[IanFrelinger]

Summary

Scale accuracy pass for github.blockers 1.1.0 (post-#874) on 100 public PRs across 10 repositories, including IanFrelinger/Ashlar and other .NET-heavy hosts. Brick-raw grading only; Sonnet skipped because it adds no rubric signal beyond the payload.

Rebased onto current master after #871 merged. One commit on top of master.

Headline (CLEAN): verdicts 100/100, blockers 211/211, wrong claims 0, Anthropic spend $0 / $6 budget. No new brick bugs. Re-scored after the rebase: same numbers.

Changes

• docs/benchmarks/blockers-accuracy-scale.md — published scale result
• Pointer from docs/benchmarks/blockers-accuracy.md; CHANGELOG + harness README
• tools/blockers-accuracy/targets.json — 100 PRs / 10 repos
• fixtures-scale/ + results/2026-10-25-scale/ — snapshot and brick-raw scores
• score.py — strip github.blockers: quote text read from GitHub so it reaches the answer as data #872 inline-code backticks on check names when mapping brick reasons (without this, every required-check claim scored as fabricated)

Testing

cd tools/blockers-accuracy
python3 score.py results/2026-10-25-scale --fixtures fixtures-scale --brick-dir brick
# → verdict_correct 100/100, found 211/211, overcall/fabricated/unattributed 0

Re-score of the post-#874 44-PR snapshot after the quote strip: still 109/109. Re-run after rebase onto master: still 100/100 and 211/211.

Testing strategy (blast radius)

Docs + harness only. No product code change.

• N/A kernel / prod-style gates for this docs+fixtures PR

Checklist

• Documentation updated
• No product bug findings to file
• Rebased onto master after docs(benchmarks): "what is blocking this PR?" accuracy study (44 PRs, brick vs native) #871

Records / docs touched (for drift audit)

• docs/benchmarks/blockers-accuracy-scale.md
• docs/benchmarks/blockers-accuracy.md
• CHANGELOG.md
• tools/blockers-accuracy/**