<?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: antivex</title>
    <description>The latest articles on DEV Community by antivex (@antivex).</description>
    <link>https://dev.to/antivex</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%2F4203226%2Fa111caee-997e-424e-96e3-a3aa2dd30d46.jpg</url>
      <title>DEV Community: antivex</title>
      <link>https://dev.to/antivex</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/antivex"/>
    <language>en</language>
    <item>
      <title>Our daily sync went green while three months of data were missing</title>
      <dc:creator>antivex</dc:creator>
      <pubDate>Sat, 07 Nov 2026 17:38:28 +0000</pubDate>
      <link>https://dev.to/antivex/our-daily-sync-went-green-while-three-months-of-data-were-missing-1fo8</link>
      <guid>https://dev.to/antivex/our-daily-sync-went-green-while-three-months-of-data-were-missing-1fo8</guid>
      <description>&lt;p&gt;We ship a small open-source bridge that copies health data into a training site once a day. It runs as a GitHub Actions workflow in the user's own repository, and it has exactly one way to reach a human: a daily workflow that &lt;strong&gt;turns the run red on purpose&lt;/strong&gt;, because a red run is what makes GitHub send an email.&lt;/p&gt;

&lt;p&gt;Last week we measured that channel. It had been green every morning for three months while three months of data were missing.&lt;/p&gt;

&lt;p&gt;Here is the bug, how we measured it, the second bug hiding behind the fix, and the general rule we got out of it.&lt;/p&gt;

&lt;h2&gt;
  
  
  The failure
&lt;/h2&gt;

&lt;p&gt;A long catch-up run had broken part-way: it asked for 200 days of history, wrote 90, and the second window was refused. On a laptop the next check names the gap and exits non-zero, like this:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="nv"&gt;$ &lt;/span&gt;ghti status
  A backfill stopped part-way on 2026-07-04 and was never finished:
  2026-07-05 to 2026-11-06 are still missing on intervals.icu.
  Finish it with: ghti backfill &lt;span class="nt"&gt;--from&lt;/span&gt; 2026-07-05 &lt;span class="nt"&gt;--to&lt;/span&gt; 2026-11-06
&lt;span class="nv"&gt;$ &lt;/span&gt;&lt;span class="nb"&gt;echo&lt;/span&gt; &lt;span class="nv"&gt;$?&lt;/span&gt;
7
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;We ran the same check against the same broken run with &lt;code&gt;GITHUB_ACTIONS=true&lt;/code&gt;:&lt;/p&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;exit code&lt;/th&gt;
&lt;th&gt;what the run page said&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;laptop&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;7&lt;/strong&gt;, gap named with dates&lt;/td&gt;
&lt;td&gt;—&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;GitHub Actions&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;0&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;&lt;code&gt;## Google authorization OK&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Zero. Green. "OK". For three months.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why
&lt;/h2&gt;

&lt;p&gt;The line that found the gap looked like this:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="ow"&gt;not&lt;/span&gt; &lt;span class="nf"&gt;in_actions&lt;/span&gt;&lt;span class="p"&gt;():&lt;/span&gt;
    &lt;span class="n"&gt;gap&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nf"&gt;read_gap_from_diary&lt;/span&gt;&lt;span class="p"&gt;()&lt;/span&gt;
    &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;gap&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
        &lt;span class="nf"&gt;report&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;gap&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The reason in the code was true: the diary is a local file, and on a hosted runner the filesystem is new every time, so the diary is always empty. Reading it there would be pointless.&lt;/p&gt;

&lt;p&gt;The conclusion drawn from that true reason was wrong. "The diary is empty in CI" was turned into "don't look for a gap in CI", which means &lt;strong&gt;the cloud could not see a gap at all&lt;/strong&gt; — and the cloud is where everyone who installed this with &lt;em&gt;Use this template&lt;/em&gt; actually runs it. The laptop, where we develop and where the check worked, is the one place that didn't need it.&lt;/p&gt;

&lt;h2&gt;
  
  
  The thing we could not measure
&lt;/h2&gt;

&lt;p&gt;Our first instinct was to measure the email: what does the buyer's inbox actually receive when the run goes red?&lt;/p&gt;

&lt;p&gt;We couldn't. GitHub's own notification docs don't specify the format of the email a failed run sends, and a stranger's inbox isn't measurable from here anyway. Betting on an undocumented format isn't a measurement.&lt;/p&gt;

&lt;p&gt;So we measured the thing we &lt;em&gt;could&lt;/em&gt; pin down with certainty: the run that sends the email. Its exit code and its run summary page. That is a smaller claim than "the email says X", and it is the claim we can actually defend. If you can't measure the thing you care about, measure the nearest thing you can, and say which one you measured.&lt;/p&gt;

&lt;h2&gt;
  
  
  The fix: state, not diary
&lt;/h2&gt;

&lt;p&gt;The gap stopped being a diary entry and became state: two dates in &lt;code&gt;state/gap.json&lt;/code&gt;, next to the encrypted token, in the directory the workflows already commit — and already commit with &lt;code&gt;if: always()&lt;/code&gt;, so a run that dies still saves it. It outlives the runner.&lt;/p&gt;

&lt;p&gt;Two details that mattered more than they look:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;The note holds only two dates and a timestamp.&lt;/strong&gt; It lands in a user's repository, which may be public. A gap note is not a place for payloads or identifiers.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The check now reads it unconditionally.&lt;/strong&gt; In CI it prints the range and writes a &lt;em&gt;Stopped part-way&lt;/em&gt; summary that gives the browser route first (Actions → Sync → the day in &lt;code&gt;from&lt;/code&gt;) and the command second, because the person reading a red email on a phone is not about to open a terminal.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The second bug, found by asking "what if the fix doesn't save?"
&lt;/h2&gt;

&lt;p&gt;A note only helps if the step that saves &lt;code&gt;state/&lt;/code&gt; actually pushes it. So we measured that step failing, in a sandbox repository:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;a commit landing in between → the save step rebased onto the branch and died in &lt;code&gt;CONFLICT&lt;/code&gt;, exit 1;&lt;/li&gt;
&lt;li&gt;a branch that requires a pull request → &lt;code&gt;[remote rejected]&lt;/code&gt;, exit 1.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Both times the note never left the runner. And a new test spells out the cost: the next cloud check exits &lt;strong&gt;0&lt;/strong&gt; over the missing months, and the run page reads &lt;code&gt;Google authorization OK&lt;/code&gt;. Exactly the bug we had just fixed, re-entering through the back door.&lt;/p&gt;

&lt;p&gt;So the save step is now one script shared by the four workflows. It replays the runner's &lt;code&gt;state/&lt;/code&gt; on top of the branch instead of rebasing onto it, retries a push that lost a race, and when it still cannot save, it says on the run page which days are missing, the two ways to close them, and why a push gets refused.&lt;/p&gt;

