) did most of the typing: the code, the geometry, the test harness that runs real Cura headless, the measurements and the first drafts of the writing, and it made plenty of mistakes along the way, several of which are in section 7. I reviewed each version myself, in Cura and in the screenshots and layer plots, and most of the big changes in section 7 started as my review comments. The tether sag simulation is my own work (it's in the plug-in's repository with its own research note). When a different chatbot claimed the Neptune 4 bridges 150 mm, the assistant checked its sources and found the claim wasn't in them. I think saying all this makes the work more trustworthy, not less.&lt;/p&gt;

&lt;h2&gt;
  
  
  13. Licence
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;This paper (text and figures):&lt;/strong&gt; &lt;a href="https://github.com/The-Dorkknight/putting-plastic-on-a-pedestal/blob/main/LICENSE.md" rel="noopener noreferrer"&gt;Creative Commons Attribution-ShareAlike 4.0&lt;/a&gt;. Share and adapt it, credit me, and keep it under the same licence.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The cover art&lt;/strong&gt; was supplied by the author for this paper and is under the same licence as the text.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The code&lt;/strong&gt; (the plug-in, and the figure scripts in &lt;code&gt;tools/&lt;/code&gt; here): &lt;a href="https://polyformproject.org/licenses/noncommercial/1.0.0/" rel="noopener noreferrer"&gt;PolyForm Noncommercial 1.0.0&lt;/a&gt;. Free for hobbyists and anyone else not making money from it; commercial use needs my permission. That makes the code &lt;em&gt;source-available&lt;/em&gt;, not open source.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The methods in section 9&lt;/strong&gt; are dedicated to the public: anyone may use them, including commercially, in their own code. The licence above covers my code and words, not the ideas.&lt;/li&gt;
&lt;li&gt;No warranty, no liability. 3D printers are hot and full of moving parts; supports that fall over can ruin a print. Not legal advice.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  14. How to cite
&lt;/h2&gt;

&lt;blockquote&gt;
&lt;p&gt;Armstrong, M. (2026). &lt;em&gt;Putting Plastic on a Pedestal: Experimental methods of plastic waste reduction for 3D printing.&lt;/em&gt; Version 1.0, first published 10 November 2026. &lt;a href="https://github.com/The-Dorkknight/putting-plastic-on-a-pedestal" rel="noopener noreferrer"&gt;https://github.com/The-Dorkknight/putting-plastic-on-a-pedestal&lt;/a&gt;&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;A &lt;code&gt;CITATION.cff&lt;/code&gt; file is included, so GitHub shows a "Cite this repository" button. If this repository is archived on Zenodo, cite the DOI it gets there.&lt;/p&gt;

&lt;h2&gt;
  
  
  15. References
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;J. Dumas, J. Hergel, S. Lefebvre. &lt;em&gt;Bridging the Gap: Automated Steady Scaffoldings for 3D Printing.&lt;/em&gt; ACM Transactions on Graphics 33(4), SIGGRAPH 2014. &lt;a href="https://doi.org/10.1145/2601097.2601153" rel="noopener noreferrer"&gt;https://doi.org/10.1145/2601097.2601153&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;J. Vaněk, J. A. G. Galicia, B. Beneš. &lt;em&gt;Clever Support: Efficient Support Structure Generation for Digital Fabrication.&lt;/em&gt; Computer Graphics Forum 33(5), 2014.&lt;/li&gt;
&lt;li&gt;T. Freire, S. Hornus, S. Perchy, S. Lefebvre. &lt;em&gt;Procedural bridges-and-pillars support generation.&lt;/em&gt; Eurographics 2022 short papers. &lt;a href="https://diglib.eg.org/bitstream/handle/10.2312/egs20221025/029-032.pdf" rel="noopener noreferrer"&gt;https://diglib.eg.org/bitstream/handle/10.2312/egs20221025/029-032.pdf&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Autodesk Inc. &lt;em&gt;Building and attaching support structures for 3D printing.&lt;/em&gt; US 10,775,770 B2. &lt;a href="https://patents.google.com/patent/US10775770B2/en" rel="noopener noreferrer"&gt;https://patents.google.com/patent/US10775770B2/en&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;em&gt;Automatic reusable support for 3D printing.&lt;/em&gt; US 2021/0283856 A1 (not read). &lt;a href="https://patents.google.com/patent/US20210283856A1/en" rel="noopener noreferrer"&gt;https://patents.google.com/patent/US20210283856A1/en&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Prusa Research. &lt;em&gt;Organic supports.&lt;/em&gt; &lt;a href="https://help.prusa3d.com/article/organic-supports_480131" rel="noopener noreferrer"&gt;https://help.prusa3d.com/article/organic-supports_480131&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;a href="mailto:5@xes"&gt;5@xes&lt;/a&gt;. &lt;em&gt;CustomSupportCylinder&lt;/em&gt; (Cura plug-in). &lt;a href="https://github.com/5axes/CustomSupportCylinder" rel="noopener noreferrer"&gt;https://github.com/5axes/CustomSupportCylinder&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;CNC Kitchen. &lt;em&gt;Arc Overhangs in PrusaSlicer.&lt;/em&gt; &lt;a href="https://cnckitchen.com/blog/arc-overhangs-in-prusaslicer" rel="noopener noreferrer"&gt;https://cnckitchen.com/blog/arc-overhangs-in-prusaslicer&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Fabbaloo. &lt;em&gt;Support-free horizontal FFF overhangs with wave-based toolpaths.&lt;/em&gt; &lt;a href="https://www.fabbaloo.com/?p=238439" rel="noopener noreferrer"&gt;https://www.fabbaloo.com/?p=238439&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;MIT Fab Academy. &lt;em&gt;Week 3 bridging tests.&lt;/em&gt; &lt;a href="https://fab.cba.mit.edu/classes/MAS.863/EECS/Week3.html" rel="noopener noreferrer"&gt;https://fab.cba.mit.edu/classes/MAS.863/EECS/Week3.html&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;3D Mag. &lt;em&gt;3D printing bridging: settings, materials and limits.&lt;/em&gt; &lt;a href="https://www.3dmag.com/3d-wikipedia/3d-printing-bridging-settings-materials-and-limits/" rel="noopener noreferrer"&gt;https://www.3dmag.com/3d-wikipedia/3d-printing-bridging-settings-materials-and-limits/&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;CC 3D Labs. &lt;em&gt;Bridging in 3D printing: a practical guide.&lt;/em&gt; &lt;a href="https://cc3dlabs.com/bridging-in-3d-printing-a-practical-guide/" rel="noopener noreferrer"&gt;https://cc3dlabs.com/bridging-in-3d-printing-a-practical-guide/&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Makerforums. &lt;em&gt;Bridge distance just over 2" with hardly any sag.&lt;/em&gt; &lt;a href="https://forum.makerforums.info/t/check-out-this-bridge-distance-just-over-2-of-span-with-hardly-any-sag/41167" rel="noopener noreferrer"&gt;https://forum.makerforums.info/t/check-out-this-bridge-distance-just-over-2-of-span-with-hardly-any-sag/41167&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Tom's Hardware. &lt;em&gt;Elegoo Neptune 4 review.&lt;/em&gt; &lt;a href="https://www.tomshardware.com/reviews/elegoo-neptune-4" rel="noopener noreferrer"&gt;https://www.tomshardware.com/reviews/elegoo-neptune-4&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;M. Armstrong. &lt;em&gt;Fin-ally Supported: shark-fin supports for long, flat, tilted overhangs.&lt;/em&gt; 2026. &lt;a href="https://dev.to/open-madscience/fin-ally-supported-363n"&gt;https://dev.to/open-madscience/fin-ally-supported-363n&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;M. Armstrong. &lt;em&gt;Hybrid Supports&lt;/em&gt; (Cura plug-in), version 0.6.6-wip, with its research notes (&lt;code&gt;docs/RESEARCH.md&lt;/code&gt;) and the tether sag simulation (&lt;code&gt;tools/tethersim/&lt;/code&gt;). &lt;a href="https://github.com/The-Dorkknight/cura-hybrid-supports" rel="noopener noreferrer"&gt;https://github.com/The-Dorkknight/cura-hybrid-supports&lt;/a&gt;
&lt;/li&gt;
&lt;/ol&gt;




&lt;p&gt;Project started 2026-10-03 (the plug-in) · paper first published 2026-11-10 · Figures: &lt;code&gt;tools/figures.py&lt;/code&gt; from the data in &lt;code&gt;data/&lt;/code&gt; · PDF: &lt;code&gt;tools/make_pdf.py&lt;/code&gt;&lt;/p&gt;

</description>
      <category>curaplugin</category>
      <category>supportstructures</category>
      <category>filamentsaving</category>
      <category>defensivepublication</category>
    </item>
    <item>
      <title>Gyroid Filter</title>
      <dc:creator>Matthew Armstrong</dc:creator>
      <pubDate>Mon, 09 Nov 2026 20:05:13 +0000</pubDate>
      <link>https://dev.to/open-madscience/gyroid-filter-1nga</link>
      <guid>https://dev.to/open-madscience/gyroid-filter-1nga</guid>
      <description>&lt;p&gt;Pick a &lt;strong&gt;flat face&lt;/strong&gt; or a &lt;strong&gt;cylinder&lt;/strong&gt; on a model and make part or all of it &lt;strong&gt;open-cell gyroid&lt;/strong&gt;, for printing &lt;strong&gt;fillable carbon filters&lt;/strong&gt;. The gyroid replaces the solid wall, so air can pass through while the carbon granules stay inside.&lt;/p&gt;

