<?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: Lucas</title>
    <description>The latest articles on DEV Community by Lucas (@_9848c5582063b42abecb7).</description>
    <link>https://dev.to/_9848c5582063b42abecb7</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%2F3915788%2F4474d8b0-095a-44ee-bc5c-0166ad76d683.png</url>
      <title>DEV Community: Lucas</title>
      <link>https://dev.to/_9848c5582063b42abecb7</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/_9848c5582063b42abecb7"/>
    <language>en</language>
    <item>
      <title>Mass Effect: Make the Consequence Point Back to the Choice</title>
      <dc:creator>Lucas</dc:creator>
      <pubDate>Mon, 16 Nov 2026 23:23:51 +0000</pubDate>
      <link>https://dev.to/_9848c5582063b42abecb7/mass-effect-make-the-consequence-point-back-to-the-choice-24pa</link>
      <guid>https://dev.to/_9848c5582063b42abecb7/mass-effect-make-the-consequence-point-back-to-the-choice-24pa</guid>
      <description>&lt;p&gt;A game can remember a decision perfectly and still leave the player wondering why the story changed. Saving a flag is one job. Showing the player what that flag meant is another.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Mass Effect&lt;/em&gt; gives us two ways to think about this. In the &lt;a href="https://www.ea.com/ea-play/news/gameplay-calibrations" rel="noopener noreferrer"&gt;Legendary Edition gameplay notes&lt;/a&gt;, the team explains that choices across the trilogy help determine who fights alongside Shepard in the third game. &lt;a href="https://blog.bioware.com/2012/06/22/mass-effect-3-extended-cut-2/" rel="noopener noreferrer"&gt;BioWare’s original Extended Cut notes&lt;/a&gt; describe a different job for the epilogue: adding scenes to explain what the decisions meant for the people and places involved, without fundamentally replacing the endings.&lt;/p&gt;

&lt;p&gt;One changes what’s there. The other makes the change easier to interpret. Those jobs don’t have to happen in the same scene.&lt;/p&gt;

&lt;h2&gt;
  
  
  Give the callback a concrete detail
&lt;/h2&gt;

&lt;p&gt;Imagine a station crew offers you its last working tug. You can send it to recover a stranded shuttle or keep it to clear the station’s docking lane. Show the immediate result: the shuttle gets home, or the dock reopens. Much later, the shuttle pilot can show up to help, or a cargo captain can thank you for making the dock usable. Let the returning character mention the tug once, in a way that fits the scene. That gives the player a detail to recognize instead of a banner telling them their choices mattered.&lt;/p&gt;

&lt;h2&gt;
  
  
  Test whether players make the connection
&lt;/h2&gt;

&lt;p&gt;Try the same follow-up scene with and without that one detail. Afterward, ask players what changed and why they think it changed. If they notice the new arrival but attribute it to the main plot, the branch works in code but isn’t yet doing the storytelling job. If the payoff is meant to be a mystery, don’t give the cause away early. Put the callback in the reveal itself.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;AI disclosure: This article was researched and drafted with AI assistance using the sources linked above.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>gamedev</category>
    </item>
    <item>
      <title>Elite Dangerous: Show When a Jump Can Still Be Cancelled</title>
      <dc:creator>Lucas</dc:creator>
      <pubDate>Sun, 15 Nov 2026 23:52:23 +0000</pubDate>
      <link>https://dev.to/_9848c5582063b42abecb7/elite-dangerous-show-when-a-jump-can-still-be-cancelled-1lp2</link>
      <guid>https://dev.to/_9848c5582063b42abecb7/elite-dangerous-show-when-a-jump-can-still-be-cancelled-1lp2</guid>
      <description>&lt;p&gt;AI assisted with source research and drafting. The prototype below is a design proposal, not a completed playtest or a claim of personal experience with Elite Dangerous.&lt;/p&gt;

&lt;p&gt;A spaceship's jump command has more than one useful response. The cockpit can show that charging has started, how far it has progressed, and whether the pilot can still change their mind.&lt;/p&gt;

&lt;p&gt;Those messages answer different questions. A progress bar can keep moving while leaving the last question unanswered.&lt;/p&gt;

&lt;h2&gt;
  
  
  A documented boundary in the jump sequence
&lt;/h2&gt;

&lt;p&gt;Frontier's &lt;a href="https://hosting.zaonce.net/elite/website/assets/ELITE-DANGEROUS-GAME-MANUAL.pdf" rel="noopener noreferrer"&gt;Elite Dangerous Players Guide v1.05&lt;/a&gt;, pages 74–75, describes a hyperspace jump sequence with a HUD charge bar. It also explains that the pilot can cancel before the five-second countdown begins by using the frame shift drive command again. Once that countdown starts, cancellation is no longer available.&lt;/p&gt;

&lt;p&gt;This is a reading of the older guide's documented sequence, not a claim that its timings describe every current ship or drive.&lt;/p&gt;

&lt;p&gt;The design interpretation is that the display has two jobs: communicate progress and communicate the actions still available. A pilot waiting for a drive to charge and a pilot committed to a jump may both be waiting, but they have different control over what happens next.&lt;/p&gt;

&lt;p&gt;The guide establishes the rule. It does not establish how well players understand it. That is the question worth testing.&lt;/p&gt;

&lt;h2&gt;
  
  
  Give the waiting state a clear meaning
&lt;/h2&gt;

&lt;p&gt;Consider a fictional shuttle with a departure button. Pressing it starts a short preparation phase. During preparation, the player can cancel. After preparation, the shuttle enters a final countdown and departs automatically.&lt;/p&gt;

&lt;p&gt;Use three explicit states in a paper prototype or a small interactive mockup:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Preparing: show progress and an available Cancel control.&lt;/li&gt;
&lt;li&gt;Departure committed: replace the cancel affordance with a clear message that the countdown cannot be stopped.&lt;/li&gt;
&lt;li&gt;Departed: show the changed location rather than leaving the completed progress bar on screen.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;These are proposed interface choices, not a description of Elite Dangerous's exact cockpit presentation.&lt;/p&gt;

&lt;p&gt;If the player cancels during preparation, return to a clearly idle state. If they try after commitment, explain why the shuttle continues. An input animation alone cannot tell them whether cancellation succeeded.&lt;/p&gt;

&lt;p&gt;Keep the commitment message visible long enough to read. A brief flash or a change from one color to another may look dramatic while leaving the rule ambiguous.&lt;/p&gt;

&lt;h2&gt;
  
  
  Test comprehension at the boundary
&lt;/h2&gt;

&lt;p&gt;Compare two versions with identical timings and cancellation rules. One shows a progress bar and the final countdown. The other adds the explicit preparation, commitment and cancellation messages.&lt;/p&gt;

&lt;p&gt;Ask players to stop the departure at different moments. Include an early attempt and an attempt after commitment. Record whether they think the cancellation worked, whether the shuttle actually stops, and whether they repeatedly press a control while trying to understand its state.&lt;/p&gt;

&lt;p&gt;Then ask a plain question: “Can you still stop the shuttle now?”&lt;/p&gt;

&lt;p&gt;The result to look for is agreement between the player's answer and the actual rule. Faster button pressing alone would not show that the interface became clearer.&lt;/p&gt;

&lt;p&gt;Repeat the task with sound muted. Important control changes should remain understandable when a spoken announcement is unavailable. If the text version becomes cluttered, simplify the wording before adding another animation.&lt;/p&gt;

&lt;h2&gt;
  
  
  Decide whether commitment belongs here
&lt;/h2&gt;

&lt;p&gt;The test assumes that a non-cancellable phase serves the game. That assumption needs its own review. For a frequently repeated menu action, an irreversible countdown might merely waste time. A short departure sequence might need no cancellation boundary at all.&lt;/p&gt;

&lt;p&gt;Even if the second display improves understanding, it would not prove that the countdown is enjoyable or that a spaceship needs more cockpit indicators.&lt;/p&gt;

&lt;p&gt;The useful lesson from the guide is specific: when an action changes from something the player can stop into something they must wait through, make that change legible. The machinery can feel elaborate while the answer to “Did it accept my input?” stays simple.&lt;/p&gt;