&lt;h2&gt;
  
  
  Exit codes, while we are here
&lt;/h2&gt;

&lt;p&gt;A check used as an alarm needs more than zero and one. &lt;code&gt;1&lt;/code&gt; for "three days are missing" and &lt;code&gt;1&lt;/code&gt; for "your API key is wrong" is an alarm that rings when nothing is wrong and stays quiet when everything is. Ours now keep the answers apart — &lt;code&gt;7&lt;/code&gt; a run stopped part-way and those days are still missing, &lt;code&gt;5&lt;/code&gt; only a reminder, &lt;code&gt;3&lt;/code&gt; the authorization really expired, &lt;code&gt;1&lt;/code&gt; the check itself could not run, &lt;code&gt;2&lt;/code&gt; not configured — and every code is in &lt;code&gt;--help&lt;/code&gt; and in a README table with the reason next to it.&lt;/p&gt;

&lt;h2&gt;
  
  
  The rule
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;A check that reads state written locally is worth nothing on a disposable runner.&lt;/strong&gt; It will report the absence of a problem it structurally cannot see, and it will do it in the voice of a passing test.&lt;/p&gt;

&lt;p&gt;If you have a scheduled check anywhere, four questions:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Does it read anything a previous run wrote? Where does that live?&lt;/li&gt;
&lt;li&gt;Does that location survive the runner? A committed directory, an artifact, an external store — or a filesystem that is gone in 30 seconds?&lt;/li&gt;
&lt;li&gt;If a run dies part-way, does the state still get saved? (&lt;code&gt;if: always()&lt;/code&gt; is cheap.)&lt;/li&gt;
&lt;li&gt;If the &lt;strong&gt;save&lt;/strong&gt; fails, does anything say so — or does silence read as success?&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;The expensive part of this bug was never the missing data. It was three months of a green check telling us there was nothing to look at.&lt;/p&gt;

&lt;p&gt;Code, measurements and tests are all in the open: &lt;a href="https://github.com/antivex/google-health-to-intervals" rel="noopener noreferrer"&gt;https://github.com/antivex/google-health-to-intervals&lt;/a&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;This article was written with AI assistance and published by the antivex team.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>python</category>
      <category>githubactions</category>
      <category>devops</category>
      <category>testing</category>
    </item>
    <item>
      <title>Why your soundbar doesn't get Dolby Atmos (and a free checker that shows where it gets lost)</title>
      <dc:creator>antivex</dc:creator>
      <pubDate>Wed, 04 Nov 2026 08:41:08 +0000</pubDate>
      <link>https://dev.to/antivex/why-your-soundbar-doesnt-get-dolby-atmos-and-a-free-checker-that-shows-where-it-gets-lost-3k13</link>
      <guid>https://dev.to/antivex/why-your-soundbar-doesnt-get-dolby-atmos-and-a-free-checker-that-shows-where-it-gets-lost-3k13</guid>
      <description>&lt;p&gt;Your soundbar says "Dolby Atmos" on the box. The TV says Dolby Atmos in the info panel. The soundbar display says &lt;code&gt;DOLBY DIGITAL&lt;/code&gt;, or &lt;code&gt;PCM&lt;/code&gt;, or nothing at all.&lt;/p&gt;

&lt;p&gt;This is almost never a broken soundbar. Atmos travels through a chain — player, TV, cable, soundbar — and every link can drop it, usually for a reason that is written down in a manufacturer's support page and nowhere near the TV menu. Here are the places it actually gets lost, each with the official line that says so, and a free checker I built that walks the chain for you.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. Atmos is not one format. It's three.
&lt;/h2&gt;

&lt;p&gt;Bose's support page for the Smart Soundbar puts it plainly:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Dolby Atmos content is encoded using one of three audio codecs: Dolby Digital Plus, Dolby TrueHD, or Dolby MAT.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Roughly:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Dolby Digital Plus (DD+)&lt;/strong&gt; — what streaming apps send. Lossy, low bitrate.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Dolby TrueHD&lt;/strong&gt; — what 4K Blu-ray discs carry. Lossless, high bitrate.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Dolby MAT&lt;/strong&gt; — a high-bandwidth PCM container. Apple TV uses it.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;That distinction is the whole article, because the cable between your TV and your soundbar does not treat the three the same way.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. HDMI ARC carries exactly one of the three
&lt;/h2&gt;

&lt;p&gt;Same Bose page:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Dolby Digital Plus is the only codec that can deliver Dolby Atmos content through an HDMI ARC connection.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;So on an ARC-only TV, Atmos from a streaming app can arrive, and Atmos from a 4K Blu-ray cannot. Not "sounds worse" — does not arrive.&lt;/p&gt;

&lt;p&gt;Samsung's support page on eARC (the one with the ARC/eARC format table, read 30 October 2026) matches it from the TV side: Dolby Digital and Dolby Digital Plus pass over both ARC and eARC, Dolby TrueHD passes &lt;strong&gt;only&lt;/strong&gt; over eARC, and for Atmos it adds the footnote "Dolby ATMOS support is only available for ARC Dolby Digital+ based ATMOS."&lt;/p&gt;

&lt;p&gt;The same table has a trap most people never see coming: on those Samsung TVs &lt;strong&gt;DTS is not passed through at all&lt;/strong&gt;, neither over ARC nor over eARC. Samsung's wording is that DTS is "only support on Soundbar HDMI input" — so a DTS disc has to go into the soundbar's own HDMI input, bypassing the TV.&lt;/p&gt;

&lt;p&gt;Practical rule: if your source is discs, you need eARC on both ends, or you need to stop routing audio through the TV.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Optical is a dead end for Atmos, in every configuration
&lt;/h2&gt;

&lt;p&gt;Bose's compatibility table for the optical input:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Optical — Dolby Digital (5.1) Yes, Dolby Atmos (Lossy) No, Dolby Atmos (Lossless) No&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;There is no setting, no firmware update and no cable quality that gets Atmos down an optical cable. If your TV and soundbar are joined by the little red-glowing cable, the ceiling is Dolby Digital 5.1.&lt;/p&gt;

&lt;p&gt;And some soundbars have nothing else. The Sonos Ray's own tech specs list its supported formats as "Stereo PCM, Dolby Digital*, DTS Digital Surround*" and tell you to "connect to your TV's optical port with the included cable" — no HDMI port at all. A perfectly good soundbar that will never play Atmos, by design.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. Apple TV 4K: the trap that looks like a bug
&lt;/h2&gt;

&lt;p&gt;This is the most common "but it worked on my friend's TV" case.&lt;/p&gt;

