# Report revision history

Dates use Asia/Shanghai. First-publication dates refer to this repository's public editions, supported by Git history and the retained edition records. Original experiment dates, publication dates and later observations are separate.

## 2026-11-13

- **Lab #07 v1.0:** published the Brand Authority study and its reviewed 60-answer coding, rules and recount script. It documents the measurement guide, workbook and individual evidence release. No post-release recommendation lift was measured.

## 2026-11-12

- **Lab #06 v1.0:** published the content optimization evidence review, with the full report, linked primary sources and review methodology. Added the report to the Learning Hub and sitemap.

## 2026-11-11

- **Interim Review, Part 2 v1.0:** published the question-framing experiment with 120 redacted consumer web-search answers, reproducible statistics, and separate resource, person and source evidence. Added the second review to the Learning Hub and shared report navigation.

## 2026-11-10

- **Interim Review, Part 1 v1.0:** published the historical replication and sample-size analysis, with 300 redacted selected answers, all attempt outcomes, normalized measurements and twelve historical claim decisions. The five original learning records retain their numbering and data.

## 2026-11-06

- **Lab #05 v1.1:** appended dated evidence for the expanded 12/72 URL monitor, successful combined run, 11 genuine IndexNow change notifications accepted with HTTP 202 (key validation pending), corrective Google Sitemap submission and seven acknowledged Google indexing requests. Initial counts and failed runs remain preserved; request acceptance does not establish indexing.

- **Lab #05 v1.0:** published a technical learning record with reviewed aggregate observations, methods and explicit indexing/crawler evidence limits. Added two genuine editorial routes and a fifth Learning Hub entry; the original four AI experiments and 60 response archives are preserved. The Sitemap and monitored inventory grow to 12 editorial and 72 canonical routes after this publication.
- **Monitoring completion work:** added an IndexNow-only validation path with deployment-gated change notifications and fixed delayed first-deployment handling. The owner approved ten-second Bing spacing for the existing weekly workflow; its Monday schedule is unchanged.

- **Deferred Bing recheck:** added a Bing-only manual workflow with ten-second spacing and a separate completion callback that verifies a successful run and committed full-batch evidence before issuing one private notification. Existing weekly monitoring is unchanged; no new live checks were run during this setup.

- **Bing fault handling:** retain sanitized terminal provider faults, distinguish HTTP-400 throttling from indexing evidence, pace serial requests, and stop remaining batch calls after a quota or access block. All planned URL slots remain explicit, including unqueried UNKNOWN results. Runtime provenance separates local baseline collection from GitHub weekly validation.

- **Explicit Bing API key mode:** added owner-approved account-wide API key authentication as an explicit private-workflow opt-in. The program remains fixed to GEO read operations, with no credential-method fallback. Both local OAuth callback registration attempts were rejected; their failure records remain private. Credentials and raw monitoring data are excluded from the public site.

- **Bing authorization setup:** added an owner-completed readonly OAuth helper with private client inputs, exact loopback/state validation, separate native Keychain storage and encrypted private Actions Secret installation. Synthetic checks do not establish provider callback compatibility or actual API authorization. Original experiment data and Google baselines are unchanged.

## 2026-11-05

- **Multi-engine monitoring:** added separate Bing Webmaster observations, OpenAI search crawl-eligibility probes and change-only IndexNow notifications under the existing private weekly workflow. The public IndexNow ownership file uses explicit project-scoped keyLocation. Existing Google data and one-time Sitemap ledger are preserved. Missing authorization or unverifiable index/crawl membership remains UNKNOWN; no AI-answer experiment was run.

- **Monitor application disclosures:** published an owner-operated index-monitor homepage and privacy policy for Google OAuth configuration. Both pages use noindex and are excluded from the research sitemap and monitored URL set. This publication does not establish Production authorization or change experiment results.

## 2026-11-04

- **Google monitoring setup:** added a public intent-only canonical inventory and extended the sitemap to the evidence index (ten important routes). An immutable private pre-operation baseline precedes this change and any new Google submission.
- **Private automation:** added reviewed official Search Console OAuth, URL Inspection and one-time Sitemaps clients, a private-repository Actions workflow template and bounded error/alert tests. Raw Google responses, manual request records and history remain private; OAuth secrets never enter Git. Configuration and tests alone do not establish live Google authorization or indexed-state observations.