</description>
      <category>gamedev</category>
    </item>
    <item>
      <title>Creeper World 4: Let the Player See Territory Becoming Unsafe</title>
      <dc:creator>Lucas</dc:creator>
      <pubDate>Sat, 14 Nov 2026 23:54:57 +0000</pubDate>
      <link>https://dev.to/_9848c5582063b42abecb7/creeper-world-4-let-the-player-see-territory-becoming-unsafe-3jog</link>
      <guid>https://dev.to/_9848c5582063b42abecb7/creeper-world-4-let-the-player-see-territory-becoming-unsafe-3jog</guid>
      <description>&lt;p&gt;AI assisted with source research and drafting. The experiment below is a design proposal, not a completed playtest or a claim of personal experience with Creeper World 4.&lt;/p&gt;

&lt;p&gt;A territory map can tell a player who controls each area. A moving threat can also show how that control is changing. The difference matters when the player's next decision is where to defend.&lt;/p&gt;

&lt;p&gt;Creeper World 4 offers an SF example. &lt;a href="https://www.knucklecracker.com/games/creeper-world-4/" rel="noopener noreferrer"&gt;Knuckle Cracker's official overview&lt;/a&gt; describes a fluid enemy flowing across a three-dimensional landscape. The player builds an economy, establishes defenses and pushes outward to reclaim ground. The studio's &lt;a href="https://knucklecracker.com/creeperworld4/cw4.php" rel="noopener noreferrer"&gt;original game page&lt;/a&gt; identifies terrain and the flowing enemy as central features.&lt;/p&gt;

&lt;p&gt;Those descriptions support a design question: can the changing shape of a threat help players choose where to act before an area is lost?&lt;/p&gt;

&lt;h2&gt;
  
  
  Make the front worth watching
&lt;/h2&gt;

&lt;p&gt;The interpretation here is that territory can communicate more than a final ownership state. An advancing edge gives the player something to compare with nearby defenses, routes and exposed space.&lt;/p&gt;

&lt;p&gt;This does not mean continuous motion is always clearer. A screen full of movement can obscure the decision. The useful signal is the relationship between what is spreading and what the player can still protect.&lt;/p&gt;

&lt;p&gt;For a small prototype, that relationship can be much simpler than a fluid simulation.&lt;/p&gt;

&lt;h2&gt;
  
  
  Test two ways to show the same advance
&lt;/h2&gt;

&lt;p&gt;Imagine a fictional moon outpost with two approaches and one movable defense unit. A hostile substance advances along both approaches according to fixed rules. The player must choose where to move the unit to protect the outpost.&lt;/p&gt;

&lt;p&gt;Use a small grid. Each cell records how close it is to becoming unsafe. When a cell crosses a defined threshold, its ownership changes. These are invented rules for the experiment, not an account of Creeper World 4's implementation.&lt;/p&gt;

&lt;p&gt;Compare two presentations:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Version A changes a cell's appearance only when ownership changes.&lt;/li&gt;
&lt;li&gt;Version B also shows the approach toward that threshold, so the player can watch the threatened area develop before it changes hands.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Keep the advance schedule, defense strength, movement speed, camera and starting positions identical. Both versions should clearly show current ownership. Use outlines or patterns alongside color so the test does not depend on distinguishing two hues.&lt;/p&gt;

&lt;p&gt;Before the first move, ask the player which approach is likely to become dangerous next and what they would protect. Then let the simulation run. Record the prediction, the first placement and any later correction.&lt;/p&gt;

&lt;p&gt;The useful observation is whether players can connect the visible advance to their choice. Winning once is weak evidence: the player might have guessed correctly. Ask what changed on the map that made them move the defense.&lt;/p&gt;

&lt;h2&gt;
  
  
  Separate warning from visual noise
&lt;/h2&gt;

&lt;p&gt;Version B may help players anticipate danger. It may also produce constant flickering that feels urgent everywhere. If players cannot tell which change matters, adding more animated detail has failed the test.&lt;/p&gt;

&lt;p&gt;A practical revision is to simplify the boundary or reduce the number of intermediate states. Another is to slow the advance while keeping the same relationship between both approaches. Test those changes separately so the cause of any improvement remains understandable.&lt;/p&gt;

&lt;p&gt;Watch for an opposite problem too: if the threatened route is always obvious and there is only one sensible placement, the display may be clear while the decision remains dull. Readability cannot create a meaningful choice by itself.&lt;/p&gt;

&lt;p&gt;Creeper World 4's documented premise makes an unusual enemy visible across the landscape. For a smaller SF prototype, the transferable question is whether players can read territory in the process of changing, then make a decision they can explain. Establish that relationship before adding more weapons, regions or visual spectacle.&lt;/p&gt;

</description>
      <category>gamedev</category>
    </item>
    <item>
      <title>Factorio: Let a Full Conveyor Explain the Next Upgrade</title>
      <dc:creator>Lucas</dc:creator>
      <pubDate>Fri, 13 Nov 2026 23:12:54 +0000</pubDate>
      <link>https://dev.to/_9848c5582063b42abecb7/factorio-let-a-full-conveyor-explain-the-next-upgrade-2igb</link>
      <guid>https://dev.to/_9848c5582063b42abecb7/factorio-let-a-full-conveyor-explain-the-next-upgrade-2igb</guid>
      <description>&lt;p&gt;AI assisted with source research and drafting. The prototype below is a design proposal, not a completed playtest or a claim of personal experience with Factorio.&lt;/p&gt;

&lt;p&gt;A production game can show a rising resource total while hiding almost everything that caused it. Another approach is to let the player watch material arrive, wait, move and become something else. Then an upgrade can answer a problem visible in the world.&lt;/p&gt;

&lt;p&gt;Factorio is a useful SF case for this question. Its &lt;a href="https://www.factorio.com/" rel="noopener noreferrer"&gt;official description&lt;/a&gt; places mining, infrastructure, research and automated production at the center of building and maintaining factories.&lt;/p&gt;

&lt;p&gt;The interesting design question is what the moving material communicates before the player opens a statistics panel.&lt;/p&gt;

&lt;h2&gt;
  
  
  Movement carries information
&lt;/h2&gt;

&lt;p&gt;In the January 2018 developer discussion &lt;a href="https://www.factorio.com/blog/post/fff-225" rel="noopener noreferrer"&gt;Bots versus belts, part 2&lt;/a&gt;, kovarex describes the appeal of seeing many filled belts operating, rather than reducing a large factory to its output number. That is a developer's stated preference in a historical design debate, not evidence that all players prefer belts or that one transport system is best today.&lt;/p&gt;

&lt;p&gt;A later development post exposes a practical constraint. In January 2024, &lt;a href="https://www.factorio.com/blog/post/fff-393" rel="noopener noreferrer"&gt;Friday Facts #393&lt;/a&gt; explained that faster belts could become difficult to read: their motion might look confusing or even reversed. The team described changes to arrow spacing and item colors to help communicate movement.&lt;/p&gt;

&lt;p&gt;The interpretation here is that visible production needs legible states. More moving objects alone do not tell a player whether a line is healthy. A full conveyor might be carrying material efficiently, or waiting for something downstream. Those situations should lead to different decisions.&lt;/p&gt;

&lt;h2&gt;
  
  
  Build one small production line
&lt;/h2&gt;

&lt;p&gt;Consider a fictional orbital workshop with three steps: a machine produces metal blanks, a conveyor carries them, and a fabricator consumes them. Give each step a small, visible buffer. Stop upstream production when its output buffer is full.&lt;/p&gt;

&lt;p&gt;These are invented prototype rules, not a reconstruction of Factorio's implementation.&lt;/p&gt;

&lt;p&gt;Start with a slow fabricator. Blanks accumulate before it, and eventually the producer has to stop. Offer two equally priced upgrades: a faster producer or a faster fabricator. Keep power, recipes and transport capacity fixed so they do not introduce other causes.&lt;/p&gt;

&lt;p&gt;Before the player spends anything, ask which upgrade they expect to increase completed output and why. Then let them try it and compare that prediction with the result. Allow a free reset so a mistaken purchase becomes an experiment rather than a punishment.&lt;/p&gt;