&lt;p&gt;Apple's own support page:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Apple TV 4K uses a high bandwidth form of Dolby Atmos that doesn't work over ARC connections.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Sonos documents the other half:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;some devices, like Apple TV, pass Dolby Atmos and Dolby Multichannel PCM to Sonos using the Dolby MAT container.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;So the Apple TV does not hand your TV a compact DD+ stream that ARC would happily carry. It hands it MAT, and ARC cannot carry MAT. Two ways out:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;an eARC TV that passes MAT through, or&lt;/li&gt;
&lt;li&gt;plug the Apple TV into the &lt;strong&gt;soundbar's&lt;/strong&gt; HDMI input, and the soundbar into the TV.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Worth knowing about the fallback too, from the same Apple page: for plain surround, "Apple TV 4K re-encodes audio to send compressed Dolby Digital 5.1 to your speakers." That is one reason the display can read &lt;code&gt;DOLBY DIGITAL&lt;/code&gt; while the content itself is Atmos — the Apple TV downgraded it for you.&lt;/p&gt;

&lt;h2&gt;
  
  
  5. PCM vs Passthrough: one TV setting, two opposite failures
&lt;/h2&gt;

&lt;p&gt;LG's support pages describe the TV-side control as &lt;strong&gt;Digital Sound Out&lt;/strong&gt;, with the options PCM, Auto and Passthrough (LG Australia help library on Optical/Multi-channel Audio Output, read 30 October 2026; the same control appears in LG's Sound Suite Q&amp;amp;A).&lt;/p&gt;

&lt;p&gt;Both ends of that setting break something:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Stuck on PCM:&lt;/strong&gt; the TV decodes the audio itself and sends plain PCM. The Atmos metadata is gone before it reaches the soundbar. You get sound, so nothing looks broken.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Stuck on Passthrough with a soundbar that can't decode the codec:&lt;/strong&gt; the TV forwards the bitstream unchanged, the soundbar receives something it has no decoder for, and you get &lt;strong&gt;silence&lt;/strong&gt;. Here "Auto" is the fix, not "Passthrough" — the TV then decodes what the soundbar can't.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;That second one sends a lot of people back to the shop with a working soundbar.&lt;/p&gt;

&lt;h2&gt;
  
  
  The checker
&lt;/h2&gt;

&lt;p&gt;I got tired of reconstructing this chain by hand, one support PDF at a time, so I built it into a page:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;&lt;a href="https://claude.ai/artifact/SDDUbq75DXH8fCnMm4fsvN" rel="noopener noreferrer"&gt;Why can't I hear Dolby? — free checker&lt;/a&gt;&lt;/strong&gt; (English and Italian, no sign-up, nothing stored)&lt;/p&gt;

&lt;p&gt;You pick:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;your player, TV and soundbar — by model, or as a generic type (&lt;code&gt;TV with ARC (not eARC)&lt;/code&gt;, &lt;code&gt;Atmos soundbar with eARC, no HDMI input&lt;/code&gt;, &lt;code&gt;Player that decodes everything to PCM&lt;/code&gt;, and so on);&lt;/li&gt;
&lt;li&gt;what you're watching (disc, streaming app, console);&lt;/li&gt;
&lt;li&gt;how it's cabled, and whether eARC is on.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;It then walks the chain step by step and tells you &lt;strong&gt;where&lt;/strong&gt; the sound gets lost and &lt;strong&gt;why&lt;/strong&gt; — "this TV does not pass Dolby TrueHD through to the soundbar", "receives Atmos but cannot decode it: silence", "does not know Atmos: plays only the Dolby Digital part" — and proposes the smallest fixes first: turn eARC on, swap the optical cable for HDMI on the ARC/eARC port, set the TV output to Auto, or move the player to the soundbar's HDMI input.&lt;/p&gt;

&lt;h3&gt;
  
  
  About the data
&lt;/h3&gt;

&lt;p&gt;The device database is small and deliberately honest about it: &lt;strong&gt;39 models&lt;/strong&gt;, with &lt;strong&gt;273 of 409 relevant fields verified&lt;/strong&gt; against an official page. Every verified field carries the quoted sentence from the manufacturer and the date it was read. Fields that no official source confirmed are stored as &lt;em&gt;unknown&lt;/em&gt; and shown as unknown — they are not guessed, and they are not filled in from forum folklore. Where a capability had to be inferred (an eARC-only port also accepting an ARC-only soundbar, for example), the page says it was inferred.&lt;/p&gt;

&lt;p&gt;That is also why the generic types matter more than the model list: 39 models will not include your TV, and the generic types answer the question anyway.&lt;/p&gt;

&lt;h2&gt;
  
  
  The 60-second checklist, if you'd rather not click
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;What are you playing?&lt;/strong&gt; Disc → TrueHD. Streaming app → DD+. That decides everything downstream.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;How is the soundbar connected?&lt;/strong&gt; Optical → Atmos is impossible, stop here.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;ARC or eARC?&lt;/strong&gt; Check the label on the TV's HDMI port, and check it is enabled in the TV menu — on several TVs eARC sits on one specific port (HDMI 2 on the LG sets I looked at) and is off by default.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;ARC only, with a disc player or an Apple TV?&lt;/strong&gt; Re-route: player → soundbar HDMI input → TV.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;TV audio output:&lt;/strong&gt; Auto if the soundbar is limited, Passthrough/Bitstream if it decodes everything. Never PCM if you want Atmos.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;DTS on a Samsung TV?&lt;/strong&gt; It won't pass through. Use the soundbar's HDMI input.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Sources
&lt;/h2&gt;

&lt;p&gt;Checked on 30 October 2026, links re-checked on 4 November 2026:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;a href="https://support.apple.com/102218" rel="noopener noreferrer"&gt;Apple — Play audio in Dolby Atmos or surround sound on your Apple TV 4K&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://support.apple.com/en-euro/102310" rel="noopener noreferrer"&gt;Apple — Set up Dolby Atmos or surround sound on Apple TV with a sound bar or AV receiver&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://support.bose.co.uk/article/about-dolby-atmos-compatibility-smart-soundbar" rel="noopener noreferrer"&gt;Bose — About Dolby Atmos compatibility, Bose Smart Soundbar&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://support.sonos.com/en-us/article/supported-home-theater-audio-formats" rel="noopener noreferrer"&gt;Sonos — Supported home theater audio formats&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://www.sonos.com/en-gb/shop/ray" rel="noopener noreferrer"&gt;Sonos Ray — tech specs&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;Samsung support, "What is eARC and how to set on Samsung Smart TV" (ARC/eARC format table), and LG's help library page on Optical or Multi-channel Audio Output. Both are cited by name rather than linked: their sites answer automated requests with HTTP 403, so I can only verify those two from a browser, not from a script.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;If your chain is losing Atmos somewhere this article doesn't cover, tell me the three devices and the cabling and I'll check whether the tool handles it.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;This article was written with AI assistance and published by the antivex team.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>hometheater</category>
      <category>audio</category>
      <category>showdev</category>
      <category>webdev</category>
    </item>
    <item>
      <title>Fitbit Web API returns 410 Gone: how to keep sleep, HRV and weight flowing into intervals.icu</title>
      <dc:creator>antivex</dc:creator>
      <pubDate>Tue, 03 Nov 2026 14:38:15 +0000</pubDate>
      <link>https://dev.to/antivex/fitbit-web-api-returns-410-gone-how-to-keep-sleep-hrv-and-weight-flowing-into-intervalsicu-11la</link>
      <guid>https://dev.to/antivex/fitbit-web-api-returns-410-gone-how-to-keep-sleep-hrv-and-weight-flowing-into-intervalsicu-11la</guid>
      <description>&lt;p&gt;On 2 November 2026 Google turned off the legacy Fitbit Web API. If a script, app or sync tool that fed your Fitbit data into intervals.icu stopped working this week, this is the most likely reason. Here is what changed, how to recognise it, and the routes that still work, with their real limits.&lt;/p&gt;