- **Technical maintenance:** standardized canonical URLs, distinct page descriptions and page-specific Open Graph metadata across all nine report/hub pages. Fixed the Stage 1 appendix and Stage 3 metadata omissions and the Stage 2 appendix social-preview mismatch.
- **Author and series markup:** added truthful Person, Article/CollectionPage and breadcrumb relationships based on visible Helen Bird attribution and recorded dates. Original AI responses are archived as responses, not attributed to Helen Bird as their author.
- **Evidence access:** added an HTML index and 60 individually addressable selected-answer pages, linked from the existing reports and hub. Original complete appendices remain expanded; raw evidence, coding, findings and download archives are unchanged.
- **Discovery:** added a nine-URL canonical sitemap with recorded revision dates. Host-level robots and training preferences were not changed; no Search Console submission or actual-index claim is made.
- **Interpretation:** these are bundled technical changes, not a single-variable GEO experiment. No AI discovery test or recommendation lift has been measured. Local audit evidence remains excluded from publication inputs.

## 2026-11-03

- **Stage 4 v1.0:** first redacted public release of 18 complete new answers, 54 candidate-by-answer observations and eight actually cited own/bylined page audits. The OpenAI arm is Responses API + native web search and cannot represent consumer ChatGPT Search. Preserved all submissions and sanitized local capture-recovery history, without exposing private originals.
- **Stage 4 peer review:** renamed the title from discoverability to recommendations, made question-level results primary, separated personal names from associated content assets, made tool-call limits visible beside comparisons, and connected observed content features to untested next-step hypotheses. No new model responses were collected for the editorial revision.
- **Learning Hub:** added a question-oriented, author-attributed entry point connecting all four reports, public datasets and complete appendices. This is a content-organization change; no effect on AI discovery or recommendations has been tested.
- **Stage 1 v1.2 / Stage 2 v1.2 / Stage 3 v1.3:** navigation and version-note updates only. Stage 1 remains at the website root. Earlier data, coding, findings and experiment qualifications are unchanged.

## 2026-11-02

- **Stage 1 v1.1:** clarified mixed recommended entities, the conditional own/bylined-content citation denominator, and date/reasoning changes in observed repeat overlap. Added post-hoc type classification and a purposive source audit. Added a concise reader and a complete expanded technical appendix.
- **Stage 2 v1.1:** clarified six planned cells, ten actual submissions, six complete selected answers and four earlier ChatGPT C technical/capture problems. Made ChatGPT B query-observation limits and manual C provenance more visible. Kept post-publication X search checks separate from the original outcomes. Added a concise reader and expanded appendix.
- **Stage 3 v1.2:** added shared navigation and explicit version metadata; preceding analysis and pending human-review status unchanged.
- **All reports:** original response records and outcome coding preserved; shared links connect all three experiments without moving the Stage 1 root URL. Original-evidence sections remain expanded, in line with the existing reading preference. No new AI test responses were collected for these revisions.
- **Stage 3 v1.1** (`2146d8f`): added matched within/across-wording comparisons, entity normalization and URL sensitivity, a bounded source audit, a pending human-review packet, and a reader/appendix split. Original strict coding preserved; no new model requests.
- **Stage 3 v1.0** (`f18647b`): first redacted public release. The OpenAI arm uses the Responses API with web search and does not represent the ChatGPT consumer web app.

## 2026-11-01

- **Stage 2 v1.0** (`1b7e89e`): first redacted public release, including six selected answers, ten actual submissions, all four earlier ChatGPT C technical/capture problems, and a qualified manual replacement.

## 2026-10-30

- **Stage 1 v1.0** (`ab0c22c`): first redacted public report and reproducible data. Later same-day commits framed the repository as an ongoing learning log and translated report commentary into English.

> Versions were introduced with the November 2 cross-report revision; earlier releases are labeled retrospectively. Git history preserves the actual preceding files. These editorial revisions do not retroactively strengthen experiment controls or establish GEO effectiveness.
