AntFleet

Disagreement · 9adffca7-anthropic-1

Step 4 momentum verdict can divide by zero / produce bogus % when stars_30d_ago is unknown

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

Opus finding

Step 4 momentum verdict can divide by zero / produce bogus % when stars_30d_ago is unknown

mediumbugmedium
  • skills/ai-framework-watch/SKILL.md:135-142
  • skills/ai-framework-watch/SKILL.md:152-156
The MOMENTUM SHIFT verdict requires a 30d-implied weekly average, but step 4 says 30d delta renders `—` if state is younger than 21 days. The spec does not say to skip this verdict in that case, and the `{pct}% wow` formula is undefined when 30d data is missing or zero. For the first ~5 weeks every run could either silently never fire #2 (acceptable) or compute a divide-by-zero / nonsensical pct (failure). The rule is ambiguous and likely to be implemented inconsistently by the LLM.

Recommendation

Make the precondition explicit: MOMENTUM SHIFT requires `stars_30d_ago` to be known AND non-zero; otherwise fall through. Define `pct = star_delta_7d / (stars_30d_ago_to_now_avg_weekly) * 100` with explicit zero-guard.

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 →