&lt;p&gt;It works with &lt;strong&gt;Cura&lt;/strong&gt;, &lt;strong&gt;PrusaSlicer&lt;/strong&gt; and &lt;strong&gt;OrcaSlicer&lt;/strong&gt;:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Slicer&lt;/th&gt;
&lt;th&gt;How you use it&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Cura&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;A plug-in adds a &lt;em&gt;Gyroid Filter&lt;/em&gt; tool. Select the model, pick the tool, click a face or cylinder.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;PrusaSlicer&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Pick in the app, press &lt;em&gt;PrusaSlicer 3MF&lt;/em&gt;, open the file. The open-cell regions are already there as modifiers.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;OrcaSlicer&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Same, with &lt;em&gt;OrcaSlicer 3MF&lt;/em&gt; (see what is tested).&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;PrusaSlicer and OrcaSlicer have no plug-in system that can pick faces, so for them the companion app does the picking and hands the slicer a normal 3MF. Your original model is never changed.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fb9ubfd4ijud3oymq0mtr.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fb9ubfd4ijud3oymq0mtr.webp" alt="The app with the example cartridge, outer wall picked" width="800" height="475"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Quick start (no terminal needed)
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Unzip, and &lt;strong&gt;double-click&lt;/strong&gt; the launcher for your computer:

&lt;ul&gt;
&lt;li&gt;Linux: &lt;code&gt;GyroidFilter-Linux.sh&lt;/code&gt; (choose &lt;em&gt;Run&lt;/em&gt; or &lt;em&gt;Run in Terminal&lt;/em&gt;)&lt;/li&gt;
&lt;li&gt;Windows: &lt;code&gt;GyroidFilter-Windows.bat&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Mac: &lt;code&gt;GyroidFilter-Mac.command&lt;/code&gt; (first time: right-click, then &lt;em&gt;Open&lt;/em&gt;, because it is unsigned)&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;Your browser opens. Press &lt;strong&gt;Load example cartridge&lt;/strong&gt; (or open your own STL/3MF).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Click&lt;/strong&gt; a face or a cylinder on the model. The picked surface turns orange; a yellow preview shows the region that will become open-cell.&lt;/li&gt;
&lt;li&gt;Press &lt;strong&gt;Add open-cell region&lt;/strong&gt;. Repeat for other surfaces.&lt;/li&gt;
&lt;li&gt;Press &lt;strong&gt;PrusaSlicer 3MF&lt;/strong&gt;, &lt;strong&gt;OrcaSlicer 3MF&lt;/strong&gt; or &lt;strong&gt;Region STLs&lt;/strong&gt;, and open the file in your slicer. For Cura see below.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Needs Python 3.8 or newer (nothing else; no packages to install). The launcher says what to do if Python is missing.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fuyp1zhmlj2clxn0lob9b.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fuyp1zhmlj2clxn0lob9b.webp" alt="Two regions added" width="800" height="475"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What the options mean
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Option&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Open all the way through the wall&lt;/strong&gt; (default)&lt;/td&gt;
&lt;td&gt;Measures the wall thickness under the picked surface and makes the region that deep, plus 0.3 mm so it breaks through. For a filter wall this is what lets air cross. Untick to choose a depth.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Gyroid density&lt;/strong&gt; (5–60 %)&lt;/td&gt;
&lt;td&gt;The slicer's infill density inside the region. Lower means bigger pores. How big the pores are for a given density is not measured yet: print a small test piece with your nozzle and your carbon before you commit.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Solid rim&lt;/strong&gt; (mm)&lt;/td&gt;
&lt;td&gt;Leaves a solid frame around the open area (for a cylinder: at both ends) so the part keeps its strength and has somewhere to glue or seal. 0 = the whole surface.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Only part of it&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Limits the region to a window on a flat face (rectangle or oval, in % of the face), or to a stretch along and around a cylinder.&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;"Open-cell" here means the region is printed with &lt;strong&gt;no walls, no top or bottom skin and no "small area becomes solid" rule&lt;/strong&gt;, so the gyroid strands &lt;em&gt;are&lt;/em&gt; the surface.&lt;/p&gt;

&lt;h3&gt;
  
  
  The settings it applies
&lt;/h3&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;Cura (cutting mesh)&lt;/th&gt;
&lt;th&gt;PrusaSlicer (modifier)&lt;/th&gt;
&lt;th&gt;OrcaSlicer (modifier)&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Pattern&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;infill_pattern&lt;/code&gt; = gyroid&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;fill_pattern&lt;/code&gt; = gyroid&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;sparse_infill_pattern&lt;/code&gt; = gyroid&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Density&lt;/td&gt;
&lt;td&gt;&lt;code&gt;infill_sparse_density&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;&lt;code&gt;fill_density&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;&lt;code&gt;sparse_infill_density&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Walls&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;wall_line_count&lt;/code&gt; = 0&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;perimeters&lt;/code&gt; = 0&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;wall_loops&lt;/code&gt; = 0&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Top/bottom skin&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;top_layers&lt;/code&gt;, &lt;code&gt;bottom_layers&lt;/code&gt; = 0&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;top_solid_layers&lt;/code&gt;, &lt;code&gt;bottom_solid_layers&lt;/code&gt;, &lt;code&gt;*_solid_min_thickness&lt;/code&gt; = 0&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;top_shell_layers&lt;/code&gt;, &lt;code&gt;bottom_shell_layers&lt;/code&gt;, &lt;code&gt;*_shell_thickness&lt;/code&gt; = 0&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Small areas&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;min_infill_area&lt;/code&gt; = 0&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;solid_infill_below_area&lt;/code&gt; = 0&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;minimum_sparse_infill_area&lt;/code&gt; = 0&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;If a 3MF does not do what you want in your slicer, you can do the same by hand: add the &lt;strong&gt;Region STLs&lt;/strong&gt; as modifiers and set these five things.&lt;/p&gt;

&lt;h2&gt;
  
  
  Cura
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;In the app open the &lt;strong&gt;Cura plug-in&lt;/strong&gt; tab, press &lt;strong&gt;Install&lt;/strong&gt; next to your Cura version, restart Cura.&lt;/li&gt;
&lt;li&gt;In Cura: select your model, choose the &lt;strong&gt;Gyroid Filter&lt;/strong&gt; tool in the left toolbar, and click a face or cylinder.&lt;/li&gt;
&lt;li&gt;The region appears as a child of the model (a &lt;em&gt;cutting mesh&lt;/em&gt; carrying the settings above). &lt;strong&gt;Slice&lt;/strong&gt;, then look at the layer view.&lt;/li&gt;
&lt;li&gt;Change density or rim in the tool panel and press &lt;strong&gt;Re-apply to last pick&lt;/strong&gt;; &lt;strong&gt;Remove gyroid regions&lt;/strong&gt; clears them.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Cura's log (&lt;code&gt;cura.log&lt;/code&gt;) gets a line &lt;code&gt;GyroidFilter: core self-test PASS&lt;/code&gt; when the plug-in loads; the Cura tab in the app reads it and shows a green tick.&lt;/p&gt;

&lt;h2&gt;
  
  
  What is tested, and what is not
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Part&lt;/th&gt;
&lt;th&gt;Status&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Face and cylinder detection, region solids, thickness probe&lt;/td&gt;
&lt;td&gt;✅ unit tests on the example cartridge, boxes, a half-pipe and a finely tessellated cylinder&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;PrusaSlicer&lt;/strong&gt; 3MF&lt;/td&gt;
&lt;td&gt;✅ sliced in a real PrusaSlicer 2.7.2: inside the region only gyroid-style infill, no perimeters; above the rim normal walls; flat oval window in a plate also checked&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;OrcaSlicer&lt;/strong&gt; 3MF&lt;/td&gt;
&lt;td&gt;⚠️ &lt;strong&gt;not opened in Orca yet.&lt;/strong&gt; Written to Orca's own layout by reading Orca's importer source and its setting names. If it misbehaves, use the Region STLs route&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Cura plug-in&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;✅ &lt;strong&gt;Tested by the author in a real Cura on Linux Mint: it works well.&lt;/strong&gt; Also tested against stand-ins for Cura's API, and every Cura setting name is checked against Cura's &lt;code&gt;fdmprinter.def.json&lt;/code&gt;. Windows and Mac are not tested yet. Please report what happens on your setup&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Web app&lt;/td&gt;
&lt;td&gt;✅ driven in headless Chrome (pick, preview, add, export, Cura tab): no JavaScript errors&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Linux launcher&lt;/td&gt;
&lt;td&gt;✅ run for real. &lt;strong&gt;Windows and Mac launchers were not run&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Printing a filter&lt;/td&gt;
&lt;td&gt;❌ nothing has been printed or filtered with this yet. Pore size versus density is unmeasured&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;7th-gen Core i3&lt;/td&gt;
&lt;td&gt;⚠️ not measured on one. On the 2-core test machine a 200,000-triangle model loads and picks in about a second and needs about 200 MB; see &lt;a href="https://github.com/The-Dorkknight/gyroid-filter/blob/main/DESIGN_NOTES.md#measurements" rel="noopener noreferrer"&gt;DESIGN_NOTES.md&lt;/a&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  More
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;a href="https://github.com/The-Dorkknight/gyroid-filter/blob/main/HOW_IT_WORKS.md" rel="noopener noreferrer"&gt;HOW_IT_WORKS.md&lt;/a&gt;: what it reads, changes and never does&lt;/li&gt;
&lt;li&gt;
&lt;a href="https://github.com/The-Dorkknight/gyroid-filter/blob/main/DESIGN_NOTES.md" rel="noopener noreferrer"&gt;DESIGN_NOTES.md&lt;/a&gt;: research notes, decisions, rejected ideas, measurements&lt;/li&gt;
&lt;li&gt;
&lt;a href="https://github.com/The-Dorkknight/gyroid-filter/blob/main/CHANGELOG.md" rel="noopener noreferrer"&gt;CHANGELOG.md&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/gyroid-filter/blob/main/docs/HISTORY.md" rel="noopener noreferrer"&gt;docs/HISTORY.md&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/gyroid-filter/blob/main/THIRD_PARTY.md" rel="noopener noreferrer"&gt;THIRD_PARTY.md&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/gyroid-filter/blob/main/CONTRIBUTING.md" rel="noopener noreferrer"&gt;CONTRIBUTING.md&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/gyroid-filter/blob/main/SECURITY.md" rel="noopener noreferrer"&gt;SECURITY.md&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Related projects by the same author: &lt;a href="https://github.com/The-Dorkknight/flatlock" rel="noopener noreferrer"&gt;flatlock&lt;/a&gt; / Shift-LoK (interlocking walls for the same three slicers) and &lt;a href="https://github.com/The-Dorkknight/cura-hybrid-supports" rel="noopener noreferrer"&gt;cura-hybrid-supports&lt;/a&gt; (another Cura plug-in). This project shares their approach of working inside the slicer's own settings rather than rewriting G-code.&lt;/p&gt;