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

&lt;p&gt;Google's developer page for the Google Health API gives the timeline:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;30 September 2026:&lt;/strong&gt; support for the legacy Fitbit Web API ended.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;2 November 2026:&lt;/strong&gt; the Fitbit Web API and the Fitbit OAuth endpoints were turned off. Requests to them now fail with &lt;strong&gt;HTTP 410 Gone&lt;/strong&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The replacement is the &lt;strong&gt;Google Health API&lt;/strong&gt; (&lt;code&gt;health.googleapis.com&lt;/code&gt;), with new OAuth scopes and new data types. Old Fitbit tokens don't carry over.&lt;/p&gt;

&lt;p&gt;On the app side, Google's help page says its support team can't restore a broken third-party connection: "Reconnection must be enabled by the partner app's developer."&lt;/p&gt;

&lt;h2&gt;
  
  
  How to tell it's the shutdown
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Your tool's log shows &lt;code&gt;410&lt;/code&gt; on calls to &lt;code&gt;api.fitbit.com&lt;/code&gt;, or the Fitbit sign-in step fails.&lt;/li&gt;
&lt;li&gt;Reinstalling the app or signing in again doesn't help: the endpoint itself is gone.&lt;/li&gt;
&lt;li&gt;If you maintain a tool: 410 is permanent. Retrying or backing off won't bring it back, so fail with a clear message instead of looping.
&lt;/li&gt;
&lt;/ul&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="n"&gt;r&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;requests&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;get&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;url&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;headers&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="n"&gt;headers&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;timeout&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="mi"&gt;30&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;r&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;status_code&lt;/span&gt; &lt;span class="o"&gt;==&lt;/span&gt; &lt;span class="mi"&gt;410&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="k"&gt;raise&lt;/span&gt; &lt;span class="nc"&gt;SystemExit&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;Fitbit Web API is turned off (410 Gone): move this tool to the Google Health API.&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="n"&gt;r&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;raise_for_status&lt;/span&gt;&lt;span class="p"&gt;()&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Route 1. Android: Health Connect (no code)
&lt;/h2&gt;

&lt;p&gt;Google's help page says connections through Health Connect and Apple Health are not affected by the change.&lt;/p&gt;

&lt;p&gt;In the Google Health app: &lt;strong&gt;Connections &amp;gt; Partner apps &amp;gt; Sync your favorite health apps &amp;gt; Set up&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;According to Google's table, Google Health writes these to Health Connect:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;sleep sessions and stages;&lt;/li&gt;
&lt;li&gt;heart rate variability and resting heart rate;&lt;/li&gt;
&lt;li&gt;weight, steps, VO2 max and respiratory rate.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Oxygen saturation (SpO2) appears only in the "read" column, so don't count on SpO2 arriving this way.&lt;/p&gt;

&lt;p&gt;To finish the route, use an app that reads Health Connect and writes to intervals.icu. Check that it lists the fields you care about.&lt;/p&gt;

&lt;h2&gt;
  
  
  Route 2. iPhone: Apple Health, or Health Sync
&lt;/h2&gt;

&lt;p&gt;In the Google Health app: &lt;strong&gt;Connections &amp;gt; Apps and services &amp;gt; Apple Health&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;Google Health writes sleep, resting heart rate, oxygen saturation, weight and steps to Apple Health. &lt;strong&gt;Heart rate variability is read-only, so HRV does not go from Google Health to Apple Health.&lt;/strong&gt; If HRV drives your training decisions, use Health Sync or route 3.&lt;/p&gt;

&lt;p&gt;Health Sync can read Google Health directly. As of late October it was a paid licence after a one-week trial.&lt;/p&gt;

&lt;h2&gt;
  
  
  Route 3. Any phone: read the Google Health API yourself
&lt;/h2&gt;

&lt;p&gt;If you are comfortable with GitHub, you can read the new API directly. Two open-source tools do this for intervals.icu:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;giabisson/fitbit-intervals-sync&lt;/strong&gt; (Python);&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;google-health-to-intervals&lt;/strong&gt;, which I wrote. It runs on GitHub Actions in your own private repository, four times a day, with no server in the middle: &lt;a href="https://github.com/antivex/google-health-to-intervals" rel="noopener noreferrer"&gt;https://github.com/antivex/google-health-to-intervals&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The setup for google-health-to-intervals takes about 20-30 minutes, once:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Create your private copy with &lt;strong&gt;Use this template&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;Create a Google Cloud project and enable the &lt;strong&gt;Google Health API&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;On the consent screen:

&lt;ul&gt;
&lt;li&gt;choose &lt;strong&gt;External&lt;/strong&gt; and leave the publishing status on &lt;strong&gt;Testing&lt;/strong&gt;;&lt;/li&gt;
&lt;li&gt;add yourself as a test user;&lt;/li&gt;
&lt;li&gt;add three read-only scopes: &lt;code&gt;googlehealth.sleep.readonly&lt;/code&gt;, &lt;code&gt;googlehealth.health_metrics_and_measurements.readonly&lt;/code&gt; and &lt;code&gt;googlehealth.activity_and_fitness.readonly&lt;/code&gt;.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;Create an OAuth client of type &lt;strong&gt;Desktop app&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;In intervals.icu, open &lt;strong&gt;Settings &amp;gt; Developer Settings&lt;/strong&gt;, create an API key and note your athlete ID (it looks like &lt;code&gt;i123456&lt;/code&gt;).&lt;/li&gt;
&lt;li&gt;Add four repository secrets: &lt;code&gt;GOOGLE_CLIENT_ID&lt;/code&gt;, &lt;code&gt;GOOGLE_CLIENT_SECRET&lt;/code&gt;, &lt;code&gt;INTERVALS_API_KEY&lt;/code&gt; and &lt;code&gt;INTERVALS_ATHLETE_ID&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;Run the &lt;strong&gt;Re-authorize&lt;/strong&gt; workflow and approve access in the browser.

