Repository navigation
feat(t3): bound the review loop and add a verifier - #4
Merged
Merged
Conversation
Only critical and high findings block, later rounds review the fix diff with one cross-family reviewer, a verifier runs the real artifact after each fix, and the loop stops after three rounds. Unnamed research delegates run on the how explorer entry. Release 1.0.6.
…fects Review found the severity gate named a level interrogate never emits, and the cap bounded review rounds while failed verifications could loop fix rounds. The verifier gains could-not-run, the implementer verifies before review, and a multi-round loop ends by running correct on the defect class.
Drops verify-before-review, the single-reviewer choice for later rounds, and the post-loop correct step: each came from one incident, and the reviewer choice was contradicted the same afternoon. Later rounds keep the panel and narrow the diff instead.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
An audit of five days of t3-pstack delegation, about $2.6k at API prices, found roughly 40% of it wasted, mostly in how implement, review, and fix loops ran rather than in model choice. Nit cascades ran on inputs that cannot occur, fix rounds repaired earlier fixes, the full panel re-reviewed small fix diffs, and some review seats added no unique accepted finding. Verifiers that ran the real artifact after each fix returned the most per dollar.
t3-tools.mdadds a review loop for Bug fix, Feature, and Refactoring. Only findings the lead puts under interrogate's Act on block, includingwarningcorrectness findings; accepted nits are batched into one final run or dropped, and each dismissal states its reason. Round 1 reviews the whole change with theinterrogate reviewerspanel, and later rounds run the same panel on the fix diff only. After every fix round afeature, refactoringverifier runs the real artifact and reportspass,fail, orcould not run;failblocks, andcould not rungoes to the user. After three fix rounds the lead stops and reports, and the user decides. Each rule answers a pattern the audit measured across several threads; rules drawn from a single incident were left out.how explorerentry.tests/t3-tools.test.mjschecks that every rolet3-tools.mddispatches to is onemodels.jsondeclares, since that file is synced from upstream and a merge can rename a role.