&lt;p&gt;Project started 2026-11-09 09:55. Built by Matthew Armstrong with Claude.&lt;/p&gt;




&lt;p&gt;&lt;strong&gt;Source code:&lt;/strong&gt; &lt;a href="https://github.com/The-Dorkknight/gyroid-filter" rel="noopener noreferrer"&gt;The-Dorkknight/gyroid-filter on GitHub&lt;/a&gt;&lt;/p&gt;

</description>
      <category>3dprintedfilter</category>
    </item>
    <item>
      <title>WARNING DONT INSTALL YET!!!!!!!!!!!!!</title>
      <dc:creator>Matthew Armstrong</dc:creator>
      <pubDate>Mon, 09 Nov 2026 18:38:58 +0000</pubDate>
      <link>https://dev.to/open-madscience/deck-deck-goose-38go</link>
      <guid>https://dev.to/open-madscience/deck-deck-goose-38go</guid>
      <description>&lt;h1&gt;
  
  
  Deck Deck Goose 0.6.7-wip
&lt;/h1&gt;

&lt;p&gt;POSSIBLY HAS A BUG WHICH DISABLES BAZZITE DESKTOP MODE, THO I SUSPECT IT IS MORE LIKELEY ANOTHER APP IM WORKING ON( BAZZITE DEEP SLEEP) CAUSING THE ISSUE&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F1owpg0wm2m5rgcmoru20.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F1owpg0wm2m5rgcmoru20.webp" alt="Deck Deck Goose: a cyborg goose riding a flying Steam Deck over a neon city" width="799" height="436"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;WORK IN PROGRESS: MANY FEATURES MAY BE BROKEN&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Source-available · Non-commercial.&lt;/em&gt; Copyright Matthew Armstrong. Licensed under the &lt;a href="https://github.com/The-Dorkknight/deck-deck-goose/blob/main/LICENSE.md" rel="noopener noreferrer"&gt;PolyForm Noncommercial License 1.0.0&lt;/a&gt;. Provided as is, &lt;strong&gt;no warranty and no liability&lt;/strong&gt;. This is a plain-English summary, not legal advice. &lt;strong&gt;Back up saves you cannot lose before moving or cleaning games.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Formerly &lt;em&gt;This PC for Linux&lt;/em&gt;. Made for Bazzite and Nobara (and should work on Linux Mint). &lt;strong&gt;The Steam library manager (Games tab) also runs on Windows and macOS&lt;/strong&gt;; everything else is Linux-only and is hidden there.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;This PC&lt;/strong&gt;: a normal-looking folder for any file manager, with your disks by name, Users, Documents and so on, instead of &lt;code&gt;/var/home/you&lt;/code&gt; and &lt;code&gt;/run/media/you/...&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Games&lt;/strong&gt;: a Steam game manager. It finds your games and shows their artwork, moves a game to another drive so Steam finds it there, clears junk, and compresses games on btrfs drives.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Game Mode tiles&lt;/strong&gt;: one click adds &lt;em&gt;Deck Deck Goose&lt;/em&gt; and &lt;em&gt;Airlock&lt;/em&gt; to Steam's library, so the app opens full screen in Game Mode on a handheld (ROG Ally, Steam Deck).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;On-screen keyboard&lt;/strong&gt;: installs (if needed) and turns on KDE's touch keyboard for Desktop Mode, which Bazzite KDE and Nobara KDE don't do by themselves.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Sound&lt;/strong&gt;: fixes the quiet ROG Ally speakers on Bazzite in one press, boosts the volume up to 150%, and lets you pick the sound output. Works in Desktop and Game Mode.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Airlock&lt;/strong&gt;: tap the &lt;em&gt;Airlock&lt;/em&gt; tile in Game Mode and a KDE desktop opens in a window &lt;strong&gt;inside Game Mode&lt;/strong&gt; while your game keeps running; close it and you are straight back. See below. &lt;em&gt;Untested on real hardware.&lt;/em&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Wake&lt;/strong&gt; (Zombie-Deck, bundled): brings a handheld properly back after sleep or hibernation, with a green or red brain on screen to show how it went. Installed and switched on or off from the Wake tab. Also closes Airlock before sleep.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Frb4l4isr7a9ljww0fbqr.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Frb4l4isr7a9ljww0fbqr.webp" alt="The Games tab in the Deck Deck Goose theme: Steam libraries, game covers, move and compress buttons" width="800" height="654"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Wake: Zombie-Deck
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F1094zxymxy0azmmihgu0.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F1094zxymxy0azmmihgu0.webp" alt="The Wake tab: green band, version, buttons to repair, turn off, show log or remove" width="800" height="646"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://github.com/The-Dorkknight/deck-deck-goose/blob/main/components/zombie-deck/README.md" rel="noopener noreferrer"&gt;Zombie-Deck&lt;/a&gt; is a small separate project (its own tests, licence and notes sit in &lt;code&gt;components/zombie-deck/&lt;/code&gt;) that fixes the "zombie" state some handhelds wake up in (wrong power limit, dead controller, Wi-Fi gone). Deck Deck Goose carries a copy and only installs, updates, removes and switches it, and shows its status. The Wake tab shows a green band when it is armed and the last wake worked, red when it failed, amber when it is installed but not armed. It never installs over a newer Zombie-Deck. Install and remove ask for your password once and only work in Desktop Mode. &lt;strong&gt;Status:&lt;/strong&gt; confirmed by the owner on a Bazzite ROG Ally after hibernation; plain suspend, Nobara Handheld and the red path are not yet confirmed on real hardware.&lt;/p&gt;

&lt;h2&gt;
  
  
  Windows and Mac
&lt;/h2&gt;

&lt;p&gt;Only the Games tab, from the same engine as the sibling GameShelf project. Differences: Steam is found in the Windows registry or &lt;code&gt;~/Library/Application Support/Steam&lt;/code&gt;; drives come from Windows or from &lt;code&gt;mount&lt;/code&gt; (the Data volume is shown as &lt;em&gt;Macintosh HD&lt;/em&gt;); Steam must be closed before moving (the window can close it for you); targets on FAT32/MSDOS (4 GB limit), network drives and read-only volumes are refused with a reason, and so are exFAT-vs-NTFS cases that would break games. Windows adds &lt;em&gt;Compress / Uncompress&lt;/em&gt; with the Windows compressor (&lt;code&gt;compact /EXE&lt;/code&gt;). &lt;strong&gt;Tested only against simulated Windows and Mac set-ups&lt;/strong&gt; (invented Steam folders, a stand-in for &lt;code&gt;compact&lt;/code&gt;, a sample &lt;code&gt;mount&lt;/code&gt; listing). Never run on a real PC or Mac yet.&lt;/p&gt;

&lt;h2&gt;
  
