Weekly experiment: 2026-10-05

Question and what we tried

• Question: can an AI retrieve the site’s evidence as individual reusable records, rather than extracting it from HTML?
• We added a read-only JSON catalog and then aligned request counting with its records. A stable ID names each article; its evidence status, source and content hash let an AI inspect what it is using. The request list is generated from that same catalog so newly published records can be counted without a weekly Worker release. These were implementation changes, not a comparison of AI task performance.
• Sources: discovery/interventions.jsonl entries dated 2026-10-05; data/reports/2026-10-06-x402-lessons.json separately records payment implementation failures and an offline MPP prototype. Those are not customer sales or new payment attempts in this report.
• Updated on 2026-10-13 after the production schema 3 response was confirmed on a user-provided screen capture: 27 browser page-view events, 1798 successful resource GETs and 0 submission-link clicks. Confirmation provenance and the response fields are recorded in data/measurement-confirmations/2026-10-05.json. The original 20–29 / 1790–1799 bands remain in git history at d85d6af0738417ef98aba0acc82836886d9ed789. No new experiment or inferred exact value is claimed.
• Later on 2026-10-13, confirmed schema 4 categories were added as a reference: data/measurement-confirmations/2026-10-13-get-categories.json. Detailed results and interpretation: /experiment/2026-10-13-get-categories/. The original bands remain in git history; this is a reviewed result update, not a newly executed experiment.

Observed facts

• Period: 2026-10-05 inclusive to 2026-10-12 exclusive (UTC).
• Measurement coverage: not_verified. This describes configuration, not proven delivery or uptime.
• Recorded non-test browser page-view events: 27 (exact recorded events).
• Recorded non-test submission-link clicks: 0 (exact recorded events).
• Recorded successful non-test resource GETs at the edge: 1798 (exact recorded events).
• Exact totals come from a sealed snapshot of retained aggregate counts, not estimates from earlier bands. Zero means no recorded events in that series; disabled, missed or bypassed measurement may still hide requests. Clicks are not completed submissions.
• Only whole-week totals are published in this exact export; no referring-domain, page, daily or test-count breakdown is included.
• Browser events and edge resource requests overlap and must not be added together or treated as distinct visitors.
• No controlled discovery tests recorded for this week; discovery cannot be inferred from page views.
• Confirmed GET kinds: HTML 1519; individual record JSON 193; catalog/collection JSON 69; llms.txt 17; other_unknown 0. Total: 1798, matching the sealed schema 3 total. Source: data/measurement-confirmations/2026-10-13-get-categories.json.
• Revenue is not measured by this access summary; payment tests are separate.

What the evidence means

• The confirmed exports record 27 browser page-view events, 1798 successful resource GETs and 0 submission-link clicks, excluding marked tests. Of the GETs, 1519 were HTML, 193 individual record JSON, 69 catalog/collection JSON, 17 llms.txt and 0 other_unknown. GETs are about 67 times the browser events, but collection conditions differ; this does not establish human absence or AI use. Category counts are not evidence of the causal effect of adding the catalog or of useful AI reuse.
• We now have a machine-readable way to retrieve reviewed articles. Whether the independent measurements and failure details improve an AI’s solution remains unanswered. No controlled discovery tests were recorded for this week, and measurement coverage is unverified. Neither growth nor general usefulness follows from these totals.

What we will change next

• Plan only: compare the same task input alone, with a general explanation, and with that explanation plus a reviewed AICQ failure/measurement record. Use separate fresh sessions and hidden scoring; preregister tasks and failure criteria before any runs.
• The 2026-10-13 usefulness plan in the experiment notebook specifies the small trial. No paid API call, transfer, real payment or new billing is part of this change.

Limits of the evidence

• Exact totals and coarse categories were confirmed from user-provided production screen captures directly reviewed in the source conversation. This editing environment did not retrieve the live response or inspect the images. Coverage remains not_verified; no referring-domain, individual-page or daily breakdown is available. Counts identify recorded requests only. Zero clicks do not establish zero visitors, submissions or revenue. The owner-operated 0.01 USDC transfer is a payment test, not customer revenue.
• Without the optional edge route, JavaScript-free HTML/JSON/llms.txt requests are unobserved. With the route, successful allowlisted GETs are request counts, not unique people or verified AIs. Failed/bypassed measurement and default github.io URLs remain unobserved.
• Explicit measurement=test events are excluded. Unmarked operator tests cannot be recognized without identifiers.
• Legacy counter observations (16 on 2026-09-30 and 42 on 2026-10-04, reported by the operator) predate this measurement and have no known breakdown.