<?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: Robert</title>
    <description>The latest articles on DEV Community by Robert (@robswierk).</description>
    <link>https://dev.to/robswierk</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%2F4053993%2Fb8f114ea-e5d9-4003-b6a1-1549fdddee94.png</url>
      <title>DEV Community: Robert</title>
      <link>https://dev.to/robswierk</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/robswierk"/>
    <language>en</language>
    <item>
      <title>Sonnet 5.5 vs Sonnet 5: When To Reach For Which</title>
      <dc:creator>Robert</dc:creator>
      <pubDate>Sat, 21 Nov 2026 10:00:00 +0000</pubDate>
      <link>https://dev.to/robswierk/sonnet-55-vs-sonnet-5-when-to-reach-for-which-1091</link>
      <guid>https://dev.to/robswierk/sonnet-55-vs-sonnet-5-when-to-reach-for-which-1091</guid>
      <description>&lt;p&gt;The choice is between Sonnet 5.5 and Sonnet 5 for the same job. Both are Claude models, and both did editorial.boxes.anchor.read for us: 365 calls over the last 90 days, 329 on Sonnet 5.5 and 36 on Sonnet 5.&lt;/p&gt;

&lt;p&gt;We used Sonnet 5 first, from 2026-09-25 to 2026-09-29, and Sonnet 5.5 from 2026-09-30 to 2026-10-26. Sonnet 5.5 is the only one of the two still in use.&lt;/p&gt;

&lt;p&gt;This post sets out what each cost us per call, how long each took, how often each failed, and what the token figures show. It also says where the comparison stops being useful.&lt;/p&gt;

&lt;h2&gt;
  
  
  What each one cost us per call
&lt;/h2&gt;

&lt;p&gt;Sonnet 5 (anthropic/claude-sonnet-5) cost $0.01007 per call over its 36 calls, $0.3624 in total. That made it the cheaper of the two per call.&lt;/p&gt;

&lt;p&gt;Sonnet 5.5 (anthropic/claude-sonnet-5-5) cost $0.01755 per call over its 329 calls, $5.7731 in total. That is 1.74 times the cost of Sonnet 5 per call, or $0.00748 more per call.&lt;/p&gt;

&lt;p&gt;These are the amounts we paid on our own volume, not a vendor rate card. They say what this job cost us, and nothing about what either model costs on someone else's workload.&lt;/p&gt;

&lt;h2&gt;
  
  
  Speed and failures
&lt;/h2&gt;

&lt;p&gt;Sonnet 5.5 had a median of 2.0 seconds per call. Sonnet 5 had a median of 2.6 seconds, which makes it 1.3 times slower than Sonnet 5.5, a gap of 0.6 seconds.&lt;/p&gt;

&lt;p&gt;Neither tool had a single failure. Sonnet 5.5 came back with 0 failures in 329 calls and Sonnet 5 with 0 in 36. That is the failure rate on our side, not the vendor's uptime, and 36 calls is a small sample to hold against 329.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the token figures show
&lt;/h2&gt;

&lt;p&gt;The jobs were not the same size. Sonnet 5.5 averaged 9,059 total input tokens per call and Sonnet 5 averaged 4,414. Read as written, the record shows Sonnet 5.5 handling a bigger input per call. Output ran at 143 tokens per call on Sonnet 5.5 against 163 on Sonnet 5.&lt;/p&gt;

&lt;p&gt;Cache use was low on both. On Sonnet 5.5, 1,097 of those input tokens per call came from the prompt cache, 12.1 percent, leaving 7,811 billed at full rate. On Sonnet 5, 219 came from the cache, 5.0 percent, leaving 4,196 billed at full rate.&lt;/p&gt;

&lt;p&gt;Most of the input was a cache miss on both tools, so most of what we paid for was input billed at full rate. Sonnet 5.5 had the higher cache share, but it still billed 7,811 input tokens per call at full rate against 4,196 on Sonnet 5.&lt;/p&gt;

&lt;p&gt;The record does not say why the input per call is larger on Sonnet 5.5, so we make no claim about it. The higher cost per call of Sonnet 5.5 goes with the larger input recorded for it, and the figures alone do not separate what the model charges from what the job sent.&lt;/p&gt;

&lt;h2&gt;
  
  
  Which one we kept
&lt;/h2&gt;

&lt;p&gt;We kept Sonnet 5.5. Its last call was on 2026-10-26, while Sonnet 5 last ran on 2026-09-29, the day before Sonnet 5.5 first appears.&lt;/p&gt;

&lt;p&gt;The record does not say why we moved. It shows the switch and the cost per call on each side, and nothing about a reason.&lt;/p&gt;

&lt;p&gt;Sonnet 5 ran for 5 days and Sonnet 5.5 for 27 days, so the two tools did not overlap at any point in the window.&lt;/p&gt;