  
  Game Mode (ROG Ally, Steam Deck-style Bazzite)
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fbkzl2clj7febe0yj6jcq.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fbkzl2clj7febe0yj6jcq.webp" alt="The Games tab in Game Mode: big buttons, Back to Steam, no Move button" width="800" height="500"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Set it up once, in Desktop Mode:&lt;/strong&gt; Games tab &amp;gt; &lt;strong&gt;Add to Game Mode&lt;/strong&gt;. Steam closes for a moment, &lt;em&gt;Deck Deck Goose&lt;/em&gt; and &lt;em&gt;Airlock&lt;/em&gt; are added to Steam as non-Steam shortcuts with their own artwork, and Steam starts again. In Game Mode they are under &lt;strong&gt;Library &amp;gt; Non-Steam&lt;/strong&gt;. An old &lt;em&gt;This PC Games&lt;/em&gt; tile from earlier versions is replaced.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fdeck-deck-goose%2Fmain%2Fdocs%2Fimages%2Fgame-mode-tile.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fraw.githubusercontent.com%2FThe-Dorkknight%2Fdeck-deck-goose%2Fmain%2Fdocs%2Fimages%2Fgame-mode-tile.webp" alt="The Deck Deck Goose and Airlock tiles" width="620" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;What you get in Game Mode:&lt;/strong&gt; the Games tab, full screen, with bigger buttons for the touchscreen. &lt;strong&gt;Estimate&lt;/strong&gt;, &lt;strong&gt;Compress&lt;/strong&gt; and the junk items marked safe work while Steam is open (Compress refuses a game that is running). &lt;strong&gt;Moving&lt;/strong&gt; isn't offered in Game Mode, because Steam has to be closed for that and in Game Mode Steam &lt;em&gt;is&lt;/em&gt; the screen; use Steam's own &lt;strong&gt;Settings &amp;gt; Storage&lt;/strong&gt; there, or this tool's &lt;strong&gt;Move&lt;/strong&gt; from Desktop Mode. Press &lt;strong&gt;Back to Steam&lt;/strong&gt; to close it.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;How it works:&lt;/strong&gt; the tile runs a small launcher script that starts the program and opens its page in your browser in kiosk (full-screen) mode; Firefox (a Flatpak on Bazzite) is used if present, otherwise Chrome, Chromium, Brave or Edge. Firefox gets its own separate, quiet profile, so your normal browsing isn't touched. The touchscreen works; for buttons, set the tile's controller layout to Steam's &lt;strong&gt;Web Browser&lt;/strong&gt; template.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Remove them&lt;/strong&gt; with &lt;strong&gt;Remove tiles&lt;/strong&gt; (Games tab), or &lt;code&gt;python3 -m ddg gamemode remove&lt;/code&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Airlock: a desktop inside Game Mode
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Feub9a81rk31oepitufdc.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Feub9a81rk31oepitufdc.webp" alt="The Airlock tab: what it does, Open Airlock, Back to Game, the fallback order" width="800" height="896"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;How it works (not a video stream):&lt;/strong&gt; Airlock starts a second copy of KDE's window manager (KWin) as a window inside Game Mode. It draws a real Plasma desktop, live, and gamescope (Game Mode's screen manager) puts that window on screen like a game. Each frame is drawn twice, which is why it is a little less smooth than Desktop Mode.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Smoothness and size (0.6.4):&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Desktop size:&lt;/strong&gt; &lt;em&gt;Bigger &amp;amp; faster&lt;/em&gt; (default) opens the desktop at about 720p (1280x720 on the Ally) and lets Game Mode scale it up: icons and text 1.5x bigger, and far fewer pixels to draw each frame. &lt;em&gt;Sharp&lt;/em&gt; uses the screen's full resolution. Earlier versions asked X11 for the size and could get a bigger virtual screen than the panel, which made everything tiny.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Connection:&lt;/strong&gt; &lt;em&gt;Automatic&lt;/em&gt; first connects straight to gamescope over Wayland (one copy less per frame), and falls back to the older X11 route by itself if that doesn't work; whichever worked is remembered.&lt;/li&gt;
&lt;li&gt;The Airlock tab shows which size and connection were used last time, and whether KWin drew on the graphics chip or on the CPU (CPU drawing caps it near 20 fps; send &lt;code&gt;nested-full.log&lt;/code&gt; if you see it).&lt;/li&gt;
&lt;li&gt;Steam treats the Airlock tile as a game, so its Quick Access &lt;strong&gt;Performance&lt;/strong&gt; settings (frame limit, refresh rate, power limit) apply to it too. Untick any frame limit there for the Airlock tile.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;Controller (0.6.5):&lt;/strong&gt; touch worked in Airlock but the d-pad, sticks and buttons did nothing. In Game Mode Steam hands the Airlock tile a &lt;em&gt;gamepad&lt;/em&gt;, and the nested desktop only understands mouse, keyboard and touch. Airlock now reads the pad itself and types for it through a virtual mouse and keyboard (&lt;code&gt;/dev/uinput&lt;/code&gt;), only while Airlock is open:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Pad&lt;/th&gt;
&lt;th&gt;Does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Left stick&lt;/td&gt;
&lt;td&gt;Mouse&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Right stick&lt;/td&gt;
&lt;td&gt;Scroll&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;A or RT&lt;/td&gt;
&lt;td&gt;Left click (hold to drag)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;X or LT&lt;/td&gt;
&lt;td&gt;Right click&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;B&lt;/td&gt;
&lt;td&gt;Esc (closes menus and the Start menu)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Y&lt;/td&gt;
&lt;td&gt;Super key (opens and closes the Start menu)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;D-pad&lt;/td&gt;
&lt;td&gt;Arrow keys&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;LB / RB&lt;/td&gt;
&lt;td&gt;Backspace / Tab&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Menu / View&lt;/td&gt;
&lt;td&gt;Enter / Alt+Tab&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Left stick click&lt;/td&gt;
&lt;td&gt;Middle click&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The Airlock tab has &lt;strong&gt;Controller&lt;/strong&gt; (on/off) and &lt;strong&gt;Pointer speed&lt;/strong&gt;, and shows whether a pad is seen. If the system won't let the app write &lt;code&gt;/dev/uinput&lt;/code&gt;, the tab offers &lt;strong&gt;Allow controller&lt;/strong&gt; (a one-time password window; adds a udev rule). &lt;code&gt;ddg airlock pad&lt;/code&gt; shows what is detected; &lt;code&gt;ddg airlock pad --watch&lt;/code&gt; prints what the pad sends for 15 seconds. Prefer Steam's own layout? Switch &lt;em&gt;Controller&lt;/em&gt; off and give the Airlock tile &lt;em&gt;Gamepad with Mouse Trackpad&lt;/em&gt; or &lt;em&gt;Keyboard (WASD) and Mouse&lt;/em&gt; in Steam. The Steam, Quick Access and Guide buttons stay with Steam. &lt;strong&gt;Not yet run on the Ally&lt;/strong&gt;; the layout is my choice, tell me what you'd change.&lt;/p&gt;

&lt;p&gt;Ported from the separate &lt;em&gt;Airlock workbench&lt;/em&gt; prototype. &lt;strong&gt;Status: work in progress, written from documentation and tested only with stand-ins; it has not run on a real ROG Ally yet.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Tap the &lt;strong&gt;Airlock&lt;/strong&gt; tile in Game Mode. It starts a second, nested KWin compositor as a window inside gamescope, with its own D-Bus session so it cannot clash with Steam's, and runs Plasma in it. Your game and Steam keep running underneath. To leave: close the desktop, or open &lt;strong&gt;Back to Game&lt;/strong&gt; from its app menu (add it with the Airlock tab &amp;gt; &lt;em&gt;Add "Back to Game"&lt;/em&gt;), or Steam button &amp;gt; Exit game.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Touch-mode button (0.6.7):&lt;/strong&gt; a small square in the top-left corner of the Airlock desktop. Tap it to cycle what a finger on the screen does; a popup in the middle says which mode is on, and the letter on the button shows it afterwards:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Letter&lt;/th&gt;
&lt;th&gt;Mode&lt;/th&gt;
&lt;th&gt;What a finger does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;M&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Mouse&lt;/td&gt;
&lt;td&gt;Tap clicks, &lt;strong&gt;drag works&lt;/strong&gt; (moving folders). Right-click with X or LT on the pad.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;T&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Native touch&lt;/td&gt;
&lt;td&gt;Real touch; press and hold for the right-click menu.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;P&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Trackpad&lt;/td&gt;
&lt;td&gt;Slide to move the pointer, tap to click.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;R&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Right-click&lt;/td&gt;
&lt;td&gt;Every tap right-clicks (tap &lt;strong&gt;M&lt;/strong&gt; again afterwards).&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fv9wgt0h8wfq5z7ammmsa.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fv9wgt0h8wfq5z7ammmsa.webp" alt="The touch-mode button and its popup for each mode (simulated screens)" width="799" height="502"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;It changes gamescope's touch setting (&lt;code&gt;STEAM_TOUCH_CLICK_MODE&lt;/code&gt;, which Steam normally sets) and &lt;strong&gt;never follows you out&lt;/strong&gt;: Airlock saves Steam's own value when it opens and puts it back when it closes, however it closes. If Airlock is killed too hard to do that, the next Airlock or Deck Deck Goose start puts it back. The button only exists inside Airlock (never in the real Desktop Mode or Game Mode), and only in Game Mode. Airlock remembers the last mode you picked for the next Airlock. Steam can change the setting itself, for example while its menu is open; the button shows the real mode within a couple of seconds, so tap it again if needed. Switch it off in the Airlock tab (&lt;strong&gt;Touch button&lt;/strong&gt;). In Dolphin, dropping a folder on another asks &lt;em&gt;Move Here / Copy Here / Link Here&lt;/em&gt;: that menu is normal. &lt;strong&gt;Not yet run on the Ally.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;If it can't open, it tries these in order&lt;/strong&gt; (and tells you which): 1. full Plasma desktop in the window; 2. &lt;em&gt;lite&lt;/em&gt; (just Dolphin and Konsole, for less memory); 3. a real switch to Desktop Mode via &lt;code&gt;steamos-session-select&lt;/code&gt; (&lt;strong&gt;closes your game, so it always asks first; no answer counts as No&lt;/strong&gt;); 4. instructions for switching by hand. It &lt;strong&gt;never reboots or logs you out by itself&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Leftovers (fixed in 0.6.1 after the first real test):&lt;/strong&gt; if a try fails, a desktop it started could keep running on Game Mode's own screen, outside Airlock, so it looked like a second, real desktop with no way out except logging out. Airlock now notes which desktop programs were running before it started and closes anything new it left behind, between tries, when you close it and when Steam's &lt;em&gt;Exit game&lt;/em&gt; is used. In Game Mode a Plasma shell never belongs there, so one left over from an earlier run is closed when you next open Airlock. Stuck anyway? &lt;strong&gt;Steam button &amp;gt; Airlock tile &amp;gt; Exit game&lt;/strong&gt;, or the Airlock tab's &lt;strong&gt;Close leftovers&lt;/strong&gt; button (&lt;code&gt;ddg airlock cleanup&lt;/code&gt;).&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Safety:&lt;/strong&gt; one Airlock at a time (lock file); a 15-second watchdog kills a start that never shows anything and moves down the ladder; a ready-file handshake so a quick close is not mistaken for a failure; the whole process group is killed when it closes or when Steam sends &lt;em&gt;Exit game&lt;/em&gt;; free-RAM checks (1500 MB for full, 600 MB for lite) and a low-battery note; everything is logged with short codes (&lt;code&gt;A0OK&lt;/code&gt;...&lt;code&gt;ACDP&lt;/code&gt;) in &lt;code&gt;~/.local/state/deck-deck-goose/portal.log&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The colour bar&lt;/strong&gt; at the top of the window shows where you are: &lt;strong&gt;amber&lt;/strong&gt; = Airlock open (game still running), &lt;strong&gt;blue&lt;/strong&gt; = Desktop Mode, &lt;strong&gt;green&lt;/strong&gt; = Game Mode. Scripts in &lt;code&gt;~/.config/deck-deck-goose/hooks/mode-change.d/&lt;/code&gt; are called with &lt;code&gt;portal&lt;/code&gt;, &lt;code&gt;desktop&lt;/code&gt; or &lt;code&gt;game&lt;/code&gt; (an example for colouring controller lights is included, disabled).&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Optional:&lt;/strong&gt; &lt;em&gt;Close Airlock before sleep&lt;/em&gt; (asks for your password once, Desktop Mode only) so you wake in plain Game Mode instead of a frozen nested desktop. With Zombie-Deck installed it is a script in &lt;code&gt;/etc/zombie-deck/sleep.d/&lt;/code&gt;; without it, a small unit on &lt;code&gt;sleep.target&lt;/code&gt;. (0.5.0 put it in &lt;code&gt;/etc/systemd/system-sleep/&lt;/code&gt;, which systemd never reads; fixed in 0.6.0.)&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Try it first:&lt;/strong&gt; Airlock tab &amp;gt; &lt;strong&gt;Check this device&lt;/strong&gt; (or &lt;code&gt;ddg airlock probe --live&lt;/code&gt;) is read-only, writes &lt;code&gt;~/deck-deck-goose-probe.txt&lt;/code&gt; and shows which KWin options your system has. Please send me that file from your Ally. Command line: &lt;code&gt;ddg airlock start|stop|status|plan|probe|cleanup|sleep-hook on|off|back-entry&lt;/code&gt;. In Desktop Mode, &lt;code&gt;start --force&lt;/code&gt; opens a test window.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Not verified:&lt;/strong&gt; whether Bazzite's and Nobara's KWin accept the flags used (KWin documents &lt;code&gt;--x11-display&lt;/code&gt;, &lt;code&gt;--socket&lt;/code&gt;, &lt;code&gt;--exit-with-session&lt;/code&gt;, &lt;code&gt;--xwayland&lt;/code&gt;; see the research notes), whether the on-screen keyboard works in the nested window, controller layout for the tile (use &lt;em&gt;Gamepad with Mouse Trackpad&lt;/em&gt;), and &lt;code&gt;steamos-session-select&lt;/code&gt; on Nobara.&lt;/p&gt;