&lt;p&gt;The goal is to find out whether the player can connect the accumulation they see with the next useful change.&lt;/p&gt;

&lt;h2&gt;
  
  
  Compare two presentations of the same rules
&lt;/h2&gt;

&lt;p&gt;In one version, show production and consumption rates beside the machines, with simple activity animations. In the other, keep those same numbers and add visible blanks moving and accumulating along the line. Do not change the economy between versions.&lt;/p&gt;

&lt;p&gt;Record the player's prediction, first upgrade, corrections and explanation. A correct click without an explanation might be a guess. A slower decision accompanied by a clear account of the blockage could show better understanding.&lt;/p&gt;

&lt;p&gt;Also ask whether watching the process was enjoyable. Diagnosing the line and enjoying its motion are separate outcomes.&lt;/p&gt;

&lt;p&gt;Use labels or distinct machine poses for “working,” “waiting for input” and “output full.” Color alone should not carry those meanings. If movement is visually crowded, test a simpler display rather than adding more particles.&lt;/p&gt;

&lt;h2&gt;
  
  
  Make the improvement observable
&lt;/h2&gt;

&lt;p&gt;After the fabricator upgrade, the waiting pile should change according to the simulation. If output remains capped elsewhere, the display should show that too. A celebratory effect must not imply a production gain that never happened.&lt;/p&gt;

&lt;p&gt;This is a narrow prototype: it tests understanding of one constrained line, not long-term progression or an entire factory economy. Once the cause and effect are readable, more machines and competing upgrades may become worthwhile.&lt;/p&gt;

&lt;p&gt;Factorio's developer discussions suggest a useful starting point for an SF workshop: make the machinery worth watching, then check whether watching it helps the player decide what to improve.&lt;/p&gt;

</description>
      <category>gamedev</category>
    </item>
    <item>
      <title>Prey: Let a Disguise Change What the Player Can Do</title>
      <dc:creator>Lucas</dc:creator>
      <pubDate>Mon, 09 Nov 2026 23:45:19 +0000</pubDate>
      <link>https://dev.to/_9848c5582063b42abecb7/prey-let-a-disguise-change-what-the-player-can-do-4k00</link>
      <guid>https://dev.to/_9848c5582063b42abecb7/prey-let-a-disguise-change-what-the-player-can-do-4k00</guid>
      <description>&lt;p&gt;AI assisted with source research and drafting. The disguise prototype below is a design proposal, not a completed playtest or a claim of personal experience with Prey.&lt;/p&gt;

&lt;p&gt;A disguise can change how a character looks, how an observer reacts, and which actions become possible. Those changes need to agree. If the player becomes a small object but still moves through the world exactly like a person, the new appearance may promise more than the game supports.&lt;/p&gt;

&lt;p&gt;Prey's Mimic Matter offers a concrete SF example of connecting a borrowed appearance to play.&lt;/p&gt;

&lt;h2&gt;
  
  
  Give the borrowed form a practical use
&lt;/h2&gt;

&lt;p&gt;In Bethesda's February 2017 &lt;a href="https://bethesda.net/en-US/news/prey-mimic-madness" rel="noopener noreferrer"&gt;Mimic Madness feature&lt;/a&gt;, lead designer Ricardo Bare describes using Mimic Matter to hide from an alien, prepare an ambush, or become small enough to enter a security booth through its slot. The same article describes copying everyday objects such as a mug or lamp.&lt;/p&gt;

&lt;p&gt;The March 2017 &lt;a href="https://bethesda.net/en-US/news/prey-powers-guide-and-video" rel="noopener noreferrer"&gt;powers guide&lt;/a&gt; adds an important qualification: the player can move while disguised, but movement may attract attention. It also documents a continuing psi cost.&lt;/p&gt;

&lt;p&gt;These are historical descriptions of the game's mechanics. They support a design question rather than a claim about what caused its success: what does the disguise allow, and what must the player do to keep benefiting from it?&lt;/p&gt;

&lt;p&gt;The small-object example changes a spatial possibility. The warning about movement changes behavior around observers. Together, they make the copied form more than a costume selection.&lt;/p&gt;

&lt;h2&gt;
  
  
  Write down the disguise's promises
&lt;/h2&gt;

&lt;p&gt;For a new prototype, separate three decisions:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Appearance:&lt;/strong&gt; What does the player become?&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Permission:&lt;/strong&gt; What can that form do that the ordinary character cannot?&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Exposure:&lt;/strong&gt; Which behavior gives the disguise away?&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;This is a proposed design checklist, not a description of Prey's internal implementation.&lt;/p&gt;

&lt;p&gt;A convincing visual transformation sets expectations before a tutorial says anything. A small rolling object suggests different movement from a tall, walking character. A stationary container suggests different behavior from a maintenance robot. The game does not need to simulate every property of either object, but it should make its chosen rules understandable.&lt;/p&gt;

&lt;p&gt;Keep the initial promise narrow. One form with one useful permission is enough to test whether players understand why they would choose it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Test one form in one room
&lt;/h2&gt;

&lt;p&gt;Imagine a fictional space-station storeroom. A normal-sized character can reach the exit by taking a long corridor. A small disguised form can instead pass through a low opening. Both routes work.&lt;/p&gt;

&lt;p&gt;Add a patrolling observer near the opening. In this invented prototype, the observer ignores the disguised player while they remain still, but notices movement within its visible viewing area. Give the player clear feedback when noticed, plus a nearby place to retreat and try again.&lt;/p&gt;

&lt;p&gt;These are prototype rules, not a reconstruction of Prey's enemy detection.&lt;/p&gt;

&lt;p&gt;First test the room without a timer or resource drain. That lets you examine whether players understand the form and its exposure rule before combining them with another constraint. Explain the controls, then let them choose a route without prescribing the solution.&lt;/p&gt;

&lt;p&gt;Watch for three different failures:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;They never consider the opening because they do not connect the small form with fitting through it.&lt;/li&gt;
&lt;li&gt;They understand the route but cannot tell when the observer can see them.&lt;/li&gt;
&lt;li&gt;They understand both rules but find stopping and moving tedious.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Those failures need different revisions. Making the opening clearer will not fix a dull waiting pattern. Shortening the patrol will not explain an unclear disguise.&lt;/p&gt;

&lt;p&gt;Afterward, ask what becoming the object let them do, what could expose them, and why they chose their route. Fast completion alone does not show that the disguise was satisfying.&lt;/p&gt;

&lt;h2&gt;
  
  
  Keep the alternative worth choosing
&lt;/h2&gt;

&lt;p&gt;The longer corridor matters because it lets a player decline the disguise. If everyone takes the small opening, that could mean the disguise is enjoyable, or simply that the other route is unreasonably slow.&lt;/p&gt;

&lt;p&gt;Vary the corridor length before adding more forms. Look for a situation where players can explain a preference rather than identify one mandatory answer. Some may enjoy timing their movements around an observer. Others may prefer the predictable walk.&lt;/p&gt;

&lt;p&gt;A later test could add a limited duration or resource cost, but that would answer a new question. First establish that the basic disguise is readable and worth using.&lt;/p&gt;

&lt;p&gt;Prey's documented ability suggests a useful starting point: give the borrowed form a concrete consequence, then let the player act under its rules. Before expanding a catalogue of disguises, find out whether one transformation changes a decision in a way players understand and enjoy.&lt;/p&gt;

</description>
      <category>gamedev</category>
    </item>
    <item>
      <title>Starfield: Make Room for the Player Inside the Ship</title>
      <dc:creator>Lucas</dc:creator>
      <pubDate>Sun, 08 Nov 2026 23:54:41 +0000</pubDate>
      <link>https://dev.to/_9848c5582063b42abecb7/starfield-make-room-for-the-player-inside-the-ship-169e</link>
      <guid>https://dev.to/_9848c5582063b42abecb7/starfield-make-room-for-the-player-inside-the-ship-169e</guid>
      <description>&lt;p&gt;AI assisted with source research and drafting. The cabin prototype below is a design proposal, not a completed playtest or a claim of personal experience with Starfield.&lt;/p&gt;

