Summary
After #5827, /discovery:research outcome-gate row 7 reads: "Every accepted claim is HIGH confidence; a MEDIUM or LOW claim listed in the Gaps section is not accepted". Rows 7 and 12 are both quantified over accepted claims. An artifact that lists every claim as a Gap therefore passes both rows with zero accepted claims, and nothing in the gate or the artifact tells the reader that the run settled nothing.
Row 4 also says "Every claim has ≥2 INDEPENDENT current corroborators". The intent is "every accepted claim", because a Gap claim is by definition not corroborated. As written, row 4 contradicts the Gap route that rows 7 and 12 allow.
Evidence (PR #5827 head)
plugins/discovery/skills/research/SKILL.md:75, row 4: "Every claim has ≥2 INDEPENDENT current corroborators ..."plugins/discovery/skills/research/SKILL.md:78, row 7: "Every accepted claim is HIGH confidence; a MEDIUM or LOW claim listed in the Gaps section is not accepted"plugins/discovery/skills/research/SKILL.md:83, row 12: "Every accepted claim follows jointly from its cited sources ..."- No row requires at least one accepted claim, and none requires the artifact to state that it accepted zero claims.
Related, not a duplicate: #5805 covers the acceptance-bar wording and single-publisher facts.
Design question
Pick one of these for an artifact with zero accepted claims:
- (a) A gate row that fails it.
- (b) A row that requires a verdict line in the artifact (for example
accepted: 0 — inconclusive) and leaves it passing.
- (c) An explicit "inconclusive" outcome distinct from pass and fail.
Recommendation: (b). Basis: judgment. A research run that honestly finds nothing is a valid result. A run that does not tell the reader it found nothing is the defect.
Acceptance criteria
Summary
After #5827,
/discovery:researchoutcome-gate row 7 reads: "Every accepted claim is HIGH confidence; a MEDIUM or LOW claim listed in the Gaps section is not accepted". Rows 7 and 12 are both quantified over accepted claims. An artifact that lists every claim as a Gap therefore passes both rows with zero accepted claims, and nothing in the gate or the artifact tells the reader that the run settled nothing.Row 4 also says "Every claim has ≥2 INDEPENDENT
currentcorroborators". The intent is "every accepted claim", because a Gap claim is by definition not corroborated. As written, row 4 contradicts the Gap route that rows 7 and 12 allow.Evidence (PR #5827 head)
plugins/discovery/skills/research/SKILL.md:75, row 4: "Every claim has ≥2 INDEPENDENTcurrentcorroborators ..."plugins/discovery/skills/research/SKILL.md:78, row 7: "Every accepted claim is HIGH confidence; a MEDIUM or LOW claim listed in the Gaps section is not accepted"plugins/discovery/skills/research/SKILL.md:83, row 12: "Every accepted claim follows jointly from its cited sources ..."Related, not a duplicate: #5805 covers the acceptance-bar wording and single-publisher facts.
Design question
Pick one of these for an artifact with zero accepted claims:
accepted: 0 — inconclusive) and leaves it passing.Recommendation: (b). Basis: judgment. A research run that honestly finds nothing is a valid result. A run that does not tell the reader it found nothing is the defect.
Acceptance criteria
research-verifier.mdgrades it.contract.test.shpin and an eval case cover an artifact whose every claim is a Gap.plugins/discoveryis version-bumped and has a CHANGELOG entry.