&lt;h2&gt;
  
  
  On-screen keyboard (Desktop Mode)
&lt;/h2&gt;

&lt;p&gt;KDE has two touch keyboards: the new &lt;strong&gt;Plasma Keyboard&lt;/strong&gt; (&lt;code&gt;plasma-keyboard&lt;/code&gt;, the default on Fedora-based systems such as Nobara from Plasma 6.6) and the older &lt;strong&gt;Maliit&lt;/strong&gt;. Bazzite KDE and Nobara KDE (including Nobara Handheld on a ROG Ally) don't switch either on, and Nobara may not have one installed. If none is installed, &lt;strong&gt;Install keyboard&lt;/strong&gt; adds &lt;code&gt;plasma-keyboard&lt;/code&gt; (falling back to &lt;code&gt;maliit-keyboard&lt;/code&gt;) with the system's package manager through &lt;code&gt;pkexec&lt;/code&gt;, so KDE shows its normal password window; on Bazzite (&lt;code&gt;rpm-ostree&lt;/code&gt;) a restart is needed before it can be turned on. It refuses in Game Mode, where no password window can appear. &lt;strong&gt;This PC tab &amp;gt; On-screen keyboard &amp;gt; Turn on&lt;/strong&gt; sets it as KDE's virtual keyboard (the same as System Settings &amp;gt; Keyboard &amp;gt; Virtual Keyboard). Then tapping a text box with the touchscreen brings it up; a keyboard icon also appears in the system tray. If it doesn't appear straight away, log out and back in. It also adds a &lt;em&gt;Show on-screen keyboard&lt;/em&gt; entry to the app menu for when no text box is selected (this may not work on every Plasma version). &lt;strong&gt;Turn off&lt;/strong&gt; undoes it. Only &lt;code&gt;~/.config/kwinrc&lt;/code&gt; is changed (a backup is kept). If no keyboard can be installed, Steam's own keyboard still works in Desktop Mode while Steam is open (Steam button + X).&lt;/p&gt;

&lt;h2&gt;
  
  
  Sound (quiet ROG Ally speakers)
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2oa369acndj93ii5amwv.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2oa369acndj93ii5amwv.webp" alt="The Sound tab: the ROG Ally output, the speakers underneath at 40% and muted, Fix quiet sound, volume boost buttons" width="800" height="1018"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;On Bazzite the ROG Ally shows two outputs: &lt;strong&gt;ROG Ally&lt;/strong&gt;, which is Bazzite's speaker sound processing (an EQ), and &lt;strong&gt;Family 17h/19h/1ah HD Audio Controller&lt;/strong&gt;, the real speakers it plays through. The two volumes &lt;strong&gt;multiply&lt;/strong&gt;, so if the Family one sits at 40% the speakers never get above 40% of their real level, however high the ROG Ally slider goes. That is why the Family device sounds louder when you pick it directly.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Fix quiet sound&lt;/strong&gt; sets the built-in speakers underneath to 100% and unmutes them, and leaves your ROG Ally slider where it was, so you keep Bazzite's EQ and gain the lost volume. It never touches HDMI, USB headsets or Bluetooth.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Volume 100 / 115 / 130 / 150%&lt;/strong&gt; sets the output in use. Above 100% is a software boost (loud parts can distort, so go a step at a time); KDE's slider is allowed to 150% as well (&lt;em&gt;Raise maximum volume&lt;/em&gt;).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Use this&lt;/strong&gt; picks another output, e.g. the Family device directly: louder at the same slider position, but without the EQ, so it can sound thinner.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Re-apply at every login&lt;/strong&gt; adds a small user service (&lt;code&gt;ddg-sound.service&lt;/code&gt;) that puts these back after a restart, in Desktop and Game Mode. &lt;strong&gt;Back to normal&lt;/strong&gt; sets 100%, puts KDE's slider limit back and removes the service.&lt;/li&gt;
&lt;li&gt;Command line: &lt;code&gt;ddg sound status|fix|boost 130|use &amp;lt;name&amp;gt;|keep on|off|reset&lt;/code&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  This PC tab
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;This PC/
  Disks/      Games, Backup, USBSTICK ...   (your disks, by name)
  Users/      you, other accounts
  Documents   Downloads   Pictures   Music   Videos   Desktop
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fozjd2kdqh4tcas7bllfh.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fozjd2kdqh4tcas7bllfh.webp" alt="This PC tab after installing" width="800" height="1374"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Button&lt;/th&gt;
&lt;th&gt;What happens&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Install / update&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Creates &lt;code&gt;~/This PC&lt;/code&gt;, a timer that refreshes it every 30 seconds (new disks appear), a sidebar entry in Nemo/Thunar/Nautilus and Dolphin, and Flatpak access for the file managers you ticked. Then reads everything back to check.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Check&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Re-checks all of that without changing anything.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Remove everything&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Removes only what it added, and resets the Flatpak access it granted.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Mount&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Mounts an unmounted disk (&lt;code&gt;udisksctl&lt;/code&gt;).&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Split window: Dolphin and Nemo split with &lt;strong&gt;F3&lt;/strong&gt;; Double Commander and Krusader are two-pane by default (as far as we can tell neither is on stable Flathub, so install them natively or in a container and bookmark &lt;em&gt;This PC&lt;/em&gt; yourself).&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Flatpak access.&lt;/strong&gt; Flatpak's own documentation says &lt;code&gt;--filesystem=host&lt;/code&gt; still hides &lt;code&gt;/var&lt;/code&gt;, &lt;code&gt;/run&lt;/code&gt; and other system folders, except &lt;code&gt;/run/media&lt;/code&gt;. So each ticked app gets &lt;code&gt;flatpak override --user --filesystem=host --filesystem=home --filesystem=/run/media&lt;/code&gt; (plus &lt;code&gt;/var/mnt&lt;/code&gt; if it exists). That weakens its sandbox, so only tick file managers you use.&lt;/p&gt;

&lt;h2&gt;
  
  
  Games tab (Steam)
&lt;/h2&gt;

