# v2.0 — 2026-11-10

- New independent 1,000-domain robots/llms/home window; no new/old response merge.
- Immutable attempt provenance, wire and decoded bodies, hashes, finite retries,
  explicit protocol defaults and unknown states; no browser fallback or paid AI calls.
- Fixed malformed-agent group boundaries, BOM/URI handling and adult-label false positives;
  errors, HTML, compressed-body problems and truncation remain separate.
- Root llms.txt minimum-H1 candidates replace the old unarchived MIME/length heuristic.
- Recovered original Tranco 94P72 by exact ranking CSV comparison; froze published cohort.
- Completed 1,000 home observations and 16 targeted manual reviews; no domain exclusions.
- Added exact-final-robots-URL sensitivity, signal syntax counts and common-subset comparison.
- 11,956 Google-reference comparisons; two documented nonstandard Disallow:* differences;
  both implementations give the same AI-any result, 249/854.
- New English/Chinese article; v1.1 remains available at /v1.1 and original versioned downloads.
- An initial pilot/aborted collection used an earlier intermediate cohort; excluded entirely.
  Only full-02 contributes to v2.0, following the corrected 30-domain pilot.

The time, research UA and validation method changed together. This is not a policy-change
causal estimate. See metrics.json for all collection-to-analysis reclassifications.