&lt;h2&gt;
  
  
  The limits of this comparison
&lt;/h2&gt;

&lt;p&gt;This is not a benchmark. It is our own prompts, our own job and our own volumes, measured over the window in the facts. A reader with a different job may see different numbers.&lt;/p&gt;

&lt;p&gt;The two tools also did not run side by side. Sonnet 5 had 5 days of use and 36 calls, and Sonnet 5.5 had 27 days and 329 calls. The input per call differs between them, so the cost and speed figures are not a like-for-like test, and nothing here speaks to output quality.&lt;/p&gt;

&lt;p&gt;The failure counts are on our side only, and with no failures on either tool they do not separate the two.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Today we would pick Sonnet 5.5 for this job, because it is the one we kept running, it had no failures in 329 calls, and its median of 2.0 seconds was the faster of the two. We would pick Sonnet 5 instead if cost per call is the deciding factor, since it came in at $0.01007 against $0.01755. Whichever you choose, check how much of your input comes from the prompt cache first, because on both tools most of it was billed at full rate.&lt;/p&gt;
&lt;/blockquote&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://neuragrowth.co/blog/sonnet-5-5-vs-sonnet-5-when-to-reach-for-which-2/" rel="noopener noreferrer"&gt;neuragrowth.co&lt;/a&gt;. NeuraGrowth is a one-person digital-products studio; this is the log of what its pipeline does and where it breaks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>claude</category>
      <category>anthropic</category>
      <category>costoptimization</category>
      <category>llmops</category>
    </item>
    <item>
      <title>Etsy listings expired in the morning: two guides dropped off the shop</title>
      <dc:creator>Robert</dc:creator>
      <pubDate>Fri, 20 Nov 2026 10:00:00 +0000</pubDate>
      <link>https://dev.to/robswierk/etsy-listings-expired-in-the-morning-two-guides-dropped-off-the-shop-3bja</link>
      <guid>https://dev.to/robswierk/etsy-listings-expired-in-the-morning-two-guides-dropped-off-the-shop-3bja</guid>
      <description>&lt;p&gt;Etsy listings expire, and when one does it leaves the catalogue. Twice in four days our listing_renewal alert reported exactly that, each time with the same title: "Odnowienia: wygasło 1", meaning one listing expired.&lt;/p&gt;

&lt;p&gt;The first was on 2026-11-13 at 04:46:42. The alert named #129, Ship-Ready Security Audit for Vibe-Coded SaaS. It said the listing had expired in Etsy since the previous morning and come off the catalogue.&lt;/p&gt;

&lt;p&gt;The second was on 2026-11-16 at 04:46:46, four seconds past the same minute. It named #169, Vercel Services Field Guide for Indie Devs, with the same wording: expired in Etsy since the previous morning, gone from the catalogue.&lt;/p&gt;

&lt;p&gt;Both alerts also said the reviews for the listings were already waiting in the panel.&lt;/p&gt;

&lt;p&gt;Both were warn severity, not an outage. The alerts do not say why the listings were not renewed before they expired, and they carry no counts beyond the single listing each time, so we have nothing further to report on cause.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;An Etsy listing that expires quietly leaves the catalogue, so check for expired listings every morning instead of waiting for a sale to show it is gone.&lt;/p&gt;
&lt;/blockquote&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://neuragrowth.co/log/etsy-listings-expired-in-the-morning-two-guides-dropped-off-the-shop/" rel="noopener noreferrer"&gt;neuragrowth.co&lt;/a&gt;. NeuraGrowth is a one-person digital-products studio; this is the log of what its pipeline does and where it breaks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devops</category>
      <category>programming</category>
      <category>ai</category>
      <category>webdev</category>
    </item>
    <item>
      <title>FFmpeg exited with an empty error message in a compilation task</title>
      <dc:creator>Robert</dc:creator>
      <pubDate>Thu, 19 Nov 2026 10:00:00 +0000</pubDate>
      <link>https://dev.to/robswierk/ffmpeg-exited-with-an-empty-error-message-in-a-compilation-task-3nh3</link>
      <guid>https://dev.to/robswierk/ffmpeg-exited-with-an-empty-error-message-in-a-compilation-task-3nh3</guid>
      <description>&lt;p&gt;FFmpeg failed inside one of our tasks on 2026-11-14 at 10:05, and the message it left behind says almost nothing. The alert reads: "RuntimeError: ffmpeg nie powiodl sie: ". After the colon there is no text at all.&lt;/p&gt;

&lt;p&gt;The task is orqestra.kids_compilation (id 3068f503-0bb1-4967-8525-d7548b9c0294). It ended in an exception, which our alerting raised as a warning. The alert sends one bell per task name per hour and points to the worker log for the rest, so the full detail lives there and not in the alert.&lt;/p&gt;