&lt;p&gt;The buttons are on each game's card; tapping the card itself does nothing. When a button is greyed out, the card now says why in plain text underneath (for example &lt;em&gt;Move: close Steam first&lt;/em&gt;, &lt;em&gt;Compress: needs a btrfs drive&lt;/em&gt;, &lt;em&gt;Installed twice&lt;/em&gt;), because a touchscreen never shows hover tips.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fkxa18b4rloso46zdwhak.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fkxa18b4rloso46zdwhak.webp" alt="Game cards with the reason under each greyed-out button, and Delete this copy on a game installed twice (simulated library)" width="800" height="1240"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;What&lt;/th&gt;
&lt;th&gt;How&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Find games&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Reads Steam's own files: native Steam (&lt;code&gt;~/.local/share/Steam&lt;/code&gt;) and Flatpak Steam, every library listed in &lt;code&gt;libraryfolders.vdf&lt;/code&gt;, and each &lt;code&gt;appmanifest_&amp;lt;id&amp;gt;.acf&lt;/code&gt;. Proton and the Steam runtimes are hidden unless you tick &lt;em&gt;Show Proton and runtimes&lt;/em&gt;.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Pictures&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Uses artwork Steam has already downloaded (newer per-game folders in &lt;code&gt;appcache/librarycache/&amp;lt;id&amp;gt;/…&lt;/code&gt;, the older &lt;code&gt;&amp;lt;id&amp;gt;_library_600x900.jpg&lt;/code&gt; files, and custom art you set in Steam). Nothing is downloaded. Games without a picture get coloured initials.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Move&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Pick a drive and press &lt;strong&gt;Move&lt;/strong&gt;. The game folder, its &lt;strong&gt;Proton prefix (where Windows games keep saves)&lt;/strong&gt;, shader cache and workshop items are copied, every file is compared, Steam's library list is updated (a new &lt;code&gt;SteamLibrary&lt;/code&gt; folder is created and registered if the drive has none), and only then is the old copy deleted. On the same drive it is an instant rename.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Delete this copy&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Shown when a game is &lt;strong&gt;installed twice&lt;/strong&gt; (in two libraries), which blocks Move and Compress. Deletes only the copy on that card: Steam's record of it (&lt;code&gt;appmanifest&lt;/code&gt;, a backup is kept in &lt;code&gt;~/.config/deck-deck-goose/deleted-manifests/&lt;/code&gt;), its game folder, shader cache and workshop files in that library. Its &lt;strong&gt;Proton prefix (saves) is kept&lt;/strong&gt;, and the other copy is never touched. Steam must be closed (Desktop Mode). Refused if both copies point at the same folder.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Uninstall&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;For a game installed once: opens &lt;strong&gt;Steam's own uninstaller&lt;/strong&gt; (&lt;code&gt;steam://uninstall/&amp;lt;id&amp;gt;&lt;/code&gt;), which asks you to confirm. Deck Deck Goose deletes nothing itself here.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Clear junk&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Scans for shader caches of uninstalled games, old download leftovers, Steam's web cache and logs, unfinished copies, game folders and Proton prefixes left behind by uninstalls. Safe items are ticked. Items that may hold saves or mods are not, and are marked.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Estimate&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;On a &lt;strong&gt;btrfs&lt;/strong&gt; drive, a read-only check of how much compressing would save. It reads a few small samples of each file (64 MB at most per game, so even a big game takes seconds) and shows e.g. &lt;em&gt;save ~9 GB (19%) · worth it&lt;/em&gt; or &lt;em&gt;not worth it&lt;/em&gt;.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Compress&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Unlocks after Estimate, unless the verdict was &lt;em&gt;not worth it&lt;/em&gt;. Works with Steam open, as long as that game isn't running. Compresses &lt;strong&gt;only the files that will actually shrink&lt;/strong&gt; with zstd (&lt;code&gt;btrfs filesystem defragment -czstd&lt;/code&gt;), and sets the folder so future updates are compressed too. Not available on ext4, NTFS or exFAT, which have no such feature here.&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;strong&gt;What gets compressed (and what doesn't).&lt;/strong&gt; Videos (&lt;code&gt;.bk2&lt;/code&gt;, &lt;code&gt;.bik&lt;/code&gt;, &lt;code&gt;.mp4&lt;/code&gt;, &lt;code&gt;.webm&lt;/code&gt;...), audio (&lt;code&gt;.ogg&lt;/code&gt;, &lt;code&gt;.mp3&lt;/code&gt;...), images and archives are already compressed by their format, so they are skipped. Every other file is sampled; files that would shrink by less than 3% are skipped too. The big wins are usually uncompressed level and asset data. Program files (&lt;code&gt;.exe&lt;/code&gt;, &lt;code&gt;.dll&lt;/code&gt;, &lt;code&gt;.so&lt;/code&gt;) do get compressed when they shrink. They are small and only read at launch, and zstd unpacks far faster than a drive can read, so you are unlikely to notice. Parts the drive has already compressed (Bazzite and Fedora usually mount btrfs with &lt;code&gt;compress=zstd:1&lt;/code&gt;) are detected and not counted twice. Shader caches live outside the game folder and are rebuilt; clear them with &lt;strong&gt;Clear junk&lt;/strong&gt; instead.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Why no link or manifest editing?&lt;/strong&gt; Steam's &lt;code&gt;appmanifest_&amp;lt;id&amp;gt;.acf&lt;/code&gt; stores only the game's folder name, not a full path. Steam finds a game by looking for its manifest in each library's &lt;code&gt;steamapps&lt;/code&gt; folder. So "replanting" a game means putting the manifest and folders in the new library and making sure that library is in Steam's list, which this tool does. No symlink is needed. (Steam also has its own one-at-a-time mover under game Properties &amp;gt; Installed Files.)&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Rules it follows for safety&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Steam must be closed (there is a &lt;strong&gt;Close Steam&lt;/strong&gt; button). Steam rewrites these files while running.&lt;/li&gt;
&lt;li&gt;Only fully installed games move. Games needing an update or downloading are refused until Steam is done.&lt;/li&gt;
&lt;li&gt;Targets on &lt;strong&gt;exFAT/FAT&lt;/strong&gt; are refused: they cannot store the symlinks and Linux permissions Proton prefixes need. &lt;strong&gt;NTFS&lt;/strong&gt; targets are refused by default (Proton often fails there; the command line has &lt;code&gt;--allow-ntfs&lt;/code&gt;).&lt;/li&gt;
&lt;li&gt;It needs the game's size plus 1 GB free on the target.&lt;/li&gt;
&lt;li&gt;A journal records each move. If the power fails mid-move, the next start either finishes it or puts the game back.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Flatpak Steam&lt;/strong&gt; is given access to the new drive automatically.&lt;/li&gt;
&lt;li&gt;It refuses targets that are not a real, writable data drive: a RAM disk (e.g. a drive that is not mounted, so its folder is in memory), a read-only or network mount, or a folder whose parent doesn't exist.&lt;/li&gt;
&lt;li&gt;Games with an odd manifest (no usable folder name, or installed in two libraries at once) are left for Steam to sort out.&lt;/li&gt;
&lt;li&gt;Read-only game folders are copied with their permissions and still removed afterwards. A full disk, an unreadable file or a failed check undoes the copy and leaves the game untouched.&lt;/li&gt;
&lt;li&gt;Only one window (or command line) changes Steam files at a time.&lt;/li&gt;
&lt;li&gt;A disk mounted under &lt;code&gt;/run/media/...&lt;/code&gt; is only there while it is mounted. Set it to mount at startup (e.g. Disks app &amp;gt; Edit Mount Options), or Steam will show the game as not installed.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  Command line
&lt;/h3&gt;