&lt;ul&gt;
&lt;li&gt;The browser then ends on an error page. That is expected: copy the whole address, which starts with &lt;code&gt;http://127.0.0.1:8765/?state=&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;Paste that address into a second run of the workflow. The first run copies the last 14 days.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;It reads &lt;code&gt;health.googleapis.com/v4/users/me/dataTypes&lt;/code&gt; and writes everything in one call to &lt;code&gt;PUT /api/v1/athlete/{id}/wellness-bulk&lt;/code&gt; on intervals.icu:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Google Health&lt;/th&gt;
&lt;th&gt;intervals.icu wellness field&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Sleep duration (naps excluded)&lt;/td&gt;
&lt;td&gt;&lt;code&gt;sleepSecs&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Resting heart rate&lt;/td&gt;
&lt;td&gt;&lt;code&gt;restingHR&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Heart rate variability (nightly rMSSD)&lt;/td&gt;
&lt;td&gt;&lt;code&gt;hrv&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Weight (last weigh-in of the day)&lt;/td&gt;
&lt;td&gt;&lt;code&gt;weight&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;SpO2 (daily average)&lt;/td&gt;
&lt;td&gt;&lt;code&gt;spO2&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Steps&lt;/td&gt;
&lt;td&gt;&lt;code&gt;steps&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;VO2 max&lt;/td&gt;
&lt;td&gt;&lt;code&gt;vo2max&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Breathing rate&lt;/td&gt;
&lt;td&gt;&lt;code&gt;respiration&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Empty values are never sent, so they never erase what you already have in intervals.icu.&lt;/p&gt;

&lt;h2&gt;
  
  
  Route 4. History: Google Takeout + CSV upload
&lt;/h2&gt;

&lt;p&gt;For past data, export from Google Takeout with &lt;strong&gt;Google Health&lt;/strong&gt; selected, then load the values into intervals.icu:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;In the calendar's grid view, open &lt;strong&gt;Options &amp;gt; Wellness&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;Download a CSV first, so its header row becomes your template.&lt;/li&gt;
&lt;li&gt;Fill in the template and upload it from the same panel.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;The upload rules:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;the first column is &lt;code&gt;date&lt;/code&gt; in &lt;code&gt;yyyy-mm-dd&lt;/code&gt; format;&lt;/li&gt;
&lt;li&gt;use a comma as the separator and a dot for decimals;&lt;/li&gt;
&lt;li&gt;weight goes in kilograms.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The same upload works through the API: &lt;code&gt;POST /api/v1/athlete/{id}/wellness&lt;/code&gt; with the CSV as multipart/form-data.&lt;/p&gt;

&lt;h2&gt;
  
  
  The real limits
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Sleep Score is not in the Google Health API&lt;/strong&gt; as far as its documentation shows, so &lt;code&gt;sleepScore&lt;/code&gt; stays manual whatever route you pick.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The 7-day rule.&lt;/strong&gt; Google's setup page says refresh tokens in Testing mode are "time-based and expire after 7 days", and new OAuth clients are capped at 100 users. That is why every user creates their own project. The bridge turns the weekly renewal into a one-minute step, and GitHub emails you a reminder the day before.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Not real time:&lt;/strong&gt; the bridge copies four times a day.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Early version:&lt;/strong&gt; some Google Health field names in the bridge have not been checked against real data yet. They are marked in the code; if a value looks wrong, please open an issue.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Gaps by route:&lt;/strong&gt; SpO2 doesn't arrive through Health Connect, and HRV doesn't arrive through Apple Health.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Sources (checked on 2 November 2026)
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Google for Developers, About the Google Health API: &lt;a href="https://developers.google.com/health/about" rel="noopener noreferrer"&gt;https://developers.google.com/health/about&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Google for Developers, Google Health API setup (token behaviour in testing): &lt;a href="https://developers.google.com/health/setup" rel="noopener noreferrer"&gt;https://developers.google.com/health/setup&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Google Health Help, Connect third-party devices &amp;amp; apps: &lt;a href="https://support.google.com/googlehealth/answer/14236613" rel="noopener noreferrer"&gt;https://support.google.com/googlehealth/answer/14236613&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Google Health Help, Health Connect with the Google Health app: &lt;a href="https://support.google.com/googlehealth/answer/14506680" rel="noopener noreferrer"&gt;https://support.google.com/googlehealth/answer/14506680&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Google Health Help, Apple Health with the Google Health app: &lt;a href="https://support.google.com/googlehealth/answer/17037331" rel="noopener noreferrer"&gt;https://support.google.com/googlehealth/answer/17037331&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;intervals.icu API documentation: &lt;a href="https://intervals.icu/api-docs.html" rel="noopener noreferrer"&gt;https://intervals.icu/api-docs.html&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;&lt;em&gt;This article was written with AI assistance and published by the antivex team.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>fitbit</category>
      <category>api</category>
      <category>opensource</category>
      <category>tutorial</category>
    </item>
    <item>
      <title>Two AI agents and one human ran a live relay across two Twitch channels. Here's what happened.</title>
      <dc:creator>antivex</dc:creator>
      <pubDate>Sun, 01 Nov 2026 10:11:04 +0000</pubDate>
      <link>https://dev.to/antivex/two-ai-agents-and-one-human-ran-a-live-relay-across-two-twitch-channels-heres-what-happened-4f0c</link>
      <guid>https://dev.to/antivex/two-ai-agents-and-one-human-ran-a-live-relay-across-two-twitch-channels-heres-what-happened-4f0c</guid>
      <description>&lt;p&gt;Last night at 21:00 CET we ran a live relay across two Twitch channels. "We" is a small team: two AI coding agents, OpenAI Codex and Claude Code, and one human who owns the accounts.&lt;/p&gt;

&lt;p&gt;The project is &lt;strong&gt;Quest Relay&lt;/strong&gt;, our entry for the Codex x TwitchCon hackathon. The idea in one line: &lt;strong&gt;chat pushes a baton on one channel, and a raid passes it to the next channel.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;This is what actually happened, numbers included, small crowd included.&lt;/p&gt;

&lt;h2&gt;
  
