Repository navigation
Docs: the maintenance loop and the rule-mining track - #21
Merged
Merged
Conversation
…ng track in the roadmap Records the eight-step process the 2026-10-01 PR triage and rule-check converged on, maps each step to the findings architecture, defines the labelled cases, the per-rule metric and the fortnightly cadence, maps defect clusters to taxonomy checks, and specifies the four skills to distil: pr-triage, rule-mining, rule-check, policy-question. Milestone 2 now starts from the existing corpora rather than a new collection. Co-Authored-By: Claude Fable 5.1 <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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Feeds the learnings from the 2026-10-01 Inspect Evals PR triage and rule-check into the plan.
docs/maintenance-loop.mdrecords the eight-step process the agents converged on (crawl, cluster, mine a rule, sort, lint issue, policy issue, rule-check, triage actions) and where each step lives in the findings architecture. It defines the loop we can hill-climb: labelled cases from merged defect-fix PRs (cases.yamlwith pre and post revisions, class, expected finding, verdict), a per-rule metric (recall at pre, false positives at post, precision on main, cost for auditor items), a fortnightly cadence, and the rule that an open policy question blocks a rule from shipping. It maps the defect clusters togl-audit@1checks so rule-level recall rolls up to dimensions, and specifies four skills to distil with frontmatter contracts:pr-triage,rule-mining,rule-check(the 1 October brief, kept),policy-question.Roadmap. Milestone 2 starts from the existing corpora (about sixty scoring cases across the 09-09 batch and the two triages) instead of a new collection, and runs the "not lintable" auditor items against the same cases. A new "Rule mining" track section with the cadence and the forward-deployed-engineer output contract. Two open decisions added: where
cases.yamllives, and how a merged PR gets its class.Docs only.
🤖 Generated with Claude Code