&lt;p&gt;Run from the &lt;code&gt;src&lt;/code&gt; folder (or set &lt;code&gt;PYTHONPATH=src&lt;/code&gt;):&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;python3 -m ddg games                 # list games and ids
python3 -m ddg targets &amp;lt;appid&amp;gt;       # where that game can go
python3 -m ddg move-game &amp;lt;appid&amp;gt; /var/mnt/games/SteamLibrary
python3 -m ddg junk                  # list junk; add --clean to remove the ticked items
python3 -m ddg estimate-game &amp;lt;appid&amp;gt;  # read-only: how much would compressing save?
python3 -m ddg compress-game &amp;lt;appid&amp;gt;  # compresses only files that shrink
python3 -m ddg close-steam
python3 -m ddg install|check|refresh|uninstall|disks|--version
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  What exists and what doesn't
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Exists&lt;/th&gt;
&lt;th&gt;Does not exist yet&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;This PC folder, sidebar entries, Flatpak access&lt;/td&gt;
&lt;td&gt;Instant refresh when a disk is plugged in (up to 30 s)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Steam game list with local artwork, native and Flatpak Steam&lt;/td&gt;
&lt;td&gt;Heroic, Lutris, GOG or Epic games&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Move with copy check, rollback, crash recovery, new library registration&lt;/td&gt;
&lt;td&gt;Moving several games in one go (do them one after another)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Junk scan and clean&lt;/td&gt;
&lt;td&gt;Measuring compression precisely afterwards (needs &lt;code&gt;compsize&lt;/code&gt; and root; we show the drive's free-space change)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Estimate, then btrfs zstd compression of only the files that shrink&lt;/td&gt;
&lt;td&gt;Compression on ext4/NTFS (not supported by those file systems here)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;/td&gt;
&lt;td&gt;"Estimate all games" in one go&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Sound fix, boost to 150%, output choice&lt;/td&gt;
&lt;td&gt;A real loudness/EQ preset editor (Bazzite's own EQ is kept as it is)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Airlock code, fallback ladder, colour bar (simulated tests)&lt;/td&gt;
&lt;td&gt;Airlock proven on a real handheld; a shipped Steam Input layout for it&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  Testing honesty
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Windows and Mac:&lt;/strong&gt; 70 tests drive the same engine against simulated Windows and macOS set-ups, and check the window starts without any Linux-only Python modules. A real Windows PC, a real Mac, real NTFS compression and a real Steam client there have &lt;strong&gt;not&lt;/strong&gt; been tried.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Wake / Zombie-Deck:&lt;/strong&gt; its 35 own tests (fake power files, systemd unit check, overlay) pass here, and 20 more cover the Wake tab and installer wiring using a stand-in installer. The sleep-hook unit is checked as text, not by a real suspend.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;200+ automated tests pass&lt;/strong&gt; (as root; the 2 permission tests that root skips also pass as a normal user), but they run on &lt;strong&gt;simulated data&lt;/strong&gt;: a pretend Steam install with invented games and artwork (&lt;code&gt;tests/fake_steam.py&lt;/code&gt;), simulated disks, and a stand-in for &lt;code&gt;flatpak&lt;/code&gt;, &lt;code&gt;systemctl&lt;/code&gt;, &lt;code&gt;udisksctl&lt;/code&gt;, &lt;code&gt;steam&lt;/code&gt; and &lt;code&gt;btrfs&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;A move to a &lt;strong&gt;genuinely different file system&lt;/strong&gt; (a tmpfs) was run for real via the command line: files, permissions, symlinks and a Proton save arrived, the library list was updated, and the moved script ran. Without the test override, the same tmpfs target is refused as "not a real data drive", as intended.&lt;/li&gt;
&lt;li&gt;Edge cases covered by tests: unsafe or empty folder names in manifests (cannot reach outside Steam's folders), duplicate installs, Unicode names, RAM-disk, read-only and missing targets, read-only game folders, a full disk mid-copy, unreadable files, a cross-subvolume rename falling back to copying, broken or missing library lists, junk that is a symlink to your own files, and two windows at once.&lt;/li&gt;
&lt;li&gt;The estimate uses zlib as a stand-in for zstd on Python before 3.14. Measured on sample data, it comes out equal to or slightly below what zstd achieves, so it errs on the low side. Detecting already-compressed parts uses the Linux FIEMAP call; it was checked on ext4 (reports nothing compressed) but not on a real btrfs drive.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Game Mode tile:&lt;/strong&gt; the launcher was run for real here, end to end, with a small script standing in for the browser (it loaded the page in Game Mode layout and pressed Back to Steam, which closed everything). Adding the shortcut was tested against Steam's binary shortcut file format as documented, &lt;strong&gt;not with a real Steam client or a real Game Mode session&lt;/strong&gt;. Whether Firefox's kiosk mode and the touchscreen behave well under Game Mode needs your Ally.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Airlock, first real run (ROG Ally, Bazzite, 2026-11-07):&lt;/strong&gt; it opened a desktop from Game Mode, but the full desktop did not count as started, the simple one (Dolphin) opened on top of it, and the full one could not be closed without logging out. 0.6.1 closes such leftovers. &lt;strong&gt;Second real run (2026-11-09):&lt;/strong&gt; works as expected and quick enough for general management, but around 20 fps with dropped frames, tiny icons (needed 150-200% scaling) and the Start-menu (Kickoff) popup stayed open until its button was pressed again. 0.6.4 adds the Wayland connection and the 720p default; the Start-menu issue is not fixed yet (see Known issues).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Airlock:&lt;/strong&gt; tested with a stand-in &lt;code&gt;kwin_wayland&lt;/code&gt; script and stand-in apps: the full ladder (opens and closes cleanly, compositor dying at once, hung start killed by the watchdog, second Airlock refused, switch asks first and &lt;em&gt;no&lt;/em&gt; means no, Back to Game closes it), the hooks, the lock, the sleep hook install/remove, the web API and the colour bar. &lt;strong&gt;Never run with a real KWin, gamescope or Game Mode session.&lt;/strong&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Sound:&lt;/strong&gt; tested against a stand-in for &lt;code&gt;pactl&lt;/code&gt; that copies the ROG Ally layout described by the Bazzite wiki (an EQ output over the Family speakers, plus HDMI, a USB headset and Bluetooth), and a sample &lt;code&gt;pw-dump&lt;/code&gt;. &lt;strong&gt;Not run on a real Ally&lt;/strong&gt;: the real output names may differ, and Steam in Game Mode may pull a boosted volume back to 100% when you move its slider.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;On-screen keyboard:&lt;/strong&gt; tested against example &lt;code&gt;kwinrc&lt;/code&gt; files and a simulated package manager only; not on a real KDE session, and the real &lt;code&gt;pkexec dnf install&lt;/code&gt; has not been run.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Not yet tested&lt;/strong&gt;: against a real Steam client (does it pick up the moved game without fuss?), real Flatpak Steam, real btrfs compression (no btrfs was available to the build machine; the btrfs commands are standard ones, see the research notes), Dolphin and Nemo sidebar entries, and any real Bazzite, Nobara or Mint machine. The GitHub workflows have not been run.&lt;/li&gt;
&lt;li&gt;The screenshots show invented games with generated covers and simulated sizes and estimates.&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;&lt;strong&gt;Source code:&lt;/strong&gt; &lt;a href="https://github.com/The-Dorkknight/deck-deck-goose" rel="noopener noreferrer"&gt;The-Dorkknight/deck-deck-goose on GitHub&lt;/a&gt;&lt;/p&gt;

</description>
    </item>
    <item>
      <title>Twoo STEP CAD Toolbox</title>
      <dc:creator>Matthew Armstrong</dc:creator>
      <pubDate>Wed, 04 Nov 2026 10:57:46 +0000</pubDate>
      <link>https://dev.to/open-madscience/twoo-step-cad-toolbox-4ilf</link>
      <guid>https://dev.to/open-madscience/twoo-step-cad-toolbox-4ilf</guid>
      <description>&lt;p&gt;&lt;em&gt;(The name is a play on **to&lt;/em&gt;&lt;em&gt;, **two&lt;/em&gt;* and &lt;strong&gt;too&lt;/strong&gt;; it stands for the all-in-one approach.)*&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Possibly the most advanced and functional free (for non-commercial use) solid conversion software in the world:&lt;/strong&gt; STL meshes, hardpoint STL files, OpenSCAD models and photos of flat parts to real, editable STEP solids, offline, on Linux, macOS and Windows.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Source-available · Non-commercial&lt;/em&gt; — licensed under the &lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/LICENSE.md" rel="noopener noreferrer"&gt;PolyForm Noncommercial License 1.0.0&lt;/a&gt;.&lt;br&gt;
Provided &lt;strong&gt;as is, with no warranty and no liability&lt;/strong&gt;. Check converted files before relying on them.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;This is the long-term-support successor of STL2STEP / EVERYTHING to solid.&lt;/strong&gt;&lt;br&gt;
That app (last version 0.10.3, which is 0.10.2 plus the graphics-card fix) is frozen as a finished, dead-end project and stays online for anyone who wants to read it:&lt;br&gt;
&lt;strong&gt;&lt;a href="https://github.com/The-Dorkknight/stl2step" rel="noopener noreferrer"&gt;https://github.com/The-Dorkknight/stl2step&lt;/a&gt;&lt;/strong&gt;. This repository carries all of its history (the full changelog,&lt;br&gt;
research notes and logs are in &lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/tree/main/docs/history" rel="noopener noreferrer"&gt;&lt;code&gt;docs/history&lt;/code&gt;&lt;/a&gt; and &lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/tree/main/docs/research" rel="noopener noreferrer"&gt;&lt;code&gt;docs/research&lt;/code&gt;&lt;/a&gt;) and what comes next.&lt;br&gt;
The story of the split is in &lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/LEGACY.md" rel="noopener noreferrer"&gt;LEGACY.md&lt;/a&gt;; what "long-term support" does and does not promise is in&lt;br&gt;
&lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/LTS_POLICY.md" rel="noopener noreferrer"&gt;LTS_POLICY.md&lt;/a&gt;.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fzsc6ob56bv7tnkfh6913.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fzsc6ob56bv7tnkfh6913.webp" alt="Preview of a converted part" width="800" height="610"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;One portable folder that turns things into &lt;strong&gt;real STEP solids&lt;/strong&gt; (true planes, cylinders, spheres and fitted&lt;br&gt;
NURBS, not thousands of triangles). Every tool in it is a &lt;strong&gt;plug-in&lt;/strong&gt; with a small manifest; the web page, the&lt;br&gt;
command line, a FreeCAD workbench and a Fusion add-in are all built on the same five.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Plug-in&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;th&gt;Status&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Mesh → STEP&lt;/strong&gt; (&lt;code&gt;mesh2step&lt;/code&gt;)&lt;/td&gt;
&lt;td&gt;STL / OBJ / PLY / 3MF mesh → STEP solid, with four presets (Draft, Quick, Standard, Fine) and &lt;em&gt;Advanced…&lt;/em&gt; for every setting. Also &lt;em&gt;Extrude outline&lt;/em&gt; and &lt;em&gt;Loft&lt;/em&gt; for flat-profile and tapered parts.&lt;/td&gt;
&lt;td&gt;stable core, experimental engines&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Hardpoint STL&lt;/strong&gt; (&lt;code&gt;hardpoint&lt;/code&gt;)&lt;/td&gt;
&lt;td&gt;Reads an STL that also remembers which CAD face each triangle came from and rebuilds the solid from those faces instead of guessing; or turns a plain STL into one.&lt;/td&gt;
&lt;td&gt;stable&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;OpenSCAD → STEP&lt;/strong&gt; (&lt;code&gt;scad2step&lt;/code&gt;)&lt;/td&gt;
&lt;td&gt;Rebuilds an OpenSCAD model as real CAD geometry (true cylinders, spheres, swept surfaces, exact booleans). Customizer fields, preview, live link.&lt;/td&gt;
&lt;td&gt;stable&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Jetwash&lt;/strong&gt; (&lt;code&gt;jetwash&lt;/code&gt;)&lt;/td&gt;
&lt;td&gt;Cleans slivers, 0.0001 mm steps, thin fins, specks and tiny holes out of an existing STEP file; every removal is checked and rolled back if it would damage the part.&lt;/td&gt;
&lt;td&gt;stable&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Image → STEP&lt;/strong&gt; (&lt;code&gt;image2step&lt;/code&gt;)&lt;/td&gt;
&lt;td&gt;Traces a photo, scan or drawing of a flat part into CAD curves and saves a STEP: a solid of the thickness you type, or the flat shape.&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;experimental&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;It runs in your web browser from one portable folder; nothing is installed on your system. Mac, Windows and Linux.&lt;/p&gt;

&lt;h3&gt;
  
  
  What the different ways of converting look like
&lt;/h3&gt;

