docs(evidence): re-verify q5 ground truth drop of retired Agent Desk (2026-08-14)

.lane/reports/docs-q5-ground-truth-reverify-2026-08-14.md

@@ -0,0 +1,35 @@
+# Lane report: q5 ground-truth re-verify (2026-08-14)
+
+Lane: tinystudio-io lane 1
+Branch: `docs/q5-ground-truth-reverify-2026-08-14`
+Item: `a792ca1847` — "[unreviewed-by-opus] Replace the retired "Agent Desk" ground truth in the AI-search controlled questions (q5) — fi"
+
+## Outcome
+
+Closed. The retired Agent Desk ground truth in the `q5` controlled question has already been replaced on `origin/main` — commit `ed62202` ("fix(evidence): align AI-search q5 ground truth with the current offer, not the retired Agent Desk", 2026-08-11), merged via PR #43 (`ad9cee3`) — and this lane re-verified the replacement against the current head (`e9fc96a`) and the live deployment of that head. No code change was needed; the closeout evidence is recorded.
+
+## Verification performed
+
+1. **Fixture**: `evidence-fixtures/ai-search/controlled-questions.json` — the `q5` truth reads "tinystudio.io is TinyStudio's own site: the free leak audit of high-ticket service homepages, and the human-reviewed desk that closes what the audit finds." No "Agent Desk" appears in any question truth.
+2. **Bundle == fixture**: the embedded bundle in `public/audit.html` is byte-identical to `{"questions": …, "evidence": …}` serialised from the two fixture files (`JSON.stringify` equality → true). Runs: 11 (9 `wrong`, 2 `not-tested`, 0 `found`), `testedOn: 2026-08-09`. No run relabelled; the `q5`/google run's stale Agent Desk answer is preserved verbatim as evidence.
+3. **Source guard**: `npm run check` → exit 0, "TinyStudio.io checks passed." The AI-search evidence-artifact guard (`scripts/check-site.mjs` lines 751-807) enforces git-tracked fixtures, audit-page markers, the four strict states in `public/audit.js`, and byte-for-byte bundle/fixture equality.
+4. **Full suite**: `npm test` → 117 tests, 0 failures (heading-hierarchy 6/6, sitemap 7/7, agent-worker 76/76, agent-ui 16/16, product-contract 8/8, first-viewport-audience 4/4). Only pre-existing, out-of-scope note: `/` overflows at 240/260px (does not gate exit code, unrelated to q5).
+5. **Live served == source bytes** (2026-08-14): fresh HTTPS fetches of `/audit`, `/llms.txt`, `/offer.md` are byte-identical to their `public/` sources (md5 match on all three). The Cloudflare release-state record pins the deployed SHA at `e9fc96a`, the same commit as this lane's head.
+6. **Tied surfaces**: homepage `#identity` `q5` row carries `data-ai-question="q5-what-is-tinystudio-io"` with the leak-audit wording; `llms.txt` and `offer.md` carry the matching `## Identity` block and the `q5` → homepage answer-readiness mapping.
+
+## Files changed
+
+- `docs/evidence/ai-search/2026-08-14-q5-ground-truth-reverify.md` — new evidence receipt recording the closeout on the current head and live site (the lane's claimed file).
+- `.lane/reports/docs-q5-ground-truth-reverify-2026-08-14.md` — this report.
+
+## Verification commands
+
+- `npm run check` → exit 0, "TinyStudio.io checks passed."
+- `npm test` → exit 0, 117 tests / 0 failures.
+- `git diff --check` → clean.
+- Bundle equality probe (node) → `bundle === serialised fixtures: true`; `q5` truth printed verbatim.
+- `curl https://tinystudio.io/{audit,llms.txt,offer.md} | md5sum` → identical to `public/` sources.
+
+## Honest boundary
+
+This lane claims nothing about any engine's answer. The `q5`/google run from 2026-08-09 still records `wrong` with the verbatim retired-Agent-Desk answer; only a future controlled re-run through the same fixture can measure whether engines now read the corrected yardstick and the live pages.

