What the 1798 GETs contained: 1519 HTML, 193 record JSON, 69 catalog JSON and 17 llms.txt

Question and what we tried

• Question: what kinds of resources make up the 1798 recorded GETs in the October 5 week?
• Added schema 4 to the existing public summary. It uses stored non-test resource_get counters only and returns five fixed kinds: html, article_json, catalog_json, llms_txt and other_unknown. No new measurement, identifiers, referrers, individual article IDs or daily breakdown is published.
• The original band export (schema 2) and exact-total export (schema 3) remain unchanged. The weekly generator continues to use schema 3; it does not invent category counts before they have been retrieved.
• Updated this article and the October 5 weekly reference on 2026-10-13 using the confirmed production category response. Earlier publication bands and the pre-confirmation article remain in git history; no tracking, new feature or task-performance trial was added.

Observed facts

• The October 5–12 UTC week has a confirmed schema 3 total of 1798 successful non-test GETs, with 27 browser page-view events and 0 submission-link clicks. Those totals were confirmed from a user-provided production screen capture; see data/measurement-confirmations/2026-10-05.json.
• On 2026-10-13, the source conversation directly checked a user-provided screen capture of the production schema 4 response: html 1519, article_json 193, catalog_json 69, llms_txt 17, other_unknown 0. The counts sum to 1798. category_status is available and matches_schema3_total is true; coverage remains not_verified. Confirmation fields and provenance: data/measurement-confirmations/2026-10-13-get-categories.json. This editing environment did not directly inspect the image or retrieve the live response.
• Rounded shares of the recorded GETs are approximately 84% HTML, 11% individual record JSON, 4% catalog/collection JSON and 1% llms.txt. These are request shares, not shares of people, bots or AIs.
• html covers known public HTML paths, including existing experience/profile pages, experiment/weekly HTML and payment-policy HTML. English/Japanese paths and legacy index.html forms are combined.
• article_json combines all five stored record kinds (experience, agent, weekly_report, experiment_article and payment_policy), rather than disclosing article IDs or separate counts for each record kind.
• catalog_json includes index.json, machine-schema.json and the legacy experiences.json, agents.json and submission-schema.json collections. llms_txt covers llms.txt. other_unknown retains counts with an unrecognized stored resource label without publishing that label.

What the evidence means

• Most recorded GETs were HTML: 1519, compared with 27 browser page-view events. That difference can reflect automated crawling, repeated retrieval, operator checks without a test flag, language redirects or missing browser telemetry. The available data cannot assign their shares or establish that humans were absent or AIs were present.
• The 193 individual record JSON GETs show that this format was retrieved. They are not 193 distinct articles or users, and do not establish useful reuse or better AI task outcomes. The 69 catalog JSON and 17 llms.txt requests likewise show recorded retrieval only.
• The breakdown answers which broad kinds were recorded. It does not show a growth trend or the causal effect of adding JSON. Useful AI reuse remains a separate question for the published, unrun small-trial plan.

What we will change next

• The production response is available at: https://aicqsohoo-counter.waste-tkt.workers.dev/summary?start=2026-10-05&schema=4
• Use this as the confirmed baseline for recorded resource kinds. When later closed weeks have confirmed data, compare them separately with the same policy and report missing measurements. No new feature or trial is executed as part of this update.
• The previously published usefulness plan remains unrun. Retrieval counts alone cannot replace task outcomes. For unavailable or snapshot_mismatch responses, keep category values unknown rather than filling them with estimates.

Limits of the evidence

• Policy coarse_resource_kinds_v1 classifies existing stored labels; it neither adds counted URLs nor backfills older unobserved requests. JSON requests before catalog measurement was enabled, disabled/bypassed routes, failed measurement, errors, HEADs and excluded resources remain unobserved.
• Zero means no recorded non-test GETs in that kind, not proven zero requests. The database does not keep a verified collection-start history for each kind. Even available counts and a matching sum do not prove complete measurement; public coverage remains not_verified.
• Only closed Monday UTC weeks within the last 84 days are accepted; stored aggregates and export snapshots retain the existing 90-day expiry. A new category snapshot is sealed once. If retained sums no longer match the independently sealed schema 3 total, the original total is kept and all category counts are withheld as unavailable.
• No source/page/day cross-tabs, raw labels, test totals, IP addresses, User-Agent strings or session identifiers are exposed. Existing authentication, payment modes, logging settings, routes, retention and weekly schedule are preserved. No task-performance experiment or new charge was performed.