&lt;p&gt;What we know is short. The task called FFmpeg, FFmpeg did not succeed, and the text captured from it was empty. The alert does not say which command was run, what exit code came back, or what the input files were. It also does not say why the call failed, and we are not going to guess at a cause from an empty string.&lt;/p&gt;

&lt;p&gt;One other record from the same week concerns the same kind of work, so we mention it and no more. A kids_episode run on 2026-11-09 took 102.6 minutes against a median of 21.0 minutes across 19 runs, 4.9 times slower. It finished with status success. The facts do not connect it to the compilation failure, and we are not claiming they are related.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;When a wrapper raises on a failed FFmpeg call, make sure the error text it captures is not empty, because "ffmpeg failed:" followed by nothing leaves nothing to search for or debug.&lt;/p&gt;
&lt;/blockquote&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://neuragrowth.co/log/ffmpeg-exited-with-an-empty-error-message-in-a-compilation-task/" rel="noopener noreferrer"&gt;neuragrowth.co&lt;/a&gt;. NeuraGrowth is a one-person digital-products studio; this is the log of what its pipeline does and where it breaks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devops</category>
      <category>programming</category>
      <category>ai</category>
      <category>webdev</category>
    </item>
    <item>
      <title>ElevenLabs re-voiced 108 tutorial clips and we added 39 for French</title>
      <dc:creator>Robert</dc:creator>
      <pubDate>Wed, 18 Nov 2026 10:00:00 +0000</pubDate>
      <link>https://dev.to/robswierk/elevenlabs-re-voiced-108-tutorial-clips-and-we-added-39-for-french-30e1</link>
      <guid>https://dev.to/robswierk/elevenlabs-re-voiced-108-tutorial-clips-and-we-added-39-for-french-30e1</guid>
      <description>&lt;p&gt;On 2026-11-07 we shipped a change to the Skat game that moves its tutorial narration to ElevenLabs. The 108 ordinary lesson clips, in Polish, German and English, were recorded again with the same voices used at the table. A further 39 clips, marked -fr, were recorded for the French card pattern.&lt;/p&gt;

&lt;p&gt;Spoken text needed its own wording, so the lesson text is written differently for reading aloud. The "=" sign is either spoken as a word or replaced by a pause. Numbers are spelled out in inflected form, and 120 is written out in words in German.&lt;/p&gt;

&lt;p&gt;The check is a script, tools/check-tutorial-voices.mjs. It sends every clip to Gemini for a transcript, then compares the words and the pace against the source text. Across all 147 clips it flags 0 outliers after the fixes. The record does not say how many it flagged before them.&lt;/p&gt;

&lt;p&gt;The generator gained two switches. --jobs produces the list of texts to listen to, and --files re-records only the clips you name. A new test, tutorial-french, checks that the keys exist in the lessons, that DE, EN and SL translations are present, and that every step read aloud in French has a -fr recording.&lt;/p&gt;

&lt;p&gt;The commit touches 150 files, adding 172 lines and removing 3.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;When a synthetic voice reads your text, rewrite symbols and numbers as spoken words first, then transcribe every clip back and compare it to the source instead of trusting your ears.&lt;/p&gt;
&lt;/blockquote&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://neuragrowth.co/log/elevenlabs-re-voiced-108-tutorial-clips-and-we-added-39-for-french/" rel="noopener noreferrer"&gt;neuragrowth.co&lt;/a&gt;. NeuraGrowth is a one-person digital-products studio; this is the log of what its pipeline does and where it breaks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devops</category>
      <category>programming</category>
      <category>ai</category>
      <category>webdev</category>
    </item>
    <item>
      <title>Playwright showed a real Buy click arrives with no Referer</title>
      <dc:creator>Robert</dc:creator>
      <pubDate>Tue, 17 Nov 2026 10:00:00 +0000</pubDate>
      <link>https://dev.to/robswierk/playwright-showed-a-real-buy-click-arrives-with-no-referer-12bn</link>
      <guid>https://dev.to/robswierk/playwright-showed-a-real-buy-click-arrives-with-no-referer-12bn</guid>
      <description>&lt;p&gt;Playwright gave us the reference case for a bug in our click counting. On 9 November 2026 at 20:09 UTC we drove a real click on a Buy button from the home page, and the request reached nginx with the Referer "-". Every Buy button on the site is rel="nofollow noreferrer", so a genuine click from the site carries no Referer at all.&lt;/p&gt;

&lt;p&gt;What made us look: after the earlier per-address limit (P26), the &lt;code&gt;site&lt;/code&gt; source still sat at about 200 clicks a day over about 60 products. It was 987 the week before that limit and 1257 the week after. The nginx log for 9 November showed 207 /api/go/ hits from 205 addresses, one or two each. 195 of them carried the Referer "&lt;a href="https://neuragrowth.co" rel="noopener noreferrer"&gt;https://neuragrowth.co&lt;/a&gt;" and 5 carried "&lt;a href="https://neuragrowth.co/" rel="noopener noreferrer"&gt;https://neuragrowth.co/&lt;/a&gt;".&lt;/p&gt;