&lt;p&gt;A spaceship can express a player's choices through its engines, weapons and hull. Its interior offers another opportunity: arranging a place they want to return to. Housing design can matter aboard a ship even when a chair does not improve its combat statistics.&lt;/p&gt;

&lt;p&gt;Starfield's ship-decoration update provides a concrete case for examining that opportunity.&lt;/p&gt;

&lt;h2&gt;
  
  
  An interior becomes something the player can arrange
&lt;/h2&gt;

&lt;p&gt;Bethesda's &lt;a href="https://bethesda.net/en-US/news/starfield-may-update-highlights" rel="noopener noreferrer"&gt;May 2024 feature overview&lt;/a&gt; describes expanded ship-interior customization using a system similar to outpost decoration. It also introduces empty hab modules so players can begin decorating from scratch.&lt;/p&gt;

&lt;p&gt;The &lt;a href="https://bethesda.net/en-US/news/starfield-update-1-11-36-may-15-2024" rel="noopener noreferrer"&gt;May 15, 2024 release notes for update 1.11.36&lt;/a&gt; confirm the addition of Ship Decoration mode. These are historical features, not a claim about a forthcoming update.&lt;/p&gt;

&lt;p&gt;The design interpretation is straightforward: choosing a ship's components and arranging its living space invite different kinds of decisions. One player may care about a tidy work area; another may want a crowded room that looks inhabited. A housing system gives those preferences somewhere to appear.&lt;/p&gt;

&lt;p&gt;The sources establish the feature. They do not establish how much attachment it creates, whether every player uses it, or whether it explains the game's commercial performance.&lt;/p&gt;

&lt;h2&gt;
  
  
  Start with one cabin and a few objects
&lt;/h2&gt;

&lt;p&gt;For a small prototype, imagine a fictional courier ship with one cabin. Give the player a bed, a table, a chair, a lamp and a storage cabinet. Fix the cabin's size and the position of its door.&lt;/p&gt;

&lt;p&gt;Let the player move and rotate the furniture, then leave editing mode and walk through the result. Do not add a room score, stat bonuses or a requirement to place every item in this first test. The purpose is to examine whether arranging the room is interesting on its own.&lt;/p&gt;

&lt;p&gt;These are proposed rules, not a description of Starfield's editor or furniture behavior.&lt;/p&gt;

&lt;p&gt;The first useful observation is what the player tries to change. Do they arrange objects around an activity, such as reading near the lamp? Do they group things by appearance? Do they want to leave an area empty? Those choices can suggest what the next set of housing features should support.&lt;/p&gt;

&lt;p&gt;An object count alone would miss this. A player who places only three pieces might have made a deliberate composition, while another who uses all five might simply be following an assumed completion rule.&lt;/p&gt;

&lt;h2&gt;
  
  
  Separate a difficult tool from an uninteresting choice
&lt;/h2&gt;

&lt;p&gt;A blank cabin can offer freedom, but it can also leave someone unsure where to begin. Prepare a second starting state with the same furniture already arranged. Let the player modify it freely.&lt;/p&gt;

&lt;p&gt;Compare the two starting states with different participants, or vary their order if each person tries both. Keep the available objects and editing controls the same. Ask which changes they wanted to make and which changes the editor prevented.&lt;/p&gt;

&lt;p&gt;If they have clear ideas but struggle to place the chair, improve placement before expanding the furniture catalogue. If they can edit easily but have no interest in the result, adding more objects may not solve the problem. Some players may prefer a finished cabin and spend their time elsewhere.&lt;/p&gt;

&lt;p&gt;Also inspect the room outside the editor. Can the player enter, see the arrangement at the normal camera angle, and reach the places they intended to use? An attractive editing view can hide an awkward everyday space.&lt;/p&gt;

&lt;h2&gt;
  
  
  Give the arrangement a chance to matter
&lt;/h2&gt;

&lt;p&gt;After a short unrelated activity, bring the player back to the cabin without asking them to decorate again. Notice whether they recognize their changes, choose to adjust something, or pass straight through. Ask what, if anything, makes the room feel like theirs.&lt;/p&gt;

&lt;p&gt;Treat those observations as reasons to revise the prototype, not proof of long-term attachment. Decorating once and caring about a space over many sessions are different outcomes.&lt;/p&gt;

&lt;p&gt;Starfield's documented ship-decoration tools suggest a useful SF housing question: where can the player make a small, visible choice about how life aboard the ship looks? Test that in one usable room before building a hundred furniture options.&lt;/p&gt;

</description>
      <category>gamedev</category>
    </item>
    <item>
      <title>Ratchet &amp; Clank: Rift Apart Gives Small Pickups a Physical Response</title>
      <dc:creator>Lucas</dc:creator>
      <pubDate>Sat, 07 Nov 2026 23:40:26 +0000</pubDate>
      <link>https://dev.to/_9848c5582063b42abecb7/ratchet-clank-rift-apart-gives-small-pickups-a-physical-response-5d83</link>
      <guid>https://dev.to/_9848c5582063b42abecb7/ratchet-clank-rift-apart-gives-small-pickups-a-physical-response-5d83</guid>
      <description>&lt;p&gt;AI assisted with source research and drafting. The prototype described below is a proposal, not a completed playtest or a claim of personal gameplay experience.&lt;/p&gt;

&lt;p&gt;Explosions are obvious places to spend an action game's feedback budget. A routine pickup is easier to overlook. Yet it may happen far more often, so its response deserves attention before another spectacular effect is added.&lt;/p&gt;

&lt;p&gt;Ratchet &amp;amp; Clank: Rift Apart provides a concrete SF example: the player can feel a small reward arriving through the controller.&lt;/p&gt;

&lt;h2&gt;
  
  
  Look at the ordinary action
&lt;/h2&gt;

&lt;p&gt;In a &lt;a href="https://blog.playstation.com/2021/05/13/devs-reveal-their-upcoming-dualsense-wireless-controller-implementations/" rel="noopener noreferrer"&gt;May 2021 PlayStation Blog contribution&lt;/a&gt;, Insomniac Games describes using haptics for more than weapon fire and wrench impacts. Collecting bolts produces small impulses accompanied by subtle controller audio. Highlighting items in the weapon wheel or map also produces a faint tactile click.&lt;/p&gt;

&lt;p&gt;The team presents these details as ways to make ordinary interactions more satisfying. That is the developer's description, not a measured result reported by this article.&lt;/p&gt;

&lt;p&gt;The interesting design choice is the scale of the response. A pickup does not need to borrow the force of an explosion to acknowledge the player's action. A short, restrained response can give a small interaction its own texture.&lt;/p&gt;

&lt;p&gt;That suggests a useful question for an SF prototype: what should collecting a piece of fictional machinery feel like, and how often will the player feel it?&lt;/p&gt;

&lt;h2&gt;
  
  
  Review a sequence, not one perfect pickup
&lt;/h2&gt;

&lt;p&gt;Imagine a fictional repair game in which a maintenance robot collects loose energy cells. Each cell disappears into its container and increments a visible count. Keep that basic confirmation intact.&lt;/p&gt;

&lt;p&gt;For a first experiment, make two versions:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;A uses the existing animation, count update and short pickup sound.&lt;/li&gt;
&lt;li&gt;B keeps those elements and adds a brief, low-intensity controller pulse when the cell is actually collected.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;These are proposed prototype rules, not a reconstruction of Rift Apart. Keep the reward value, movement and pickup locations unchanged. Otherwise, a more generous reward could be mistaken for better feedback.&lt;/p&gt;

&lt;p&gt;Begin with one cell, then try several collected in quick succession. A pulse that feels pleasant in isolation may become a continuous buzz when events overlap. The sequence matters because the player experiences the combined output, not the individual effect assets.&lt;/p&gt;

&lt;p&gt;If the rapid sequence becomes muddy, test a shorter response or a grouped response for closely spaced pickups. Treat grouping as another hypothesis: it might reduce clutter, but it could also make separate rewards harder to notice. Do not silently change the reward count to match the number of pulses.&lt;/p&gt;