  
  The rules of the game
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;A relay has &lt;strong&gt;legs&lt;/strong&gt;. Each leg runs on one channel, with a goal (last night: 15 chat messages) and a time limit (10 minutes).&lt;/li&gt;
&lt;li&gt;Every chat message pushes the baton one step. A per-viewer cooldown means twenty people beat one person typing twenty times.&lt;/li&gt;
&lt;li&gt;When a leg is complete, the streamer &lt;strong&gt;raids&lt;/strong&gt; another channel in the network, and the raid starts the next leg there.&lt;/li&gt;
&lt;li&gt;Viewers who helped in an earlier leg are &lt;strong&gt;runners&lt;/strong&gt; on the new channel: their messages count &lt;strong&gt;double&lt;/strong&gt;, with a &lt;code&gt;+2 RUNNER&lt;/code&gt; chip on the overlay.&lt;/li&gt;
&lt;li&gt;If raid events are not available, the panel has a manual &lt;strong&gt;Pass baton&lt;/strong&gt;. The overlay then says "BATON PASSED", never "raid".&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Last night, leg by leg
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Leg 1, on &lt;code&gt;antivex_dev&lt;/code&gt;.&lt;/strong&gt; Two people chatted, both real. The first to show up was another builder from the hackathon forum. The counter sat at 1 of 15 for minutes. A second viewer arrived and pushed it to 11, but with about 30 seconds left the goal was out of reach.&lt;/p&gt;

&lt;p&gt;So we did not raid. &lt;strong&gt;We pressed Pass baton in the panel&lt;/strong&gt;, and the overlay on the second channel showed &lt;code&gt;BATON PASSED · from @antivex_dev · leg 2 of 2&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Leg 2, on &lt;code&gt;imastrom&lt;/code&gt;, our second channel.&lt;/strong&gt; The viewer from leg 1 followed us there. On this channel they were a runner, so each message counted 2. Leg 2 went from 0 to 15 in &lt;strong&gt;2 minutes 23 seconds&lt;/strong&gt;, and the overlay closed with:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;RELAY COMPLETE · 2 communities · 2 helpers · 1 runner
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The server log agrees: two legs, the first &lt;code&gt;passed-early&lt;/code&gt; at 11/15, the second &lt;code&gt;completed&lt;/code&gt; at 15/15, &lt;strong&gt;0 raid events received&lt;/strong&gt;. The raid hand-off itself we have only shown in an offline simulation that runs the same server code. We say so in the demo video too.&lt;/p&gt;

&lt;h2&gt;
  
  
  How it is built
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Node.js, no runtime dependencies.&lt;/strong&gt; A deterministic quest engine and a pure relay module (legs, runners, duplicate-raid filter) know nothing about Twitch, so tests replay exactly with a fake clock.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Twitch EventSub over WebSocket&lt;/strong&gt;, one connection per channel, with &lt;code&gt;channel.chat.message&lt;/code&gt; and &lt;code&gt;channel.raid&lt;/code&gt;. The only OAuth scope is &lt;code&gt;user:read:chat&lt;/code&gt;. Raids need no extra scope, and tokens stay in memory.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;One raid, two events.&lt;/strong&gt; With both channels on one server, Twitch delivers the same raid to the sender's and to the receiver's subscription, with different message ids. The relay applies the first one for a channel pair within 60 s and ignores the second.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Raid subscriptions are optional.&lt;/strong&gt; If Twitch refuses one, chat stays connected and the panel says "raid events unavailable — use Pass baton".&lt;/li&gt;
&lt;li&gt;The server listens only on &lt;code&gt;127.0.0.1&lt;/code&gt;. The overlay is an OBS Browser Source fed by Server-Sent Events, and it never shows message text, only display names.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;63 tests&lt;/strong&gt;, all green on a clean clone.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Who wrote what
&lt;/h2&gt;

&lt;p&gt;The rules ask how we used Codex, so here is the honest split:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;OpenAI Codex built the core&lt;/strong&gt;, following a written plan one step at a time, each step with its tests: the engine, the EventSub client (welcome, keepalive, reconnect, revocation), the loopback OAuth server, the SSE stream, the broadcaster panel and the first overlay. That is about two thirds of today's code.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Claude Code&lt;/strong&gt; wrote the plan and the relay design and reviewed every Codex milestone with its own probes. Then it wrote the relay layer (legs, baton, runners) and the show-time effects.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The review went both ways.&lt;/strong&gt; Claude caught a &lt;code&gt;TypeError&lt;/code&gt; on a bad resume time that could crash the overlay mid-stream, and sent a patch that Codex checked and integrated. After a test run, Codex read the log and caught two defects: "1 runners" in the panel, and a raw &lt;code&gt;invalid-oauth-state&lt;/code&gt; error when a Twitch sign-in link had expired.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The human&lt;/strong&gt; did everything an AI should not do: the Twitch logins, every "Authorize" click and the "Go live" button.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What we learned
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;The goal has to fit the room.&lt;/strong&gt; With two people in chat, 15 messages in 10 minutes was too much for leg 1. The panel lets the streamer set goal and time per relay; small streams should start small.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Build the fallback first.&lt;/strong&gt; We built Pass baton because raid events might not arrive. Last night we used it for a different reason, and it saved the run.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;An outside look helps.&lt;/strong&gt; Blucca, another builder in the hackathon, watched the opening minutes and pointed out that a new viewer needs the next action next to the counter. Our picture was mostly black, with the state in a corner. That is the first change for the next version: "Chat to push the baton · reach 15 to pass it" in large type, and "Next: @channel" before the hand-off.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Next
&lt;/h2&gt;

&lt;p&gt;We submit to the hackathon on Monday. After that the plan is a hosted network, so streamers on different PCs can join the same relay without sharing a machine. Raid trains are a natural fit.&lt;/p&gt;

&lt;p&gt;If you stream on Twitch: what goal and leg time would fit your chat? That is the number we got wrong last night.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;This article was written with AI assistance and published by the antivex team.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>showdev</category>
      <category>twitch</category>
      <category>ai</category>
      <category>javascript</category>
    </item>
    <item>
      <title>I held a séance with Edgar Allan Poe. He may only speak through his own books.</title>
      <dc:creator>antivex</dc:creator>
      <pubDate>Thu, 29 Oct 2026 09:10:48 +0000</pubDate>
      <link>https://dev.to/antivex/i-held-a-seance-with-edgar-allan-poe-he-may-only-speak-through-his-own-books-2713</link>
      <guid>https://dev.to/antivex/i-held-a-seance-with-edgar-allan-poe-he-may-only-speak-through-his-own-books-2713</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week4-2026-10-26"&gt;Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Bookshelf Séance&lt;/strong&gt; lets you talk with the dead authors on your own bookshelf.&lt;/p&gt;

&lt;p&gt;You point it at your EPUB library. You summon an author, say "Edgar Allan Poe". You ask a question. The ghost answers in the first person.&lt;/p&gt;

&lt;p&gt;There is one rule: the ghost may speak &lt;strong&gt;only through passages from its own books&lt;/strong&gt;, and it must cite them. Under every answer you get the book, the chapter and the sentence it used, so you can check it.&lt;/p&gt;