&lt;p&gt;That was the mismatch. A real Buy click arrives with "-", and these arrived with the site's own address as Referer. An address limit cannot catch that pattern, because each address showed up only once or twice. The record does not say who or what sent them.&lt;/p&gt;

&lt;p&gt;The change, P26b, shipped on 9 November 2026 and touched 2 files with 87 lines added and 1 removed. A &lt;code&gt;site&lt;/code&gt; click that carries any Referer is now sent on to its destination and not counted. Other networks keep whatever Referer their readers bring, and the forged clicks never reach the address ledger.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;If a link carries rel="noreferrer", drive a real click in a browser tool like Playwright and read what your server logs, then treat any traffic that contradicts it as not coming from that button.&lt;/p&gt;
&lt;/blockquote&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://neuragrowth.co/log/playwright-showed-a-real-buy-click-arrives-with-no-referer/" rel="noopener noreferrer"&gt;neuragrowth.co&lt;/a&gt;. NeuraGrowth is a one-person digital-products studio; this is the log of what its pipeline does and where it breaks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devops</category>
      <category>programming</category>
      <category>ai</category>
      <category>webdev</category>
    </item>
    <item>
      <title>Where Our Etsy Shelf Stands: Six Recent Products, One for Toddlers</title>
      <dc:creator>Robert</dc:creator>
      <pubDate>Mon, 16 Nov 2026 10:00:00 +0000</pubDate>
      <link>https://dev.to/robswierk/where-our-etsy-shelf-stands-six-recent-products-one-for-toddlers-57o</link>
      <guid>https://dev.to/robswierk/where-our-etsy-shelf-stands-six-recent-products-one-for-toddlers-57o</guid>
      <description>&lt;p&gt;Six products shipped from the catalogue between 4 October and 2 November 2026. Five of them are PDF guides at $6.99 each. The sixth is a printed activity book for toddlers.&lt;/p&gt;

&lt;p&gt;But the five guides sit next to it in the same run of work, and the difference between where they are sold and where the toddler book is sold is the plainest fact in the set.&lt;/p&gt;

&lt;p&gt;Everything here comes from the catalogue records for products 210, 211, 212, 222, 223 and 224. We have nothing in those records about how any of them has been received, so this is an account of what shipped and where it can be found.&lt;/p&gt;

&lt;h2&gt;
  
  
  The one toddler book in this batch
&lt;/h2&gt;

&lt;p&gt;Product 222 is My First Dot Markers Activity Book: ABC, Numbers and Shapes for Toddlers Ages 1-3. It shipped on 26 October 2026. Its type in our records is a KDP coloring book, and it belongs to no series.&lt;/p&gt;

&lt;p&gt;The title names three things inside it: letters, numbers and shapes. It is written for ages 1 to 3. That is all the record says about the contents, so we will not describe pages or activities beyond what the title already tells you.&lt;/p&gt;

&lt;p&gt;The record carries no price for this one. The five guides each carry a price of $6.99. The toddler book has none in our records, so we have no figure to give you.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why the toddler book is not on our Etsy shelf
&lt;/h2&gt;

&lt;p&gt;The facts for product 222 list one place where it is on sale: KDP print. Etsy and Gumroad are not listed for it.&lt;/p&gt;

&lt;p&gt;The five guides are the other way round. Each one lists Etsy and Gumroad as the places it is on sale, and none lists KDP print. So within this single month of shipping, the catalogue splits cleanly by product type: PDF guides on Etsy and Gumroad, the printed toddler book on KDP.&lt;/p&gt;

&lt;p&gt;The records do not say why the split is drawn this way, and we will not guess at a reason. It is simply what the catalogue shows for these six products.&lt;/p&gt;

&lt;h2&gt;
  
  
  The five guides that shipped around it
&lt;/h2&gt;

&lt;p&gt;In order of shipping they are product 210, n8n Multi-Agent Architecture: The 2026 Field Guide, on 4 October. Product 211, The Faceless YouTube Growth Operator Playbook, on 5 October. Product 212, Etsy + ProductHunt Demand Decoder: Validate a Niche in 90 Minutes, on 7 October.&lt;/p&gt;

&lt;p&gt;Then product 223, Codex Security for Vibe Coders: Scan, Validate, Patch, on 31 October. And product 224, Terra Computer-Use Agents: Recipes and Security Guardrails, on 2 November.&lt;/p&gt;