&lt;p&gt;Ask players what they noticed and whether the collection felt satisfying, distracting or tiring. Also check whether they can tell when collection has finished. An enthusiastic reaction to the first pulse is not enough to settle how it works across repeated encounters.&lt;/p&gt;

&lt;h2&gt;
  
  
  Leave room to turn the effect down
&lt;/h2&gt;

&lt;p&gt;Insomniac's &lt;a href="https://support.insomniac.games/hc/en-us/articles/46716208107795-What-Accessibility-options-does-Ratchet-Clank-Rift-Apart-feature" rel="noopener noreferrer"&gt;accessibility documentation&lt;/a&gt; lists Experiential, Functional and Off vibration settings, along with overall vibration intensity. Functional limits vibration to useful gameplay cues, while Off disables it. The documentation separately lists controller-speaker controls.&lt;/p&gt;

&lt;p&gt;For the proposed repair game, keep the reward and its visible confirmation available with vibration disabled. Let participants choose a comfortable setting before comparing versions. If someone prefers no vibration, that is a valid result, not a failed appreciation of the effect.&lt;/p&gt;

&lt;p&gt;There is no universal pulse strength that can be inferred from the developer's account. Hardware, event frequency and individual preference all remain part of the design problem.&lt;/p&gt;

&lt;p&gt;The practical lesson from Rift Apart is modest: review the small actions that fill the space between big scenes. A tiny response, tied to an actual pickup and comfortable to repeat, may contribute more to everyday game feel than another layer of spectacle. Test that possibility in a short sequence before extending it across the whole game.&lt;/p&gt;

</description>
      <category>gamedev</category>
    </item>
    <item>
      <title>My sub-second sprite judge can't tell a profile from a 3/4 view</title>
      <dc:creator>Lucas</dc:creator>
      <pubDate>Sat, 07 Nov 2026 05:18:16 +0000</pubDate>
      <link>https://dev.to/_9848c5582063b42abecb7/my-sub-second-sprite-judge-cant-tell-a-profile-from-a-34-view-4eg3</link>
      <guid>https://dev.to/_9848c5582063b42abecb7/my-sub-second-sprite-judge-cant-tell-a-profile-from-a-34-view-4eg3</guid>
      <description>&lt;p&gt;I run a pipeline that draws 2D fighting-game characters with image and video models, then checks every drawing before it ships. A check that fails means paying to redraw. For a while the checks went to a cloud vision model, which takes about 10–15 seconds per call. So I wrote a NumPy-only "pixel judge" that answers in under a second. Then I found the catch. On side-view takes, the expensive model still gets called almost every time, because the pixel judge can't tell a true profile from a three-quarter view.&lt;/p&gt;

&lt;p&gt;This post is the research I did before fixing it. &lt;strong&gt;Nothing below is built yet, and I have no new benchmark numbers.&lt;/strong&gt; It's a plan with sources, and I'd like you to poke holes in it.&lt;/p&gt;

&lt;h2&gt;
  
  
  The setup
&lt;/h2&gt;

&lt;p&gt;Each fighter starts as a full-body front painting. From that front, the pipeline draws a side view, then 11 motions (idle, walk, run, attacks, block, hit, fall, and so on). Two kinds of judgement happen along the way:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Side take:&lt;/strong&gt; is this new drawing the same character, seen from the side, with whole limbs?&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Loop:&lt;/strong&gt; do the walk and run cycles actually alternate legs and loop cleanly?&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The worker supports three judge modes:&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;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;codex&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Every check goes to a cloud VLM (~10–15 s per call)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;pixels&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Only the NumPy judge (&lt;code&gt;pixels-v1&lt;/code&gt;, under 1 s). A check it can't decide counts as a &lt;strong&gt;pass&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;hybrid&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Pixels first. A pixel &lt;em&gt;fail&lt;/em&gt; is final. Only &lt;code&gt;unknown&lt;/code&gt; checks escalate to the VLM&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;code&gt;hybrid&lt;/code&gt; is meant to be the fast-and-safe default. It isn't fast for side takes.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where the pixel judge goes blind
&lt;/h2&gt;

&lt;p&gt;Reading the code made the problem obvious. The pixel side-check always returns &lt;code&gt;side: unknown&lt;/code&gt;. Colour regions, silhouettes and frame differences can't separate "face pointing left" from "face turned 45° toward the camera". So in &lt;code&gt;hybrid&lt;/code&gt;, every side take that passes the other checks escalates anyway, and the time saving there is close to zero.&lt;/p&gt;

&lt;p&gt;In &lt;code&gt;pixels&lt;/code&gt; mode it's worse, because &lt;code&gt;unknown&lt;/code&gt; means pass. That lets through:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;three-quarter takes labelled as side views&lt;/li&gt;
&lt;li&gt;face changes&lt;/li&gt;
&lt;li&gt;parts that change size between frames (the gloves shrink mid-walk)&lt;/li&gt;
&lt;li&gt;wrong limb shapes&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Two other judgements never touch pixels at all: the &lt;strong&gt;front check&lt;/strong&gt; (exactly one character, whole from head to soles, facing the viewer) and the &lt;strong&gt;look judge&lt;/strong&gt;. Both always go to the cloud model.&lt;/p&gt;

&lt;p&gt;The thresholds that do exist came from 6 probe fighters and 20 verdicts (8 takes, 12 loops). Worse, the work folders are deleted after 7 days, so verdicts never pile up into anything I could measure against.&lt;/p&gt;

&lt;h2&gt;
  
  
  Research A: small local models for anime art
&lt;/h2&gt;

&lt;p&gt;I wanted CPU-friendly models, ideally ONNX, with permissive licences.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;&lt;a href="https://github.com/deepghs/imgutils" rel="noopener noreferrer"&gt;deepghs/imgutils&lt;/a&gt;&lt;/strong&gt; (MIT): anime person, head and face detectors, plus &lt;strong&gt;CCIP&lt;/strong&gt;, a contrastive model for "is this the same character?". The detectors are tiny (a few million parameters). Some of its models and weights carry OpenRAIL licences, so check each one.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;&lt;a href="https://huggingface.co/SmilingWolf/wd-swinv2-tagger-v3" rel="noopener noreferrer"&gt;WD14 tagger v3, SwinV2&lt;/a&gt;&lt;/strong&gt; (Apache-2.0, ONNX). Its tag list has exactly the vocabulary I need: &lt;code&gt;profile&lt;/code&gt;, &lt;code&gt;from_side&lt;/code&gt;, &lt;code&gt;facing_viewer&lt;/code&gt;, &lt;code&gt;solo&lt;/code&gt;, &lt;code&gt;full_body&lt;/code&gt;, &lt;code&gt;multiple_views&lt;/code&gt;, &lt;code&gt;feet_out_of_frame&lt;/code&gt;, &lt;code&gt;head_out_of_frame&lt;/code&gt;. I didn't find a three-quarter-view tag, which matters below.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;&lt;a href="https://github.com/Tau-J/rtmlib" rel="noopener noreferrer"&gt;rtmlib&lt;/a&gt;&lt;/strong&gt; (Apache-2.0): RTMPose and DWPose on ONNX Runtime, no mmcv. It was trained on photos (COCO), so its accuracy on anime limbs is an open question.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;DINOv2-small&lt;/strong&gt; (Apache-2.0, 22M params) for appearance similarity. The &lt;a href="https://arxiv.org/abs/2208.12242" rel="noopener noreferrer"&gt;DreamBooth&lt;/a&gt; authors reported that DINO similarity tracks human judgement better than CLIP image similarity (about 0.32 vs 0.27 correlation).&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The honest gap: &lt;strong&gt;I found no validated model that estimates the yaw (left-right turn) of an anime face.&lt;/strong&gt; So profile-vs-3/4 has to come from several weak signals combined:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;WD14 scores for &lt;code&gt;profile&lt;/code&gt;, &lt;code&gt;from_side&lt;/code&gt; and &lt;code&gt;facing_viewer&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;where the face box sits inside the head box, and whether a face is detected at all&lt;/li&gt;
&lt;li&gt;optionally, facial landmarks: is only one eye visible?&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Research B: a small local VLM, used carefully
&lt;/h2&gt;