&lt;p&gt;Pick a &lt;strong&gt;quality preset&lt;/strong&gt; (or &lt;em&gt;Advanced…&lt;/em&gt; for every setting). Each button shows how hard it works your computer (red)&lt;br&gt;
and how good the result is (green):&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ft2gzz7efnuabmxdsjika.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ft2gzz7efnuabmxdsjika.jpg" alt="The four quality presets as shown in the app: Draft, Quick, Standard, Fine, and Advanced" width="800" height="854"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The same three meshes converted the normal way (Classic, the original converter) and with &lt;strong&gt;roughing first&lt;/strong&gt;&lt;br&gt;
(the Staged way, which the Quick and Standard presets now use for turned and flat-sided parts): far fewer, cleaner&lt;br&gt;
faces, and faster.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F8ld6ovb9qbtjo82q3nni.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F8ld6ovb9qbtjo82q3nni.webp" alt="Three parts as input mesh, Classic result and Staged result: vase 8,665 faces against 14, torus 605 against 9, tray 1,927 against 131" width="799" height="722"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;An organic test model with &lt;em&gt;Guesstimate&lt;/em&gt; on: flat faces and holes stay exact, freeform areas are allowed to sit a&lt;br&gt;
little further from the mesh, so it finishes in minutes instead of grinding for hours.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fwttp9w78bksku3bz9pzw.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fwttp9w78bksku3bz9pzw.webp" alt="The ghost test model: input mesh, Classic with Guesstimate, Staged with Guesstimate" width="799" height="286"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;More pictures, and every measurement behind them, are in the research paper&lt;br&gt;
&lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/docs/research/SOLID_EVIDENCE.md" rel="noopener noreferrer"&gt;SOLID evidence&lt;/a&gt; (&lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/docs/research/SOLID_EVIDENCE.pdf" rel="noopener noreferrer"&gt;PDF&lt;/a&gt;).&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Reporting a problem?&lt;/strong&gt; The 3D preview carries a &lt;strong&gt;Diagnostic&lt;/strong&gt; panel: the job's settings and result as a QR code and&lt;br&gt;
in plain words. One screenshot of it says it all (&lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/docs/guide/diagnostic-code.md" rel="noopener noreferrer"&gt;how it works&lt;/a&gt;).&lt;/p&gt;

&lt;h2&gt;
  
  
  Command line (optional)
&lt;/h2&gt;

&lt;p&gt;Pass arguments to the start file, or run &lt;code&gt;python run_tostep.py …&lt;/code&gt; with the private Python:&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;./twoo-step-cad-toolbox-linux.sh part.stl                         # → part.step
./twoo-step-cad-toolbox-linux.sh convert part.stl out.step --preset quick
./twoo-step-cad-toolbox-linux.sh wash old.step --threshold 0.05
./twoo-step-cad-toolbox-linux.sh scad model.scad -D width=40
./twoo-step-cad-toolbox-linux.sh image photo.jpg --part-above-token 10 --thickness 10
./twoo-step-cad-toolbox-linux.sh hardpoints part.stl
./twoo-step-cad-toolbox-linux.sh diag screenshot.png
./twoo-step-cad-toolbox-linux.sh plugins                          # what is installed, and anything broken
./twoo-step-cad-toolbox-linux.sh solidcraft                       # selected mesh bodies in SolidCraft -&amp;gt; solids
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;

&lt;p&gt;Which commands exist comes from the plug-ins. The classic converter stays the default on the command line and in the CAD plug-ins; new engines are opt-in.&lt;br&gt;
More: &lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/docs/guide/using.md" rel="noopener noreferrer"&gt;docs/guide/using.md&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  FreeCAD and Fusion
&lt;/h2&gt;

&lt;p&gt;Open the &lt;strong&gt;Install&lt;/strong&gt; tab and click &lt;em&gt;Install FreeCAD plug-in&lt;/em&gt; or &lt;em&gt;Install Fusion add-in&lt;/em&gt; (or run &lt;code&gt;install freecad&lt;/code&gt; / &lt;code&gt;install fusion&lt;/code&gt;). Both are thin front ends:&lt;br&gt;
they send the mesh to this folder, it converts in the background and the solid comes back into your model. Start the toolbox once first so its setup is done.&lt;br&gt;
Details and manual install: &lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/docs/guide/cad-plugins.md" rel="noopener noreferrer"&gt;docs/guide/cad-plugins.md&lt;/a&gt;. &lt;strong&gt;Neither has been run in a real FreeCAD or Fusion yet&lt;/strong&gt; (see below).&lt;/p&gt;

&lt;h2&gt;
  
  
  SolidCraft (experimental)
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://github.com/tangency-labs/solidcraft" rel="noopener noreferrer"&gt;SolidCraft&lt;/a&gt;, the new open-source Fusion-style CAD app, has no plug-in&lt;br&gt;
system yet, so the toolbox links to it instead: select a mesh body in SolidCraft, click &lt;strong&gt;or take the selected mesh&lt;br&gt;
body from SolidCraft&lt;/strong&gt; in the Mesh tab, convert, and &lt;strong&gt;Send to SolidCraft&lt;/strong&gt; puts the solid back as a new component.&lt;br&gt;
Command line: &lt;code&gt;tostep solidcraft&lt;/code&gt;. Turn on &lt;em&gt;Preferences › Assistant › Agents through MCP › Let AI agents connect&lt;/em&gt; in&lt;br&gt;
SolidCraft first. Details: &lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/docs/guide/solidcraft.md" rel="noopener noreferrer"&gt;docs/guide/solidcraft.md&lt;/a&gt;. &lt;strong&gt;Not yet run against a real&lt;br&gt;
SolidCraft&lt;/strong&gt;, only against a stand-in that follows its documented protocol.&lt;/p&gt;

&lt;h2&gt;
  
  
  What to read next
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/tree/main/docs/guide" rel="noopener noreferrer"&gt;docs/guide&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;One page per feature: hardpoints, OpenSCAD, images, engines, performance, SolidCraft, troubleshooting&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/HOW_IT_WORKS.md" rel="noopener noreferrer"&gt;HOW_IT_WORKS.md&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;The pipeline, step by step&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/ARCHITECTURE.md" rel="noopener noreferrer"&gt;ARCHITECTURE.md&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/docs/PLUGIN_CONTRACT.md" rel="noopener noreferrer"&gt;docs/PLUGIN_CONTRACT.md&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/docs/ADDING_A_PLUGIN.md" rel="noopener noreferrer"&gt;docs/ADDING_A_PLUGIN.md&lt;/a&gt;
&lt;/td&gt;
&lt;td&gt;How it is put together, and how to add a tool&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/MAINTAINING.md" rel="noopener noreferrer"&gt;MAINTAINING.md&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/docs/DECISIONS.md" rel="noopener noreferrer"&gt;docs/DECISIONS.md&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/LTS_POLICY.md" rel="noopener noreferrer"&gt;LTS_POLICY.md&lt;/a&gt;
&lt;/td&gt;
&lt;td&gt;For whoever looks after it next&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/docs/KNOWN_LIMITATIONS.md" rel="noopener noreferrer"&gt;docs/KNOWN_LIMITATIONS.md&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;What has and has not been tested, honestly&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/tree/main/docs/research" rel="noopener noreferrer"&gt;docs/research&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/docs/research/SOLID_EVIDENCE.md" rel="noopener noreferrer"&gt;SOLID_EVIDENCE&lt;/a&gt;
&lt;/td&gt;
&lt;td&gt;The research notes and the defensive publication of every method&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/CHANGELOG.md" rel="noopener noreferrer"&gt;CHANGELOG.md&lt;/a&gt; · &lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/tree/main/docs/history" rel="noopener noreferrer"&gt;docs/history&lt;/a&gt;
&lt;/td&gt;
&lt;td&gt;What changed, including the whole STL2STEP history&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/docs/MIGRATING_FROM_STL2STEP.md" rel="noopener noreferrer"&gt;docs/MIGRATING_FROM_STL2STEP.md&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;New names for old commands, settings and environment variables&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  Honest status
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Work in progress.&lt;/strong&gt; Expect rough edges. Version 0.11.0-wip moved the whole 0.10.2 app into a plug-in structure &lt;em&gt;without changing what the converters do&lt;/em&gt;: on the test knob the new layout writes a STEP file of the same size and face count as 0.10.2, and the old test-suite ports pass. The release gate (&lt;code&gt;tools/golden.py&lt;/code&gt;) is described in &lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/benchmarks/README.md" rel="noopener noreferrer"&gt;benchmarks/README.md&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Linux is the only tested platform.&lt;/strong&gt; The Windows and macOS start files and the FreeCAD and Fusion plug-ins have &lt;strong&gt;not&lt;/strong&gt; been run on real Windows, macOS, FreeCAD or Fusion. Version 1.0 is gated on that: &lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything/blob/main/LTS_POLICY.md" rel="noopener noreferrer"&gt;LTS_POLICY.md&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;A plain 96-sided cylinder converted with the default Classic engine stays 98 flat strips; Standard and Quick make it 3 true faces. Known, measured and locked in the release gate so it can only get better.&lt;/li&gt;
&lt;li&gt;The result is a solid of surfaces, not an editable feature history (no sketches / extrudes).&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;&lt;strong&gt;Source code:&lt;/strong&gt; &lt;a href="https://github.com/The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything" rel="noopener noreferrer"&gt;The-Dorkknight/Twoo-STEP-CAD-Toolbox-0.11.1-wip-everything on GitHub&lt;/a&gt;&lt;/p&gt;

</description>
      <category>cad</category>
      <category>3dprinting</category>
      <category>stltostepconversion</category>
      <category>phototocadsketch</category>
    </item>
  </channel>
</rss>
