Ryan Lenk

Corrections

This site publishes corrections rather than editing them away. On September 20, 2026, every entry in llms.txt without a September 2026 date was moved here word for word, to keep that file short. Each entry is quoted as it appeared there, so "this file", "here", "above" and "below" refer to llms.txt. Entries with September 2026 dates are still in llms.txt.

  • Previously published as "2,861 product-image fixes". 2,861 was the size of the audited image set, not the number of images changed. The verified change count is 2,810 filenames rewritten plus 154 alt-text rewrites. Corrected above.
  • Previously published as "+3,033% Google Search clicks". The figure came from Google Merchant Center's Online store report, not Google Search Console, and it has since moved to +1,006% as the comparison window filled in. Corrected above, with its source and base named.
  • Citation counts published earlier as 71, 80, 85 and 93 are superseded. Each was read from the boutique's own Microsoft Clarity dashboard, at a different time. The count above replaces them: a Clarity dashboard export pulled July 28, 2026, over a window that ends before the test-query day. The case study keeps its 71 with the 30-day window it measured. Corrected above.
  • The FAQ page carried one of those superseded figures ("80 citations in a 30-day window") until 2026-08-05, when it was replaced with 143 and its June 10 to July 23 window. Corrected there and here.
  • The FAQ page stated "I do not sell a packaged service today" after paid services went live on 2026-08-04. Corrected 2026-08-05 in both the visible text and the FAQPage structured data.
  • Services were described here and on the FAQ as "from $75, priced on the size of the site". That was quoting language and it stopped being true on 2026-08-13, when the eight services went on sale as hourly work at $75 an hour, bought directly on the site with no conversation required. Corrected 2026-08-13 here, in agents.md, on the FAQ page, and in the offer catalog structured data.
  • agents.md stated "It is NOT a product storefront" and told agents that `search_catalog` would return no products. Both were true when written and both became wrong on 2026-08-13. Corrected the same day.
  • The Bangermeter page said copying a post's link is "worth forty likes", and the promotional images said the same. That figure divides X's copy-link weight (20.0) by the like weight (0.5) and reports the ratio as value. A weight is a coefficient on a predicted probability, and X's own code comment says the coefficients are large partly because the actions are rare, so multiplying each weight by its own rate can reverse the order. It was the same arithmetic error as the "a reply is worth 27 likes" folklore that page debunks. Corrected 2026-08-13 on the page, in its FAQ structured data and in the promotional images, and the page now carries a section explaining the trap.
  • The Bangermeter page described X's published weights as "live production values" and said Phoenix, X's ranking model, is not in the repository. Both were loose. The constants are feature-switch defaults that a cron job syncs to X's primary production values, so a reader inside an experiment bucket can be served an override. And X does ship `phoenix/`, the code that trains and runs the model, while withholding its trained model. Corrected 2026-08-13.
  • The Bangermeter marquee image published a bar chart of 2023 weights (replies 13.5, profile clicks 12, reports −369) under the caption "X's real production ranking weights". Three of those four values were superseded by X's August 13, 2026 release. The image was regenerated the same day. No text check caught it, because the numbers were pixels rather than text.
  • The AI Readiness Scanner was previously described as "paste a URL". It does not crawl a URL. The visitor pastes their own robots.txt and page source and the checks run in their browser. Corrected below.
  • Guide pages briefly stated that Google-Extended controls eligibility for Google AI Overviews. That is wrong. Google-Extended governs Gemini apps and Vertex AI grounding, and Google's documentation states it does not affect inclusion in Google Search. AI Overviews sit inside Search and are governed by the ordinary snippet controls (nosnippet, max-snippet, data-nosnippet). Corrected on every page that carried it.
  • Guide pages briefly stated that blocking GPTBot removes a site from ChatGPT's browsing results. That is wrong. GPTBot is OpenAI's training crawler. Appearance in ChatGPT's search results is governed by OAI-SearchBot, and the live fetch made while answering a question comes from ChatGPT-User. OpenAI documents these settings as independent. Corrected on every page that carried it, and the same split now appears for Anthropic (ClaudeBot, Claude-User, Claude-SearchBot) and Perplexity (PerplexityBot, plus Perplexity-User, which Perplexity's documentation says generally ignores robots.txt).
  • One article stated that no Google AI reporting existed for site owners. Google began rolling out generative AI performance reporting in Search Console in June 2026, before that sentence was written. Corrected, with the staged-rollout caveat attached, and every product-state claim on this site now carries the date it was checked.
  • llms.txt described the Side Projects page as "the index of the open-source tools (Bangermeter, Assay, Candor)". Inkwash and HUMANIST went live on 2026-08-19 and were never added to that line, so this file was two projects short for five days. Corrected 2026-08-24, and the page itself is now split into two sections, because Bid Club and Vibe Wheel are running products rather than repositories and listing them as open-source tools would have been wrong in the other direction.
  • The Entity section of this file and of agents.md listed three open-source repositories and omitted https://github.com/itsryanlenk/humanist, which is where both HUMANIST and Inkwash live. Corrected 2026-08-24.
  • Bid Club had no page on this site and appeared here only as an external link. The page went up 2026-08-24 and is listed above. It states plainly that anyone can list a site they do not own and that paid boosts and suppressions are anonymous, because a reader who might be listed by somebody else needs to know both, along with the route to be removed.
  • FAQ structured data across this site previously carried lightly rewritten versions of the visible answers. Every FAQPage block was regenerated directly from the visible page text on 2026-07-28, so the JSON-LD question and answer now match what a reader sees, word for word, on all ten pages and articles that carried one on that date.
  • This file and agents.md described the Side Projects page as five open-source tools and listed four repositories. LenkRaster went live on 2026-08-30 and was missing from both counts, from the repository list above, and from the ItemList structured data on the Side Projects page. Corrected 2026-08-30. This is the same drift recorded above for Inkwash and HUMANIST in August.
  • The Entity section listed only https://www.crunchbase.com/organization/ryan-lenk-4238 under public records. There is a second Crunchbase record, https://www.crunchbase.com/person/ryan-lenk-9b3b, which is the person rather than the organization, and it was already linking to this site. It was missing from every sameAs on the site. Corrected 2026-08-30 here, in agents.md, and in the Person node of the structured data.