&lt;p&gt;For machines with a GPU, a small VLM could sit between the pixel judge and the cloud. &lt;strong&gt;&lt;a href="https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct" rel="noopener noreferrer"&gt;Qwen3-VL-4B Instruct&lt;/a&gt;&lt;/strong&gt; (Apache-2.0) is about 3.3 GB on &lt;a href="https://ollama.com/library/qwen3-vl" rel="noopener noreferrer"&gt;Ollama&lt;/a&gt;. It should fit an 8 GB card, but I found no published latency on an RTX 4060, so I'll have to measure it.&lt;/p&gt;

&lt;p&gt;The bigger lesson was &lt;em&gt;how&lt;/em&gt; to ask. Techniques that make VLM judging steadier:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Atomic yes/no questions.&lt;/strong&gt; Split "is this a valid side view?" into small checks (&lt;a href="https://arxiv.org/abs/2310.18235" rel="noopener noreferrer"&gt;Davidsonian Scene Graph&lt;/a&gt;).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Read the probability, not the word.&lt;/strong&gt; Score with P("Yes") from logprobs, as &lt;a href="https://arxiv.org/abs/2404.01291" rel="noopener noreferrer"&gt;VQAScore&lt;/a&gt; does, then put a threshold on a continuous number.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ask twice, swapped.&lt;/strong&gt; When comparing two images, ask in both orders to cancel position bias.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;One grid image instead of many.&lt;/strong&gt; Tile frames into a single image (&lt;a href="https://arxiv.org/abs/2403.18406" rel="noopener noreferrer"&gt;IG-VLM&lt;/a&gt;).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cap image tokens&lt;/strong&gt; with &lt;code&gt;max_pixels&lt;/code&gt; so latency stays predictable.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Research C: a cascade with calibrated thresholds
&lt;/h2&gt;

&lt;p&gt;My &lt;code&gt;pixels → cloud VLM&lt;/code&gt; setup already is a judge cascade. It just has hand-picked thresholds. &lt;strong&gt;&lt;a href="https://arxiv.org/abs/2407.18370" rel="noopener noreferrer"&gt;Trust or Escalate&lt;/a&gt;&lt;/strong&gt; formalises it: a cheap judge decides when it's confident, and escalates to a stronger judge otherwise. The confidence threshold is calibrated so that agreement with humans on the non-escalated cases meets a level you choose.&lt;/p&gt;

&lt;p&gt;Conformal risk control and Learn-then-Test can calibrate several thresholds at once. But with 20 calibration examples any guarantee is very loose. A realistic start is 50–100 labels per check, growing to hundreds. The labelling order is cheap: &lt;strong&gt;label the cases where the judges disagree first.&lt;/strong&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  The plan, in priority order
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Keep every verdict.&lt;/strong&gt; Copy each &lt;code&gt;pixels.json&lt;/code&gt; (with its measurements), each cloud verdict and a small thumbnail to an archive outside the 7-day cleanup. No user prompts included. Add an eval script that replays a labelled set and prints, per check, a confusion matrix (pixels vs VLM vs human), the escalation rate, and p50/p95 latency. Everything else depends on this.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Make &lt;code&gt;side&lt;/code&gt; decidable.&lt;/strong&gt; Add a &lt;code&gt;models&lt;/code&gt; stage (WD14 + head/face boxes, maybe eye landmarks). It answers pass/fail only when confident and otherwise stays &lt;code&gt;unknown&lt;/code&gt;. This has the biggest speed payoff.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Identity and part sizes.&lt;/strong&gt; Add CCIP or DINOv2 cosine similarity next to the colour comparison, so a matching palette with a changed face or outfit gets caught. Track the variance of head-box height over body height across frames for the shrinking-glove case.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Pre-screen the front check locally.&lt;/strong&gt; Exactly one person, &lt;code&gt;solo&lt;/code&gt; + &lt;code&gt;full_body&lt;/code&gt;, none of &lt;code&gt;multiple_views&lt;/code&gt; / &lt;code&gt;feet_out_of_frame&lt;/code&gt; / &lt;code&gt;head_out_of_frame&lt;/code&gt;, and &lt;code&gt;facing_viewer&lt;/code&gt;. Only &lt;em&gt;obvious fails&lt;/em&gt; end locally. Anything that looks like a pass still goes to the cloud model, because this is a registration gate.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Optional GPU tier.&lt;/strong&gt; Qwen3-VL-4B with atomic questions and P(yes), only after steps 1–4 and only if its measured latency and accuracy earn it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Set thresholds from data.&lt;/strong&gt; Recalibrate per check with the Trust-or-Escalate method as labels grow. Re-test the rule "a pixel fail is final", because a false fail costs a paid redraw. Until then, label current thresholds as uncalibrated.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Constraints I'm keeping: new models live in a separate host-side venv, away from the generation engine. Model versions and hashes go into the judge cache key, so changing a model marks old verdicts stale. MIT/Apache licences come first, OpenRAIL only after reading the terms, and no GPL/AGPL models.&lt;/p&gt;

&lt;h2&gt;
  
  
  Honest limits
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;None of this is measured yet. The 10–15 s and under-1 s figures are what the current judges take. Every gain above is a hypothesis until step 1's eval set exists.&lt;/li&gt;
&lt;li&gt;Profile vs 3/4 will probably stay &lt;strong&gt;partly&lt;/strong&gt; unsolved. Without a yaw model, the best case is "confident on the clear cases, escalate the rest".&lt;/li&gt;
&lt;li&gt;Photo-trained pose models may simply be wrong on chibi or stylised anatomy.&lt;/li&gt;
&lt;li&gt;A local VLM only helps machines with a GPU. CPU-only generation PCs skip that tier.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What I'd love to hear
&lt;/h2&gt;

&lt;p&gt;Have you found &lt;strong&gt;anything that reliably estimates head yaw on anime or cel-shaded art&lt;/strong&gt;? Or have you calibrated a cheap-judge → expensive-judge cascade in production? How many labels did it take before the thresholds stopped moving?&lt;/p&gt;

</description>
      <category>machinelearning</category>
      <category>computervision</category>
      <category>gamedev</category>
      <category>ai</category>
    </item>
    <item>
      <title>No Man's Sky: Give a Discovery Catalogue Something to Remember</title>
      <dc:creator>Lucas</dc:creator>
      <pubDate>Fri, 06 Nov 2026 23:45:19 +0000</pubDate>
      <link>https://dev.to/_9848c5582063b42abecb7/no-mans-sky-give-a-discovery-catalogue-something-to-remember-3fe5</link>
      <guid>https://dev.to/_9848c5582063b42abecb7/no-mans-sky-give-a-discovery-catalogue-something-to-remember-3fe5</guid>
      <description>&lt;p&gt;AI assisted with source research and drafting. The small prototype below is a design proposal, not a completed playtest or a claim of personal experience with the game.&lt;/p&gt;

&lt;p&gt;A themed collection needs more than additional entries. It needs a way for the player to recognize what makes an entry worth keeping. In an SF exploration game, a catalogue can make unfamiliar creatures and planets easier to remember without turning every discovery into equipment.&lt;/p&gt;

&lt;p&gt;No Man's Sky offers a useful example through its Wonders catalogue.&lt;/p&gt;

&lt;h2&gt;
  
  
  Give each entry a reason to stand out
&lt;/h2&gt;

&lt;p&gt;Hello Games' &lt;a href="https://www.nomanssky.com/fractal-update/" rel="noopener noreferrer"&gt;Fractal update description&lt;/a&gt; introduced a Wonders section containing notable discoveries from the player's journey. Its examples include an unusually large herbivore, an ancient tree, and a planet approaching paradise. The same update records valuable treasures and retrieved planetary glitches.&lt;/p&gt;

&lt;p&gt;These are documented features of update 4.1, not an announcement of a new release. The useful design observation is that the collection presents several ways for something to be noteworthy. A creature, a tree, and a planet do not need to compete on one shared power statistic.&lt;/p&gt;

&lt;p&gt;The later &lt;a href="https://www.nomanssky.com/interceptor-update/" rel="noopener noreferrer"&gt;Interceptor update description&lt;/a&gt; documents Personal Records, where players can bookmark favourite discoveries, including planets, alien creatures, and flowers. That adds room for individual preference alongside the catalogue's record categories.&lt;/p&gt;