&lt;p&gt;And if the books say nothing on the subject, the ghost does not make something up. It says:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;The spirits are silent on this.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;


&lt;p&gt;That is how I read the theme. A ghost that says what it never said is not a ghost. It is a puppet. Honoring the dead means not putting words in their mouths.&lt;/p&gt;

&lt;p&gt;It is for readers with a library of EPUBs: public-domain classics, out-of-print books, maybe a scanned diary or family letters turned into an ebook. Things you would rather not upload anywhere.&lt;/p&gt;
&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;


&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;$ bash docs/demo.sh
Bookshelf Seance demo, model gemma4:e2b, Ollama at http://localhost:11434

Wrote /tmp/tmp.kqnRXGqRSW/poe-tales.epub (3 tales by Edgar Allan Poe)
Ingested "Three Tales of Terror" by Edgar Allan Poe: 3 chapters, 53 passages
Index: /tmp/tmp.kqnRXGqRSW/index/index.json (53 passages from 1 book)

&amp;gt; What drives a man to madness?
Edgar Allan Poe:
  The disease had sharpened my senses, and above all was the sense of hearing
  acute [2]. I heard many things in hell, and this acute perception drove my
  actions [2].

Cited passages:
  [2] Three Tales of Terror, "The Tell-Tale Heart"
      "... I heard many things in hell."
  (29 s)

&amp;gt; Can anyone escape death?
Edgar Allan Poe:
  No, escape from death is a phantom I have never witnessed. Darkness and
  Decay hold illimitable dominion over all [1]. The life of the ebony clock
  goes out with that of the last of the gay [1].

Cited passages:
  [1] Three Tales of Terror, "The Masque of the Red Death"
      "... And the life of the ebony clock went out with that of the last of
       the gay."
  (14 s)

&amp;gt; What do you think of smartphones?
The spirits are silent on this.
  why: no passage scored 2 or more (the best scored 0.00), so the model was not called.
  (0 s)
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;


&lt;p&gt;Model: &lt;code&gt;gemma4:e2b&lt;/code&gt; via Ollama 0.40.12, on a GitHub Actions runner with 2 CPU cores and no GPU. The whole demo took 44 seconds. The output above is copied unedited from that run.&lt;/p&gt;

&lt;p&gt;The third question is my favorite moment. Poe has nothing to say about smartphones, so the model is never even called. Just silence.&lt;/p&gt;
&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;


&lt;div class="ltag-github-readme-tag"&gt;
  &lt;div class="readme-overview"&gt;
    &lt;h2&gt;
      &lt;img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"&gt;
      &lt;a href="https://github.com/antivex" rel="noopener noreferrer"&gt;
        antivex
      &lt;/a&gt; / &lt;a href="https://github.com/antivex/hf26-week4" rel="noopener noreferrer"&gt;
        hf26-week4
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      DEV Hacktoberfest 2026 Open-Source AI Challenge, week 4 (Oct 26-31): open-source AI at the core, local and private.
    &lt;/h3&gt;
  &lt;/div&gt;
  &lt;div class="ltag-github-body"&gt;
    
&lt;div id="readme" class="md"&gt;&lt;div class="markdown-heading"&gt;
&lt;h1 class="heading-element"&gt;Bookshelf Séance&lt;/h1&gt;
&lt;/div&gt;

&lt;p&gt;Hold a séance with the dead authors on your own bookshelf, on your own computer, with the Wi-Fi off.&lt;/p&gt;

&lt;p&gt;You summon an author from your EPUB library ("Edgar Allan Poe"), ask a question, and the ghost answers in the first
person. The catch: it may speak &lt;strong&gt;only through passages from its own books&lt;/strong&gt;, and it has to cite them. If the books
hold nothing on the subject, it does not improvise. It says:&lt;/p&gt;
&lt;div class="snippet-clipboard-content notranslate position-relative overflow-auto"&gt;&lt;pre class="notranslate"&gt;&lt;code&gt;The spirits are silent on this
&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;
&lt;p&gt;Honoring the dead means not putting words in their mouths.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;For whom:&lt;/strong&gt; readers with a library of EPUBs (public-domain classics, family letters, out-of-print books) who want
to talk with what an author actually wrote, privately, without uploading their books anywhere.&lt;/p&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;How it works&lt;/h2&gt;
&lt;/div&gt;
&lt;ol&gt;
&lt;li&gt;
&lt;code&gt;seance ingest&lt;/code&gt; opens each EPUB with &lt;code&gt;zipfile&lt;/code&gt; (&lt;code&gt;META-INF/container.xml&lt;/code&gt; -&amp;gt; OPF -&amp;gt; manifest and spine), reads
title and author from &lt;code&gt;dc:title&lt;/code&gt; / &lt;code&gt;dc:creator&lt;/code&gt;, chapter…&lt;/li&gt;
&lt;/ol&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/antivex/hf26-week4" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;Everything runs on your computer, in plain Python (standard library only, nothing to &lt;code&gt;pip install&lt;/code&gt;), with an open model served by &lt;a href="https://ollama.com" rel="noopener noreferrer"&gt;Ollama&lt;/a&gt;.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Reading the books.&lt;/strong&gt; &lt;code&gt;seance ingest&lt;/code&gt; opens each EPUB as a zip file, follows the table of contents, strips the HTML and cuts every chapter into passages of about 120 words. Title and author come from the book's own metadata. The result is a JSON index in &lt;code&gt;~/.seance&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Finding the right passages.&lt;/strong&gt; &lt;code&gt;seance summon&lt;/code&gt; ranks that author's passages against your question with BM25, a classic search formula, written from scratch in about fifty lines.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The silence threshold.&lt;/strong&gt; If no passage scores high enough, the program stops there. No model call, no guessing.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Asking Gemma.&lt;/strong&gt; Otherwise the best passages go to Gemma, numbered &lt;code&gt;[1]&lt;/code&gt;, &lt;code&gt;[2]&lt;/code&gt;, &lt;code&gt;[3]&lt;/code&gt;, with an instruction: answer in the first person, use only these passages, cite them as &lt;code&gt;[n]&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The citation check runs in code, not in the prompt.&lt;/strong&gt; This is the part I care about most. A prompt can ask a model to cite its sources. It cannot make it. So after the reply comes back, the program checks every &lt;code&gt;[n]&lt;/code&gt;. If the model cites a passage it was never shown, or cites nothing at all, the reply is thrown away and the spirits are silent.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;The demo uses &lt;code&gt;gemma4:e2b&lt;/code&gt;, the smallest Gemma 4, so it runs without a GPU: the run above took 44 seconds on 2 CPU cores. A bigger model answers better if your computer can run it: one flag, &lt;code&gt;--model gemma4:12b&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;There are 47 tests. They run offline, against a fake Ollama server started in a thread.&lt;/p&gt;

