feat(resolve): collapse duplicate cross-reviewer issues via DUPLICATE triage verdict - #307
Open
dean0x wants to merge 6 commits into
Open
feat(resolve): collapse duplicate cross-reviewer issues via DUPLICATE triage verdict#307dean0x wants to merge 6 commits into
dean0x wants to merge 6 commits into
Conversation
… triage verdict Add a DUPLICATE verdict to the /resolve pipeline so the same underlying defect flagged by multiple Review agents counts as one issue in resolution-summary.md, de-skewing fp_ratio convergence math in code-review without any parser change. Changes: - triage.md: duplicate grouping pre-pass before the disposition matrix; DUPLICATE verdict section in output; duplicate_of attribute (non-chainable); security gate applies to the whole group (applies ADR-006, avoids PF-026) - resolve.mds: DUPLICATE added to every verdict-enumeration site — Phase 2 parse list, completeness assertion, Phase 3 batch exclusion, Phase 9 manage-debt note, Phase 9b-1 THREAD_MAP mapping, Phase 10 Results table, Output artifact Statistics (Duplicates Collapsed row + ## Duplicates section), Architecture block, Edge Cases (avoids PF-024 spawn↔op seam; additive-only per ADR-006) - tests/build-mds.test.ts: §16b guards pin DUPLICATE, duplicate_of, "| Duplicates Collapsed | ", and "## Duplicates" in compiled resolve.md (PF-018) Co-Authored-By: Claude <noreply@anthropic.com>
Remove filler word ("immediately") from triage.md pre-pass rule.
In resolve.mds: simplify the completeness-invariant sentence for DUPLICATE
entries, drop the redundant "excluded from all batches" clause (implied by
"never dispatched"), and condense the PF-024 parenthetical in Phase 9b-1.
No behaviour or contract change — all §16b parser-contract literals preserved.
Self-review fixes on the DUPLICATE triage verdict:
- triage.md: responsibility 3 said "apply the matrix" to every issue,
contradicting the pre-pass ("matrix applies to primaries only") and
omitting DUPLICATE from ledger completeness. Reworded in place.
- triage.md: a group mixing security and non-security findings could
elect the non-security report as primary (it may be the "most specific
and complete"), handing the Code agent a style description for a
security defect. Security member is now always the primary; the
whole-group Security Gate remains the backstop.
- resolve.mds: Phase 5 never stated that DUPLICATE rows are exclusive to
## Duplicates. Without it, a duplicate of a FALSE_POSITIVE primary
could be listed in ## False Positives while the unique-only Statistics
row counted just the primary -- section and row disagreeing, and
git.md manage-debt reading duplicates out of ## Fix Separately.
- resolve.mds: the Statistics note swept `Total Issues` into the
unique-only rule, so Total would have excluded collapsed duplicates
and stopped equalling the sum of its rows. Total now counts every
triaged issue; all other rows stay unique-only.
- resolve.mds: completeness assertion demanded a `duplicate_of` field
while the ledger emits a `Duplicate Of` column -- an enum/field-name
mismatch across the spawn/op seam whose failure mode is a hard abort
(avoids PF-024). Assertion now names the ledger surface.
- resolve.mds: Triage spawn directive named only the matrix; now names
the collapse-then-matrix order it parses a DUPLICATE bucket from.
- tests: new tests/resolve/duplicate-verdict.test.ts pins the producer
side of the seam (build-mds §16b pinned only the caller side, so the
agent could stop emitting the bucket with every guard green).
Byte-stable parser contract untouched: Fixed / False Positive /
Deferred row labels and ## Fixed Issues / ## False Positives column
layouts are unchanged (applies ADR-006, additive-only).
…ambiguate statistics note - docs/commands.md: append DUPLICATE to the disposition enumeration (step 1) and note in step 7 that duplicate cross-reviewer reports are collapsed so Statistics counts reflect unique issues - CHANGELOG.md: add ### Changed entry under [Unreleased] describing the new DUPLICATE verdict and its effect on resolution-summary counts - resolve.mds: reword the ambiguous Statistics note — "every other row above Duplicates Collapsed" → "every row between Total Issues and Duplicates Collapsed" to make clear that Total Issues itself is not a unique-only count Co-Authored-By: Claude <noreply@anthropic.com>
…TE edge case In triage.md Duplicate Grouping Pre-Pass step 2, add a parenthetical tying 'security member' to the Security Gate so the term is grounded on first use in the pre-pass. In resolve.mds Edge Cases table, extend the DUPLICATE-without-duplicate_of row label to also name the chained-to-another-DUPLICATE case, matching the Phase 2 completeness-assertion prose at ~line 189. Co-Authored-By: Claude <noreply@anthropic.com>
…t-name guard 'cross-reviewer' contains 'reviewer' which trips the case-insensitive retired-name scan for the pre-PR-#253 'Reviewer' agent name. Reword to 'multiple Review agents' which carries the same meaning and passes the guard.
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
The same underlying defect flagged by 3 Review agents was counted as 3 issues — fixed once, reported as "Total Issues 3 / Fixed 3". This inflated all Statistics rows and skewed the
fp_ratioconvergence formula in/code-review(which divides false-positive count byfp_count + fixed_count + deferred_count). The fix is a DUPLICATE triage verdict that collapses cross-reviewer duplicates before the disposition matrix runs.Changes
src/assets/agents/triage.md— duplicate grouping pre-pass added before the blast-radius matrix (not as a matrix row — this is a relation between issues, not a priority). Groups issues by same root cause, selects the most complete report as primary, applies the matrix only to the primary. Non-primary members get DUPLICATE withduplicate_of: <primary-id>(non-chainable). Security gate applies to the whole group. Kept minimal (avoids PF-026).src/assets/commands/resolve.mds— DUPLICATE added to every verdict-enumeration site per PF-024 (spawn↔op seam):duplicate_of→ Triage failure with retry-then-abort)| Duplicates Collapsed | {n} |row added to Results table| Duplicates Collapsed | {n} |row added;## Duplicatessection added with| Issue | Duplicate Of | File:Line |tabletests/build-mds.test.ts— §16b guards (RED→GREEN TDD) pinDUPLICATE,duplicate_of,| Duplicates Collapsed |, and## Duplicatesin compileddist/commands/resolve.md.Parser-Contract Safety
All existing Statistics row labels (
Fixed,False Positive,Deferred) and section headings (## Fixed Issues,## False Positives) are byte-stable — labels unchanged, column layouts unchanged. TheDuplicates Collapsedrow is strictly additive and the/code-reviewconvergence parser ignores it. All parser-read rows now count unique (non-DUPLICATE) issues, which de-skewsfp_ratiowithout any change to the formula or the parser.Reviewer Focus Areas
duplicate_ofmust reference a non-DUPLICATE id — enforced in the completeness assertion and in triage.md pre-pass rule 4## Duplicatesis additive; existing## Fixed Issues/## False Positivescolumn layouts are unchanged