&lt;p&gt;This is a design interpretation, not evidence that the catalogue caused the game's commercial success. The narrower question is practical: when a player opens a collection, can they explain why its entries differ?&lt;/p&gt;

&lt;h2&gt;
  
  
  Prototype the catalogue before expanding the universe
&lt;/h2&gt;

&lt;p&gt;Consider a fictional survey game with six alien plants. Give each plant a distinct image, a short name, a location, and one observable feature. One folds its leaves during rain; another grows around warm rocks. These are invented examples, not No Man's Sky species.&lt;/p&gt;

&lt;p&gt;Let a short exploration session add the plants to a catalogue. Start with a simple entry card:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Show the plant clearly enough to recognize it again.&lt;/li&gt;
&lt;li&gt;State the feature that distinguishes it.&lt;/li&gt;
&lt;li&gt;Preserve where it was found.&lt;/li&gt;
&lt;li&gt;Make the collected state clear without obscuring the illustration.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The purpose is to connect the entry to something the player noticed. A glowing completion badge cannot carry that connection by itself.&lt;/p&gt;

&lt;p&gt;Keep the initial test small. After the player collects several plants, ask them to find the entry for one they encountered and explain what they remember about it. Then show a new view of that plant in the world. Can they match it to the catalogue without reading the answer aloud?&lt;/p&gt;

&lt;p&gt;If they remember only the rarity colour, the collection may be communicating a ranking more clearly than the plants themselves. That can be appropriate for some games. It does not yet support the goal of making the fictional ecology memorable.&lt;/p&gt;

&lt;h2&gt;
  
  
  Check what another entry actually adds
&lt;/h2&gt;

&lt;p&gt;Adding thirty more plants would test content volume before resolving the smaller design problem. First inspect the six existing entries for repeated images, indistinguishable descriptions, or details the player never had a chance to observe.&lt;/p&gt;

&lt;p&gt;A collection can also become administrative work. If players open it only to clear notification dots, the interface may be demanding attention without giving them a useful reason to return. Observe voluntary revisits, but ask what prompted them; a revisit might reflect interest or confusion.&lt;/p&gt;

&lt;p&gt;The prototype's result could be that players enjoy the illustrations while ignoring the descriptions. That would be useful direction for the next iteration, rather than a reason to force more text into the cards.&lt;/p&gt;

&lt;p&gt;The Wonders catalogue suggests a modest way to examine SF worldbuilding: give discoveries a persistent, recognizable place in the player's collection. Before adding more alien life, test whether the existing entries preserve something the player can actually remember.&lt;/p&gt;

</description>
      <category>gamedev</category>
    </item>
    <item>
      <title>Metroid Dread: Let the Opening Deliver One Complete Action</title>
      <dc:creator>Lucas</dc:creator>
      <pubDate>Thu, 05 Nov 2026 23:37:17 +0000</pubDate>
      <link>https://dev.to/_9848c5582063b42abecb7/metroid-dread-let-the-opening-deliver-one-complete-action-174m</link>
      <guid>https://dev.to/_9848c5582063b42abecb7/metroid-dread-let-the-opening-deliver-one-complete-action-174m</guid>
      <description>&lt;p&gt;AI assisted with source research and drafting. The prototype below is a design proposal, not a completed playtest or a claim of personal development experience.&lt;/p&gt;

&lt;p&gt;An SF action game can introduce a sophisticated suit by listing its abilities. It can also give the player a short situation in which one ability produces a satisfying, understandable result. Those openings ask the player to learn in different ways.&lt;/p&gt;

&lt;p&gt;Metroid Dread offers a useful example of the second approach, alongside explicit instructions.&lt;/p&gt;

&lt;h2&gt;
  
  
  Let an early enemy demonstrate the move
&lt;/h2&gt;

&lt;p&gt;In &lt;a href="https://www.nintendo.com/en-gb/News/2021/September/Metroid-Dread-Report-Vol-9-Handy-tips-for-newcomers-2047809.html" rel="noopener noreferrer"&gt;Nintendo's September 2021 report for newcomers&lt;/a&gt;, the development team describes connecting actions smoothly. The report explains Dash Melee as a running version of the Melee Counter that also damages enemies. The team notes that some early enemies can be defeated with a single Dash Melee, giving players an opportunity to experience clearing their path while moving.&lt;/p&gt;

&lt;p&gt;The same report says the game explains controls during play and introduces the story at the beginning. It does not present discovery and explanation as mutually exclusive.&lt;/p&gt;

&lt;p&gt;The design interpretation is about sequence: an early encounter can let the player complete a small action before asking them to manage a larger collection of systems. The fictional technology becomes something they have used, with a result they can recognize.&lt;/p&gt;

&lt;p&gt;This is not evidence that a particular opening increased retention or caused the game's success. It is a documented design choice worth examining at prototype scale.&lt;/p&gt;

&lt;h2&gt;
  
  
  Specify what the first success means
&lt;/h2&gt;

&lt;p&gt;Imagine a fictional station-exploration game with a powered slide. Its opening room contains one low obstruction and an exit. The slide gets the player through the obstruction; reaching the exit completes the encounter.&lt;/p&gt;

&lt;p&gt;Before adding equipment menus, write down the intended first experience:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Notice the obstruction.&lt;/li&gt;
&lt;li&gt;Learn or discover the slide input.&lt;/li&gt;
&lt;li&gt;Use the slide to pass underneath.&lt;/li&gt;
&lt;li&gt;Recognize that the encounter is complete.&lt;/li&gt;
&lt;li&gt;Reach the next decision.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;This is a proposed sequence, not a description of Metroid Dread's opening rooms.&lt;/p&gt;

&lt;p&gt;A short instruction may be appropriate at step two. The important question is whether the player then gets to perform the action and see its outcome. A prompt acknowledged by a button press is weak evidence that the player can use the move.&lt;/p&gt;

&lt;p&gt;The completion moment matters too. If the game opens three menus immediately after the slide, it may bury the result under new information. Consider letting the player reach a clear stopping point before introducing the next system.&lt;/p&gt;

&lt;h2&gt;
  
  
  Include the return to play
&lt;/h2&gt;

&lt;p&gt;The opening also needs a plan for failure. In &lt;a href="https://www.nintendo.com/en-gb/News/2021/June/Metroid-Dread-Report-Vol-2-Researching-the-E-M-M-I-1997672.html" rel="noopener noreferrer"&gt;Nintendo's June 2021 report about the E.M.M.I.&lt;/a&gt;, the developers describe nearby restart checkpoints for encounters with the pursuing robots. That is a separate documented feature, not a claim that those encounters are easy tutorials.&lt;/p&gt;

&lt;p&gt;For the invented slide room, inspect what happens after a missed input. Does the player retry the move, or repeat an unrelated introduction? The useful unit to review is the entire attempt, including recovery.&lt;/p&gt;

&lt;p&gt;Observe a few new players without coaching them through the intended sequence. Record the first point where they need help. Then offer another obstruction with a different placement. Can they recognize an opportunity to use the slide without the original prompt?&lt;/p&gt;

&lt;p&gt;These observations would guide revision; they would not establish a universal onboarding formula. A player may understand the move and simply dislike it. A story-led opening may also justify more time before control.&lt;/p&gt;

&lt;p&gt;Before expanding an SF protagonist's equipment, make one early action feel complete: a reason to try it, enough guidance to act, a visible result, and a clear way to continue. A modest first success can give the next explanation something concrete to build on.&lt;/p&gt;

</description>
      <category>gamedev</category>
    </item>
    <item>
      <title>Portal 2: A Puzzle Effect Must Preserve the Surface Underneath</title>
      <dc:creator>Lucas</dc:creator>
      <pubDate>Thu, 05 Nov 2026 03:33:30 +0000</pubDate>
      <link>https://dev.to/_9848c5582063b42abecb7/portal-2-a-puzzle-effect-must-preserve-the-surface-underneath-4ddf</link>
      <guid>https://dev.to/_9848c5582063b42abecb7/portal-2-a-puzzle-effect-must-preserve-the-surface-underneath-4ddf</guid>
      <description>&lt;p&gt;AI assisted with source research and drafting. The design review below is a proposal, not a completed playtest or a claim of personal experience developing Portal 2.&lt;/p&gt;

