AntFleet

Disagreement · f03404f9-anthropic-2

Momentum verdict formula divides by zero / undefined when 30d data is missing but still threshold-compares

solo Opus
repo 6f7fc663·PR #22·reviewed 1 week ago

Opus finding

Momentum verdict formula divides by zero / undefined when 30d data is missing but still threshold-compares

mediumbughigh
  • skills/ai-framework-watch/SKILL.md:151-156
  • skills/ai-framework-watch/SKILL.md:124-131
Verdict priority 2 requires comparing `star_delta_7d` against `3× 30d-implied weekly average`, but step 4 explicitly says the 30d delta renders `—` until state is ≥21 days old. There is no rule for what verdict 2 evaluates to during weeks 1-3 (the comparison is against `—`). Implementations will either (a) treat `—` as 0 and fire MOMENTUM SHIFT on every framework with ≥200 weekly stars (false positives in cold-start period), or (b) silently skip the verdict (the safer behavior) — but the spec doesn't say which. Given the digest is read as 'trustworthy weekly anchor', this ambiguity is high-impact.

Recommendation

Add an explicit rule in step 6 priority 2: 'Skip this verdict if 30d-implied weekly average is unavailable; fall through to priority 3.' Or define a fallback threshold for weeks 1-3.

Other reviewer

The other reviewer flagged nothing in this file/line range.

Why this didn't post

This finding didn't meet AntFleet's unanimous agreement threshold. Both frontier models review every PR independently; only findings they both flag with the same severity and category are posted to the PR. This one fell through.

read the methodology →