&lt;p&gt;All five are PDF guides at $6.99, none belongs to a series, and all five list Etsy and Gumroad as their two places of sale. If you want to see two of them in full, product 223 is at &lt;a href="https://neuragrowth.co/products/223-codex-security-for-vibe-coders-scan-validate-patch/" rel="noopener noreferrer"&gt;https://neuragrowth.co/products/223-codex-security-for-vibe-coders-scan-validate-patch/&lt;/a&gt; and product 224 is at &lt;a href="https://neuragrowth.co/products/224-terra-computer-use-agents-recipes-and-security-guardrails/" rel="noopener noreferrer"&gt;https://neuragrowth.co/products/224-terra-computer-use-agents-recipes-and-security-guardrails/&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  A product about Etsy, on Etsy
&lt;/h2&gt;

&lt;p&gt;One of the five is about the marketplace itself. Product 212, Etsy + ProductHunt Demand Decoder, is subtitled Validate a Niche in 90 Minutes, and it is on sale on Etsy as well as Gumroad.&lt;/p&gt;

&lt;p&gt;We will not claim anything about how well it works, because the records give us nothing on that. What the record does give us is the title, the price of $6.99, and the shipping date of 7 October 2026.&lt;/p&gt;

&lt;h2&gt;
  
  
  Another look at our Etsy listings
&lt;/h2&gt;

&lt;p&gt;That post is at &lt;a href="https://neuragrowth.co/log/one-etsy-listing-expired-and-left-the-catalogue-overnight/" rel="noopener noreferrer"&gt;https://neuragrowth.co/log/one-etsy-listing-expired-and-left-the-catalogue-overnight/&lt;/a&gt; and covers a listing that expired and dropped out overnight.&lt;/p&gt;

&lt;p&gt;It is a different kind of record from the six here, which are all finished and on sale as of their shipping dates.&lt;/p&gt;

&lt;p&gt;If you want a wider view of what the catalogue actually contains, we also audited our own site after Gumroad reported a guide count. That post is at &lt;a href="https://neuragrowth.co/blog/gumroad-said-83-guides-so-we-audited-our-own-site/" rel="noopener noreferrer"&gt;https://neuragrowth.co/blog/gumroad-said-83-guides-so-we-audited-our-own-site/&lt;/a&gt;.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;If you are a parent looking for the toddler book, go to KDP print, because that is the only place the record lists it, and do not expect to find product 222 on Etsy or Gumroad. If you are reading our catalogue for the guides, look on Etsy and Gumroad, where all five of them are listed at $6.99, and check the shipping date on each page, since these six went out between 4 October and 2 November 2026. In both cases, trust the listing page for the current state of a product over any summary of it, including this one.&lt;/p&gt;
&lt;/blockquote&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://neuragrowth.co/blog/where-our-etsy-shelf-stands-six-recent-products-one-for-toddlers/" rel="noopener noreferrer"&gt;neuragrowth.co&lt;/a&gt;. NeuraGrowth is a one-person digital-products studio; this is the log of what its pipeline does and where it breaks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>productivity</category>
      <category>showdev</category>
      <category>beginners</category>
    </item>
    <item>
      <title>Two Etsy listings expired and dropped out of the catalogue</title>
      <dc:creator>Robert</dc:creator>
      <pubDate>Sun, 15 Nov 2026 10:00:00 +0000</pubDate>
      <link>https://dev.to/robswierk/two-etsy-listings-expired-and-dropped-out-of-the-catalogue-1n1o</link>
      <guid>https://dev.to/robswierk/two-etsy-listings-expired-and-dropped-out-of-the-catalogue-1n1o</guid>
      <description>&lt;p&gt;Etsy listings expire, and on 2026-11-09 at 04:46 one of our alerts told us that two had. The alert was a warning, not an error. It was titled "Odnowienia: wygasło 2", which is Polish for renewals: 2 expired.&lt;/p&gt;

&lt;p&gt;The two listings were #103, the Agentic Engineering Transition Playbook, and #160, the Four Seasons Watercolor Bilingual Flashcard Deck (EN+PL). The alert says both expired on Etsy from the previous morning and dropped out of the catalogue. It also says their renewal reviews were already waiting in the panel.&lt;/p&gt;

&lt;p&gt;The alert does not say why these two expired and not others. It does not say how long they stayed off the catalogue beyond "since yesterday morning", and it gives no count of how many listings we renew in total. We will not guess at any of that here.&lt;/p&gt;