&lt;p&gt;A colorful coating can make a science-fiction puzzle look more interesting while making its next move harder to see. The problem is especially sharp when a surface already carries gameplay information before the effect arrives.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://store.steampowered.com/app/620/Portal_2/" rel="noopener noreferrer"&gt;Valve's description of Portal 2&lt;/a&gt; places its puzzles inside Aperture Science Labs, with new puzzle elements and test chambers. One of those elements gives an unusually concrete example of how fictional materials and usable visuals have to work together.&lt;/p&gt;

&lt;h2&gt;
  
  
  Two layers, two messages
&lt;/h2&gt;

&lt;p&gt;In Valve's 2011 presentation &lt;a href="https://steamcdn-a.akamaihd.net/apps/valve/2011/gdc_2011_grimes_nonstandard_textures.pdf" rel="noopener noreferrer"&gt;Making and Using Non-Standard Textures&lt;/a&gt;, the gel-development section describes an early rendering problem. The coating obscured the surface beneath it. Players needed to distinguish surfaces that accepted portals, so a visually opaque treatment could hide a useful action even when the rules still permitted it. The next implementation needed more translucency.&lt;/p&gt;

&lt;p&gt;The presentation also describes using translucent plastic and rubber references, and bubbles, to suggest the gel's thickness and elasticity. These are documented development choices, not proof that one material treatment works for every puzzle game.&lt;/p&gt;

&lt;p&gt;The useful design interpretation is that the new effect inherits an obligation to preserve existing information. The coating needs to communicate its own properties without erasing the player's understanding of the surface.&lt;/p&gt;

&lt;p&gt;That is a more specific problem than making an effect brighter or more attractive. A striking effect can succeed as an illustration and fail in the moment when someone must choose where to act.&lt;/p&gt;

&lt;h2&gt;
  
  
  Review the moment information changes
&lt;/h2&gt;

&lt;p&gt;For a fictional SF puzzle, imagine a coating applied to one of two wall panels. One panel accepts a tool; the other does not. The coating introduces an additional property, such as making an object rebound. Those are proposed rules for a prototype, not a reconstruction of Portal 2.&lt;/p&gt;

&lt;p&gt;Before polishing the effect, write down what a player must still be able to answer:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Which panel accepts the tool?&lt;/li&gt;
&lt;li&gt;Which area now has the coating's property?&lt;/li&gt;
&lt;li&gt;Where does that area end?&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Then inspect three moments: before application, while the material spreads, and after it settles. A clean final screenshot can miss a transition that briefly conceals the only useful target. Conversely, a dramatic application animation may be harmless if no decision is required until it finishes.&lt;/p&gt;

&lt;p&gt;Try a few presentation changes individually: preserve part of the base pattern, sharpen the coating's boundary, or reduce the visual density over a target. These are alternatives to investigate, not a recommendation that every material should simply become transparent. Excess transparency could make the coating itself difficult to locate.&lt;/p&gt;

&lt;p&gt;Ask someone to point out the two properties before explaining the intended answer. If they identify the coating but lose the panel's original role, the layers are competing. If they can identify both but still cannot solve the puzzle, the visual treatment may not be the problem.&lt;/p&gt;

&lt;p&gt;Color can help organize this information, but a prototype should also consider shape, pattern or another cue. Two meanings assigned to the same patch need to remain distinguishable under the conditions in which people will actually play.&lt;/p&gt;

&lt;h2&gt;
  
  
  Give fictional materials an information budget
&lt;/h2&gt;

&lt;p&gt;An SF material can feel unusual through its movement, texture and response without covering every available pixel with activity. The important constraint is what the player needs to read through it.&lt;/p&gt;

&lt;p&gt;Before adding another ripple, glow or bubble, ask what information becomes easier to see, and what becomes harder. Portal 2's documented gel iteration offers a useful reminder: visual polish includes protecting the meaning that was already there.&lt;/p&gt;

</description>
      <category>gamedev</category>
    </item>
    <item>
      <title>Subnautica: Making a Base Part of the Next Expedition</title>
      <dc:creator>Lucas</dc:creator>
      <pubDate>Tue, 03 Nov 2026 23:40:43 +0000</pubDate>
      <link>https://dev.to/_9848c5582063b42abecb7/subnautica-making-a-base-part-of-the-next-expedition-4e9f</link>
      <guid>https://dev.to/_9848c5582063b42abecb7/subnautica-making-a-base-part-of-the-next-expedition-4e9f</guid>
      <description>&lt;p&gt;AI assisted with source research and drafting. The experiment below is a proposal, not a completed playtest.&lt;/p&gt;

&lt;p&gt;When exploration supplies materials for construction, two activities already have a connection: go out, collect something, return and build. A useful next question is what the construction changes about the following trip.&lt;/p&gt;

&lt;p&gt;Subnautica provides a concrete example. &lt;a href="https://store.steampowered.com/app/264710/Subnautica/" rel="noopener noreferrer"&gt;Unknown Worlds’ official description on Steam&lt;/a&gt; says players collect ocean resources to make equipment, habitat modules and submersibles. Bases provide storage, vehicle parking and oxygen replenishment.&lt;/p&gt;

&lt;p&gt;That connection was explicit during development. In its &lt;a href="https://unknownworlds.com/en/news/coming-next-seabases" rel="noopener noreferrer"&gt;February 2015 seabase preview&lt;/a&gt;, Unknown Worlds explained that the Cyclops had already reduced the need to return to the starting Life Pod. Seabases were intended to extend that freedom by allowing players to establish operations elsewhere on the ocean floor. This is historical design evidence, not a description of an upcoming update.&lt;/p&gt;

&lt;p&gt;One design interpretation is that a base can give exploration a destination while also changing where the next expedition begins. That is an interpretation of the documented mechanics, not a claim about every player’s experience.&lt;/p&gt;

&lt;h2&gt;
  
  
  Test the connection with one small map
&lt;/h2&gt;

&lt;p&gt;A prototype does not need an entire alien ocean. Use a fictional underwater map with a starting habitat, two resource areas and one distant point of interest. Let the player collect materials and place one outpost.&lt;/p&gt;

&lt;p&gt;Compare two versions:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;A:&lt;/strong&gt; The outpost displays collected discoveries but does not provide expedition support.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;B:&lt;/strong&gt; The same outpost also replenishes the player’s air supply.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Keep construction cost, movement, air consumption and geography unchanged. These are proposed prototype rules, not a reconstruction of Subnautica.&lt;/p&gt;

&lt;p&gt;The hypothesis is that expedition support will change how players choose an outpost location and plan their next trip. Observe where they build, which route they take afterward and whether they can explain the relationship. Ask, “What did this outpost let you do differently?”&lt;/p&gt;

&lt;p&gt;Completion time alone would miss the point. A longer trip might reflect interesting exploration, or tedious travel. The player’s explanation helps distinguish those possibilities.&lt;/p&gt;

&lt;h2&gt;
  
  
  Watch for a compulsory chore
&lt;/h2&gt;

&lt;p&gt;Connecting systems does not automatically make either one enjoyable. The experiment fails to support the hypothesis if players build wherever the prompt tells them and never reconsider their route. It also needs revision if every outpost has one obvious location, or if players describe construction as a tax they pay before returning to the interesting activity.&lt;/p&gt;

&lt;p&gt;A different outcome is possible: players may prefer version A because displaying discoveries matters more to them than extending their range. That would be useful evidence about the audience rather than proof that their preference is wrong.&lt;/p&gt;

&lt;p&gt;The practical question is whether construction creates a choice the player values. More dependencies are not inherently better. Subnautica’s documented relationship between exploration and habitats offers a useful case for asking that question. Before adding more rooms, recipes or destinations, test whether one completed structure gives the player a reason to plan the next expedition differently.&lt;/p&gt;

</description>
      <category>gamedev</category>
    </item>
  </channel>
</rss>