docs/evidence/ai-search/2026-08-14-q5-ground-truth-reverify.md

@@ -0,0 +1,137 @@
+# q5 ground truth — retired Agent Desk replaced, re-verify against current main and live
+
+Date: 2026-08-14
+Scope: the review-queue item "[unreviewed-by-opus] Replace the retired
+"Agent Desk" ground truth in the AI-search controlled questions (q5)"
+(item id `a792ca1847`). The `q5-what-is-tinystudio-io` ground truth in
+`evidence-fixtures/ai-search/controlled-questions.json` still named the
+retired self-serve Agent Desk as part of the current offer. This receipt
+re-verifies the replacement of that retired ground truth against the current
+`origin/main` head (`e9fc96a`, "docs(evidence): re-verify tap-target
+finding on current main and live (2026-08-14) (#183)") and the live
+deployment of that head. It is a source-side plus served-bytes re-verify;
+it claims nothing about any engine's future answer and runs no new live
+AI-search capture.
+
+## Summary
+
+The retired Agent Desk ground truth has been replaced on source and on the
+live site. The fix landed in commit `ed62202`
+("fix(evidence): align AI-search q5 ground truth with the current offer,
+not the retired Agent Desk", 2026-08-11), merged to `origin/main` via PR #43
+(`ad9cee3`, "Merge pull request #43 from
+nish3451/fix/ai-search-rerun-entity-offer"), and is preserved verbatim on
+the current head:
+
+- `evidence-fixtures/ai-search/controlled-questions.json` — the `q5` truth
+  reads "tinystudio.io is TinyStudio's own site: the free leak audit of
+  high-ticket service homepages, and the human-reviewed desk that closes
+  what the audit finds." No "Agent Desk" appears in any question truth.
+- `public/audit.html` — the embedded AI-search bundle under
+  `<script type="application/json" id="ai-search-evidence">` is
+  byte-for-byte identical to the JSON serialisation of
+  `{"questions": controlled-questions.json, "evidence": evidence.json}`,
+  enforced by the drift guard in `scripts/check-site.mjs` (lines 751-807).
+- `docs/evidence/ai-search/2026-08-09-controlled-rerun.md` — carries the
+  2026-08-11 alignment note, so the historical receipt does not silently
+  outdate.
+- `docs/evidence/ai-search/2026-08-11-q5-ground-truth-alignment.md` — the
+  original alignment receipt, on main.
+
+No captured run was touched: `evidence.json` keeps the 2026-08-09 captures
+(`testedOn: 2026-08-09`), 11 runs — 9 `wrong` (including
+`q5`/google@2026-08-09, whose verbatim answer still describes the retired
+Agent Desk, quoted honestly with its remediation), 2 `not-tested`
+(`q1`/chatgpt, `q1`/perplexity), 0 `found`. The yardstick no longer encodes
+the retired product; the stale answers are preserved as the evidence they
+are.
+
+The tied surfaces that answer `q5` carry the same wording:
+
+- `public/index.html` `#identity` — the `q5-what-is-tinystudio-io` row
+  ("What is tinystudio.io" → "This site: the leak audit and the desk behind
+  it.") with `data-ai-question="q5-what-is-tinystudio-io"` (line 190).
+- `public/llms.txt` and `public/offer.md` — the `## Identity` block names
+  "the free leak audit of high-ticket service homepages and the
+  human-reviewed desk that closes what the audit finds"; the
+  `## Answer Readiness: Preferred Source Pages` section maps
+  `q5-what-is-tinystudio-io` to `https://tinystudio.io/` (homepage) in both
+  files, and the mirror checks fail if the pair drifts.
+
+## Source checks on the current head (`e9fc96a`)
+
+1. `node scripts/check-site.mjs` passes ("TinyStudio.io checks passed.").
