Prespecify an author-audited synthetic benchmark route - #36
Merged
Merged
Conversation
mj3b
changed the base branch from
experiment/adjudication-record-v2
to
main
September 26, 2026 03:58
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Independent human reviewers are unavailable. Add a separately named, prespecified author-audited synthetic route without changing or claiming completion of the original two-reviewer route. The author confirmed that no live Jev outputs have been observed before this amendment.
The frozen protocol narrows claims to agreement with the author's synthetic specification, controlled-change sensitivity and nuisance stability. It explicitly defers independent reliability and real-world validity. Workbooks cover all 1,654 determinations through 216 distinct question/excerpt units and 60 paired comparisons. Two actual author passes are separated by at least seven days (a practical prespecified interval, not a validated memory washout). All disagreements and concerns require preserved issue records; material defects block progression and require a new version.
The offline checker validates source bytes, coverage, saved pass hashes, dates, declared exposure, issue accounting and recorded author consistency. Blank forms cannot pass. The original v2 two-reviewer records remain open; later author-only interpretations must be separately labeled. Optional fresh-context AI criticism prompts are specified but no AI criticism or author review is claimed as performed.
Validation: all eight new tests and 12 existing adjudication tests passed; original evidence freeze verified. Blank workbook exported. No live calls. GitHub CI also runs the existing offline Jev and repository suites.
Author review is pending, and this PR does not authorize live execution. Preserve the original corpus, oracle, metrics, release findings and independent-review protocol. This PR is stacked on #35; merge #35 before changing this PR's base to main. Maintenance #34 is separate. Publish the planned v0.17.1 release only after the agreed author-audit stage is completed or its release scope is explicitly revised to disclose a pending audit.