docs(bench): phase-8 pre-registration + ABC audit + operator manual (#117)#128
Merged
Conversation
…#117) The offline docs for the headline run, plus a from-cold instruction manual. - README.md: rewritten into a full operator manual — the five conditions (C0-C3 + Cw), both run modes (single + multi), all six hypotheses, how to run the pipeline (run -> report -> oracle; author + validate_scenario), how to read every metric (success / misled / verification_rate / verified_then_correct / tokens / cost, Wilson + bootstrap + Holm), the file layout, the authoring loop, and the sharp edges (surf's identifier- renaming hashing, read-only-by-design tools, don't-stack-PRs, where spend actually is). This is the single entry point for future context. - PREREGISTRATION.md: hypotheses H1-H6 with pre-specified directions, conditions, models, tiered N, max_turns, metric definitions, the analysis plan (Wilson + bootstrap + Holm, confirmatory vs exploratory), exclusion rules, stopping rule, and a freeze gate to git-tag before the run. - ABC_CHECKLIST.md: self-audit against the Agentic Benchmark Checklist (arXiv 2507.02825) across task validity / outcome validity / reporting, with known limitations disclosed. Docs only; no code change. uv run pytest still 38. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Phase-8 offline docs for the headline run, plus a from-cold instruction manual so anyone (future me or an agent) can pick this experiment up and run/interpret it without re-deriving the design.
README.md— rewritten into a full operator manual: the five conditions (C0–C3 + Cw), both run modes (single + multi), all six hypotheses, the pipeline (run → report → oracle, plusauthor+validate_scenario), how to read every metric (success / misled / verification_rate / verified_then_correct / tokens / cost, with Wilson + bootstrap + Holm), the file layout, the authoring loop, and the sharp edges (surf's identifier-renaming hashing, read-only-by-design tools, don't-stack-PRs, where spend actually is).PREREGISTRATION.md— H1–H6 with pre-specified directions, conditions, models, tiered N, max_turns, metric definitions, the analysis plan (Wilson + bootstrap + Holm–Bonferroni, confirmatory vs exploratory), exclusion + stopping rules, and a freeze gate to git-tag before the run.ABC_CHECKLIST.md— self-audit against the Agentic Benchmark Checklist (arXiv 2507.02825) across task/outcome/reporting validity, with limitations disclosed.Verification
uv run pytest— 38 passed (docs-only, no code change);cargo fmtn/a.Scope
Implements the offline half of #117. The pre-registration is committed but not yet frozen/tagged — that waits on the scenario set being final (PRs #126 TS + #127 QA merged) and a clean oracle pass, per its own freeze gate. The paid matrix run is the remaining phase-8 work.
🤖 Generated with Claude Code