feat: Oracle baselines + blindness refined to judgments-not-ground-truth#9
Open
jirispilka wants to merge 1 commit into
Open
feat: Oracle baselines + blindness refined to judgments-not-ground-truth#9jirispilka wants to merge 1 commit into
jirispilka wants to merge 1 commit into
Conversation
Learnings from an 8-task batch run (apify-mcp-server #1066 quick-wins):
1. Oracle baselines. A green suite is not proof of correctness: one run
produced a vitest config that silently double-ran the unit suite while
staying fully green — caught only by comparing test counts against a
pre-change baseline. The inner loop now records baseline metrics
(iter-1/baseline.txt) before the first source edit, and convergence
requires green AND baseline-consistent; unexplained metric deltas fail
the oracle.
2. Blindness taxonomy. The old rule ("reviewers get pasted content only —
never file access") conflated two blindnesses. Blind-to-judgments
(claim.md, peer reviews) is essential and stays. Blind-to-repository is
harmful: a diff-only reviewer cannot verify byte-identical-to-HEAD
claims or check that a removed import is truly unused. Reviewers now
get repo/git read access; .devforge/ judgment files stay pasted-only.
The criteria author additionally gets _request_fact_check.md (verified
facts, no solution content) — previously it invented grep commands over
guessed file paths. The final reviewer's lens is now defined as
post-fix integrated state, not "same as reviewer plus working tree".
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016pXZvhEoE6Lz3DpBSL3n1Y
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
2 participants
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.
Learnings from an 8-run batch on apify/apify-mcp-server#1066 (PRs #1070–#1078 there, one devforge run each).
What we're solving
Two flow-level gaps the batch exposed:
Flow changes (not obvious from the diff)
claim.md/peer reviews stay off-limits, and.devforge/is still pasted-only. Practical effect: inner review becomes verification (runninggit show HEADcomparisons), not plausibility-reading.Follow-up (not in this PR)
Trivial-tier fast path (collapse the pipeline, not just the panel, for trivial tasks) — drafted, will land separately.
🤖 Generated with Claude Code
https://claude.ai/code/session_016pXZvhEoE6Lz3DpBSL3n1Y