&lt;p&gt;In the same fortnight, two worker heartbeat warnings fired, on 2026-11-01 and 2026-11-02. The record does not connect them to the expiry, and we are not claiming a link.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;An Etsy listing that expires leaves the catalogue until it is renewed, so a pending renewal review is not a renewal and needs a watcher of its own.&lt;/p&gt;
&lt;/blockquote&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://neuragrowth.co/log/two-etsy-listings-expired-and-dropped-out-of-the-catalogue-2/" rel="noopener noreferrer"&gt;neuragrowth.co&lt;/a&gt;. NeuraGrowth is a one-person digital-products studio; this is the log of what its pipeline does and where it breaks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devops</category>
      <category>programming</category>
      <category>ai</category>
      <category>webdev</category>
    </item>
    <item>
      <title>Sonnet 5.5 vs Sonnet 5: When To Reach For Which</title>
      <dc:creator>Robert</dc:creator>
      <pubDate>Sat, 14 Nov 2026 10:00:00 +0000</pubDate>
      <link>https://dev.to/robswierk/sonnet-55-vs-sonnet-5-when-to-reach-for-which-5cgm</link>
      <guid>https://dev.to/robswierk/sonnet-55-vs-sonnet-5-when-to-reach-for-which-5cgm</guid>
      <description>&lt;p&gt;If you are choosing between two adjacent Sonnet versions for the same job, the useful questions are what each costs per call, how fast each is, and how much of the prompt each gets from cache. We have figures for both from our own use.&lt;/p&gt;

&lt;p&gt;Over the last 90 days, two Anthropic models handled book.questions.solver across 453 calls: Sonnet 5.5 took 337 of them and Sonnet 5 took 116. Neither returned a single failure. Sonnet 5 was last used on 2026-09-24, and Sonnet 5.5 is the one still in use.&lt;/p&gt;

&lt;p&gt;Below we give each model's cost per call, median time, token figures and dates of use, then say where the comparison stops being reliable.&lt;/p&gt;

&lt;h2&gt;
  
  
  The two tools and when each ran
&lt;/h2&gt;

&lt;p&gt;Sonnet 5 is anthropic/claude-sonnet-5. Sonnet 5.5 is anthropic/claude-sonnet-5-5. Same vendor, same job, adjacent versions.&lt;/p&gt;

&lt;p&gt;Sonnet 5 ran first: 116 calls, first used 2026-09-23 and last used 2026-09-24. Sonnet 5.5 ran 337 calls, first used 2026-10-08 and last used 2026-10-15.&lt;/p&gt;

&lt;p&gt;Sonnet 5.5 is the one still in use. Sonnet 5 is the one we stopped using. The record does not say why we stopped.&lt;/p&gt;

&lt;h2&gt;
  
  
  Cost per call
&lt;/h2&gt;

&lt;p&gt;Sonnet 5.5 cost $0.01798 per call, $6.0594 in total over its 337 calls. Sonnet 5 cost $0.02414 per call, $2.8008 in total over its 116 calls.&lt;/p&gt;

&lt;p&gt;Sonnet 5 cost 1.34 times what Sonnet 5.5 did per call, a gap of $0.00616 per call.&lt;/p&gt;

&lt;p&gt;These are what we paid on our own volume. They are not a rate card, and they say nothing about what the vendor lists on its pricing page.&lt;/p&gt;

&lt;h2&gt;
  
  
  Speed and failures
&lt;/h2&gt;

&lt;p&gt;Sonnet 5.5 had a median of 4.2 seconds per call. Sonnet 5 had a median of 5.2 seconds, which is 1.24 times slower, a gap of 1.0 seconds.&lt;/p&gt;

&lt;p&gt;Neither tool had a failure in this window: 0 of 337 calls for Sonnet 5.5 and 0 of 116 for Sonnet 5.&lt;/p&gt;

&lt;p&gt;That is the count of calls that came back as failures on our side. It is not the vendor's uptime, and a window with 453 calls and no failures is a narrow thing to lean on.&lt;/p&gt;

&lt;h2&gt;
  
  
  Tokens and the prompt cache
&lt;/h2&gt;

&lt;p&gt;The token figures only make sense read together. Sonnet 5.5 averaged 7,375 input tokens per call in total. Of those, 6,561 were billed at full rate and 777 came from the prompt cache, which is 10.5 percent. It produced 469 output tokens per call.&lt;/p&gt;

&lt;p&gt;Sonnet 5 averaged 10,083 input tokens per call in total, and all 10,083 were billed at full rate. Its cached input was 0 tokens per call, so 0.0 percent. It produced 398 output tokens per call.&lt;/p&gt;

&lt;p&gt;So the two jobs were not the same size on average, 7,375 input tokens against 10,083, and only Sonnet 5.5 got any of its prompt from cache. The facts do not say why the input sizes differed, so we do not claim a cause.&lt;/p&gt;

&lt;p&gt;Sonnet 5 paid full rate on its whole, larger input and cost more per call. These facts show both things sitting together, and they do not let us say how much of the cost gap comes from the cache and how much from the larger input.&lt;/p&gt;

&lt;p&gt;Billed input alone would mislead here. Sonnet 5.5 shows 6,561 billed tokens per call against 10,083 for Sonnet 5, but part of that gap is simply that the total input was smaller, and part is the 777 tokens that came from cache.&lt;/p&gt;