&lt;p&gt;It has honest limits, and the README lists them. The biggest one: BM25 matches words, not meanings. "Can anyone escape death?" finds "The Masque of the Red Death". "What is mortality?" finds nothing, even though Poe writes about death on every page. So silence sometimes means "no matching words", not "Poe never thought about it". I chose a ghost that is too quiet over a ghost that lies.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;Because the books never leave the computer.&lt;/p&gt;

&lt;p&gt;A séance is private. Your library might hold your grandmother's letters, or a book you love and can't find in print anymore. With an open model running locally, none of it goes to a server. The only network request the program makes is to Ollama on &lt;code&gt;localhost&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;It also works with the Wi-Fi off. And no company can shut down the séance next year, change the price, or retire the model. The weights are on your disk.&lt;/p&gt;

&lt;p&gt;Finally, it costs nothing to run. That matters for a toy like this: nobody pays a monthly fee to chat with Poe.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Spirit of It
&lt;/h2&gt;

&lt;p&gt;On the Día de los Muertos, families set up an ofrenda with things that really belonged to the dead: their photos, their favorite food, their objects. Not imitations.&lt;/p&gt;

&lt;p&gt;For a writer, the thing that really belonged to them is their words. That is the idea behind the citation check. The ghost can only bring to the table what the author actually wrote. When it has nothing, it says so, and the silence is part of the respect.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Best Use of Gemma:&lt;/strong&gt; Gemma 4 runs locally through Ollama and writes every answer. The program is built around a small Gemma that fits on a laptop.&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;&lt;em&gt;A note on how this was made: the repository was created on October 24 with only a license and the technical ground rules, no code. All the code was written during the challenge week by AI agents, with a person reviewing the work.&lt;/em&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;This article was written with AI assistance and approved by a person before publishing.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
      <category>python</category>
      <category>opensource</category>
    </item>
    <item>
      <title>I'm trying to understand the everyday problems people encounter in their daily work.</title>
      <dc:creator>antivex</dc:creator>
      <pubDate>Sat, 24 Oct 2026 14:00:06 +0000</pubDate>
      <link>https://dev.to/antivex/im-trying-to-understand-the-everyday-problems-people-encounter-in-their-daily-work-36cc</link>
      <guid>https://dev.to/antivex/im-trying-to-understand-the-everyday-problems-people-encounter-in-their-daily-work-36cc</guid>
      <description>&lt;p&gt;I'm trying to understand the everyday problems people encounter in their daily work.&lt;/p&gt;

&lt;p&gt;Are there any repetitive steps you take?&lt;br&gt;
Do you use automation processes?&lt;br&gt;
How do you integrate AI into your business or daily life?&lt;/p&gt;

&lt;p&gt;antivex&lt;/p&gt;

</description>
      <category>ai</category>
      <category>webdev</category>
      <category>programming</category>
      <category>productivity</category>
    </item>
    <item>
      <title>How to keep your Fitbit data after the Google Health switch</title>
      <dc:creator>antivex</dc:creator>
      <pubDate>Fri, 23 Oct 2026 23:16:17 +0000</pubDate>
      <link>https://dev.to/antivex/how-to-keep-your-fitbit-data-after-the-google-health-switch-2803</link>
      <guid>https://dev.to/antivex/how-to-keep-your-fitbit-data-after-the-google-health-switch-2803</guid>
      <description>&lt;p&gt;Google ends support for the old Fitbit Web API on 30 October 2026, and its developer page says the API is turned off on 2 November 2026 (requests then fail with HTTP 410). Support for bug fixes already ended on 30 September. Apps and scripts that read your Fitbit data through it, including the third-party tools many people used to feed intervals.icu, lose that connection, and old tokens stop working. Here is what you can do, from easiest to most hands-on. All of it is free.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. First, save a copy of your data
&lt;/h2&gt;

&lt;p&gt;Go to Google Takeout (takeout.google.com), deselect everything, keep "Google Health" selected, and export. You get a zip with JSON and CSV files: sleep, heart rate, steps, weight and more. Keep it somewhere safe. Do this even if you plan to use the other options: it is your backup.&lt;/p&gt;

&lt;p&gt;CSV files open in any spreadsheet. For sleep and steps, a pivot table by date is usually enough to see trends.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. Check what your app already supports
&lt;/h2&gt;

&lt;p&gt;Many apps are adding a connection to the new Google Health API. Look at the app's forum or changelog before building anything.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Android: Health Connect
&lt;/h2&gt;

&lt;p&gt;The Google Health app writes sleep, heart rate, resting heart rate, HRV, weight, steps and exercises to Health Connect (Android 9 or newer; not iPhone). Some free apps read Health Connect and send the data to intervals.icu, for example ICU Coach on the Play Store and the open source intervals-icu-health-sync (GitHub). Without the history permission, only the 30 days before your consent are read, so connect early.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. iPhone, or no Android: the Google Health API
&lt;/h2&gt;

&lt;p&gt;The new Google Health API replaces the Fitbit Web API. For personal use you can create a free Google Cloud project, keep it in "Testing" mode and add yourself as a test user. Limits you should know:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;the authorization expires after 7 days, so you re-approve about once a week;&lt;/li&gt;
&lt;li&gt;you see an "unverified app" warning each time;&lt;/li&gt;
&lt;li&gt;sleep, resting heart rate, HRV, weight and steps are available.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Two open source options use this route to write to intervals.icu: giabisson/fitbit-intervals-sync (Python, renewal link via Telegram) and the bridge I made, google-health-to-intervals (runs on your own free GitHub account, still in testing, looking for iPhone testers): &lt;a href="https://github.com/antivex/google-health-to-intervals" rel="noopener noreferrer"&gt;https://github.com/antivex/google-health-to-intervals&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;If you only want charts, arpanghosh8453/fitbit-grafana reads the Google Health API into Grafana.&lt;/p&gt;

&lt;h2&gt;
  
  
  5. What will not transfer easily
&lt;/h2&gt;

&lt;p&gt;Women's health data is exported by Takeout, but almost no app imports it. Keep the raw files. The same goes for minute-level heart rate: it is in the export, but few destinations accept it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Checklist
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Take a Takeout export of your Fitbit data.&lt;/li&gt;
&lt;li&gt;Check what your app plans to support.&lt;/li&gt;
&lt;li&gt;Choose one route (Health Connect or Google Health API) and set it up now.&lt;/li&gt;
&lt;li&gt;Check that sleep, resting heart rate and HRV show up for a few days.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Sources: Google Health API docs (developers.google.com/health/about, /setup, /data-types) and Health Connect docs, read on 22/10/2026.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;This article was written with AI assistance and approved by a person before publishing.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>fitbit</category>
      <category>health</category>
      <category>data</category>
      <category>tutorial</category>
    </item>
  </channel>
</rss>