+   The "AI-search evidence artifact" guard (lines 751-807) requires the two
+   fixture files plus the README to be git-tracked, the audit page to mount
+   the `id="ai-search-evidence"` marker and the `data-ai-search-evidence`
+   consumer, `public/audit.js` to carry the four states (`found`, `wrong`,
+   `absent`, `not-tested`) and their labels (`Found`, `Wrong`, `Absent`,
+   `Not tested`), and the embedded bundle to equal
+   `{"questions": aiQuestions, "evidence": aiEvidence}` byte-for-byte.
+2. The full test suite passes (117 tests, 0 failures):
+   `node --test scripts/test-heading-hierarchy.mjs` (6/6),
+   `node --test scripts/test-sitemap.mjs` (7/7),
+   `node --test scripts/test-agent-worker.mjs` (76/76),
+   `node --test scripts/test-agent-ui.mjs` (16/16, including the
+   "every controlled question maps to a preferred source page" subtest that
+   locks the eight-question mapping),
+   `node --test scripts/test-product-contract.mjs` (8/8),
+   `node --test scripts/test-first-viewport-audience.mjs` (4/4).
+   `git diff --check` is clean.
+3. Independent re-serialisation: extracting the embedded bundle from
+   `public/audit.html` and serialising the two fixture files gives
+   byte-equal strings (`JSON.stringify(embedded) ===
+   JSON.stringify({questions, evidence})` → true); the `q5` truth inside
+   the bundle reads exactly the current-offer wording above; runs are
+   11 (9 `wrong`, 2 `not-tested`, 0 `found`), `testedOn` `2026-08-09`.
+
+## Live re-verification 2026-08-14
+
+The Cloudflare release-state record
+(`/home/nish/workspaces/agent-state/lanes/release-state-tinystudio-io.json`)
+pins the deployed SHA at `e9fc96ab018c0ed538d103e0994c8ed70c363673` — the
+same commit as this lane's `origin/main` head. Fresh HTTPS fetches confirm
+the served bytes match the source:
+
+| URL | HTTP | md5 (live) | md5 (public/ source) | match |
+|---|---|---|---|---|
+| `/audit` | 200 | `7162b128...` | `7162b128...` | yes |
+| `/llms.txt` | 200 | `d14e5d56...` | `d14e5d56...` | yes |
+| `/offer.md` | 200 | `bb06b34d...` | `bb06b34d...` | yes |
+
+The live `/audit` page embeds the same `q5` truth — "tinystudio.io is
+TinyStudio's own site: the free leak audit of high-ticket service
+homepages, and the human-reviewed desk that closes what the audit finds."
+— quoted verbatim from the served page's bundle. The live `llms.txt` and
+`offer.md` carry the matching identity and the `q5` → homepage answer-
+readiness mapping.
+
+## What this pass deliberately did not do
+
+- No new live AI-engine run was performed, and none is claimed. Whether an
+  engine now answers `q5` correctly remains a live question, honestly
+  measured only by a future controlled re-run through the same fixture.
+- No captured run was relabelled: the 9 `wrong` / 2 `not-tested` / 0
+  `found` states from 2026-08-09 stand, including the `q5`/google run whose
+  verbatim answer describes the retired Agent Desk.
+- No code, fixture, or page markup changed in this pass; the fixture and
+  the audit-page embed remain byte-identical to `origin/main`.
+
+## Verification (reproduce)
+
+```sh
+npm run check
+npm test
+git diff --check
+```
+
+## Closeout
+
+The item as stated — replace the retired "Agent Desk" ground truth in the
+AI-search controlled questions (`q5`) — is **closed against current main
+and live**: the replacement landed in `ed62202` (merged via PR #43) and is
+preserved verbatim on the current head (`e9fc96a`); the drift guard and the
+full test suite pass; the deployed `/audit` bundle, `llms.txt` and
+`offer.md` serve the current-offer wording byte-identical to source. The
+receipt now records the closeout on the current head so the item cannot be
+re-opened by tracker drift.