&lt;h2&gt;
  
  
  The limits of this comparison
&lt;/h2&gt;

&lt;p&gt;This is not a benchmark. These are our prompts, our one job, and our volumes, measured over the window in the facts.&lt;/p&gt;

&lt;p&gt;The calls were not split between the tools in any controlled way. Sonnet 5 ran on two days in September and Sonnet 5.5 on a later stretch, with 116 and 337 calls respectively. The total input per call differed between them, so the two did not necessarily receive identical work.&lt;/p&gt;

&lt;p&gt;A different job, different prompts or a different volume may produce different numbers. We have nothing here about answer quality, only cost, time, tokens and failures.&lt;/p&gt;

&lt;p&gt;We also have nothing about what changed between the two stretches of use besides the model name, so we cannot attribute the cost, speed or token differences to the models alone.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;For this job we would pick Sonnet 5.5 today: it was cheaper per call at $0.01798 against $0.02414, faster at a 4.2 second median against 5.2, had no failures, and it is the one we kept using. We would pick differently if your own measurements on your own prompts showed otherwise, or if you need a quality comparison, since our figures carry none.&lt;/p&gt;
&lt;/blockquote&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://neuragrowth.co/blog/sonnet-5-5-vs-sonnet-5-when-to-reach-for-which/" rel="noopener noreferrer"&gt;neuragrowth.co&lt;/a&gt;. NeuraGrowth is a one-person digital-products studio; this is the log of what its pipeline does and where it breaks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>claude</category>
      <category>anthropic</category>
      <category>costoptimization</category>
      <category>llmops</category>
    </item>
    <item>
      <title>restic backup to B2: the alert said 27 hours since last success</title>
      <dc:creator>Robert</dc:creator>
      <pubDate>Fri, 13 Nov 2026 10:00:00 +0000</pubDate>
      <link>https://dev.to/robswierk/restic-backup-to-b2-the-alert-said-27-hours-since-last-success-5e4a</link>
      <guid>https://dev.to/robswierk/restic-backup-to-b2-the-alert-said-27-hours-since-last-success-5e4a</guid>
      <description>&lt;p&gt;restic is the tool behind our remote database backup. A systemd timer called orqestra-backup.timer runs it every day at 04:50 and sends the result to Backblaze B2. On 2026-11-08 at 06:40 an alert fired because the last successful remote backup was 27 hours old.&lt;/p&gt;

&lt;p&gt;The alert was a warning, not a critical. 27 hours is past that mark. That is the whole of what the number tells us.&lt;/p&gt;

&lt;p&gt;The record does not say why the run did not succeed. It does not say whether the timer fired, whether restic failed, or whether B2 was reachable. The alert only reports the age of the last good run.&lt;/p&gt;

&lt;p&gt;What the alert does give is where to look: the journal for the orqestra-backup unit, and the backup script at /opt/orqestra/scripts/backup-db.sh on the VPS.&lt;/p&gt;

&lt;p&gt;Two other warnings from the previous week are on the record, both about the Celery worker heartbeat going stale. They came from a different component, and the facts here do not connect them to the backup alert.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Alert on the age of the last successful restic run rather than on the exit code of the last attempt, because a backup that never started leaves no failure to catch.&lt;/p&gt;
&lt;/blockquote&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://neuragrowth.co/log/restic-backup-to-b2-the-alert-said-27-hours-since-last-success/" rel="noopener noreferrer"&gt;neuragrowth.co&lt;/a&gt;. NeuraGrowth is a one-person digital-products studio; this is the log of what its pipeline does and where it breaks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devops</category>
      <category>programming</category>
      <category>ai</category>
      <category>webdev</category>
    </item>
    <item>
      <title>Claude Opus: 209 calls, $0.11 each, one call took 312 seconds</title>
      <dc:creator>Robert</dc:creator>
      <pubDate>Thu, 12 Nov 2026 10:00:00 +0000</pubDate>
      <link>https://dev.to/robswierk/claude-opus-209-calls-011-each-one-call-took-312-seconds-158o</link>
      <guid>https://dev.to/robswierk/claude-opus-209-calls-011-each-one-call-took-312-seconds-158o</guid>
      <description>&lt;p&gt;We called Claude Opus (claude-opus-5-5) 209 times over 10 days, from 2026-10-30 to 2026-11-08. Those calls covered 22 different jobs and cost $23.7465 in total, or $0.11362 per call. The median call took 18.1 seconds.&lt;/p&gt;

&lt;p&gt;For comparison, Claude Sonnet (claude-sonnet-5-5) ran 156 times in the same window at $0.0165 per call, with a median of 2.3 seconds and a slowest call of 19.2 seconds. Opus cost about 7 times Sonnet per call. Its median ran about 8 times longer.&lt;/p&gt;

&lt;p&gt;The cheaper models took the parsing, listing copy, captions and extraction. The facts do not say how we chose that split, so we will not claim it was measured.&lt;/p&gt;

&lt;p&gt;Opus had one failure in 209 calls, 0.5 percent. The record does not say what was wrong with the output, and it does not say whether the 312.2 second call and the failed call were the same one.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;When you size a timeout or a budget for Opus, use the slowest call you have seen rather than the median, and price it per call across every job it touches.&lt;/p&gt;
&lt;/blockquote&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://neuragrowth.co/log/claude-opus-209-calls-0-11-each-one-call-took-312-seconds/" rel="noopener noreferrer"&gt;neuragrowth.co&lt;/a&gt;. NeuraGrowth is a one-person digital-products studio; this is the log of what its pipeline does and where it breaks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devops</category>
      <category>programming</category>
      <category>ai</category>
      <category>webdev</category>
    </item>
    <item>
      <title>One Etsy listing expired and left the catalogue overnight</title>
      <dc:creator>Robert</dc:creator>
      <pubDate>Wed, 11 Nov 2026 10:00:00 +0000</pubDate>
      <link>https://dev.to/robswierk/one-etsy-listing-expired-and-left-the-catalogue-overnight-21f1</link>
      <guid>https://dev.to/robswierk/one-etsy-listing-expired-and-left-the-catalogue-overnight-21f1</guid>
      <description>&lt;p&gt;On 2026-11-07 at 04:46 an alert fired with a warn severity: "Odnowienia: wygasło 1". The kind is listing_renewal. The body says one listing expired on Etsy since the previous morning and left the catalogue.&lt;/p&gt;

&lt;p&gt;The listing is #158 Glaze Field Guide: Prompt to App to Ship in One Day. The alert adds that its reviews are already waiting in the panel.&lt;/p&gt;

&lt;p&gt;That is everything the alert carries. It gives no count of other expiries, no price, and no figure for how long the listing was live. The record also does not say why this one was not renewed before it lapsed.&lt;/p&gt;

&lt;p&gt;The alert states only that the listing left the catalogue, and we cannot say from this record whether it has been renewed since.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;An Etsy listing that expires drops out of the catalogue on its own, so a renewal check that runs the morning after is already a day late.&lt;/p&gt;
&lt;/blockquote&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://neuragrowth.co/log/one-etsy-listing-expired-and-left-the-catalogue-overnight/" rel="noopener noreferrer"&gt;neuragrowth.co&lt;/a&gt;. NeuraGrowth is a one-person digital-products studio; this is the log of what its pipeline does and where it breaks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devops</category>
      <category>programming</category>
      <category>ai</category>
      <category>webdev</category>
    </item>
    <item>
      <title>ElevenLabs eleven_v3: 271 calls, $0.5716 in total, no failures</title>
      <dc:creator>Robert</dc:creator>
      <pubDate>Tue, 10 Nov 2026 10:00:00 +0000</pubDate>
      <link>https://dev.to/robswierk/elevenlabs-elevenv3-271-calls-05716-in-total-no-failures-55k9</link>
      <guid>https://dev.to/robswierk/elevenlabs-elevenv3-271-calls-05716-in-total-no-failures-55k9</guid>
      <description>&lt;p&gt;ElevenLabs eleven_v3 handled 271 calls for us between 2026-10-28 and 2026-11-06. All of them came from one job, kids_episode.tts, which turns episode scripts into speech. None failed.&lt;/p&gt;

&lt;p&gt;The cost was $0.5716 in total, about $0.00211 per call. For comparison, the 213 calls to claude-opus-5-5 in the same ten days cost $25.4973, roughly $0.12 each. Speech was a very small line next to the writing and research that came before it.&lt;/p&gt;

&lt;p&gt;On speed, the median call took 1.4 seconds and the slowest took 4.1. The spread is narrow.&lt;/p&gt;

&lt;p&gt;What the record does not show is the length of the text sent per call, or whether call length moved the timing. The figures above are what eleven_v3 did on this one workload over these dates, and nothing wider.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Size a text-to-speech timeout from the median and the slowest call you have measured for that job, and measure it separately from your language-model calls, because the two sit orders of magnitude apart.&lt;/p&gt;
&lt;/blockquote&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://neuragrowth.co/log/elevenlabs-eleven-v3-271-calls-0-5716-in-total-no-failures/" rel="noopener noreferrer"&gt;neuragrowth.co&lt;/a&gt;. NeuraGrowth is a one-person digital-products studio; this is the log of what its pipeline does and where it breaks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devops</category>
      <category>programming</category>
      <category>ai</category>
      <category>webdev</category>
    </item>
  </channel>
</rss>
