AntFleet

Disagreement · ce6ce4c1-openai-1

Cohort definition for STALE conflicts with classification algorithm for never-run forks

mismatch
repo 6f7fc663·PR #10·reviewed 1 week ago

Primary finding

Cohort definition for STALE conflicts with classification algorithm for never-run forks

mediumdocs-gaphigh
  • skills/fork-cohort/SKILL.md:23-25
  • skills/fork-cohort/SKILL.md:69-70
  • skills/fork-cohort/SKILL.md:97-111
The table suggests a STALE case even "if no recent run record exists," which overlaps with COLD's "No Actions runs ever recorded" and conflicts with Step 5, which classifies empty last_run as ∞ and thus COLD. This discrepancy will confuse operators because narrative docs and output classifications won’t align.

Recommendation

Align the definition with the algorithm. Suggested fix: remove "even if no recent run record exists" from STALE and keep "No Actions runs ever recorded" strictly in COLD. Rephrase STALE as: "Last run ≥7 days ago and ≤365 days ago." Ensure examples match the implemented logic across the doc.

Counterpart finding

Bot allowlist is declared but never wired into list/classify steps

lowdocs-gapmedium
  • skills/fork-cohort/SKILL.md:215-216
  • skills/fork-cohort/SKILL.md:62-66
The constraint says bots should be "skipped from cohort rendering but still count in totals," but no step shows where the filter is applied. Step 2 only filters archived/disabled. Step 5 (classify) and step 8 (render) make no mention of the allowlist. An operator following the spec literally will either include bots in the rosters (violating the constraint) or filter at step 2 (deflating N_TOTAL, also violating the constraint). The two requirements (skip rendering, keep in totals) require a specific implementation point that isn't documented.

Recommendation

Add an explicit step or sub-step: "At step 5, classify bot-owned forks normally and include them in totals. At step 8 (movement subsections + POWER roster), skip any fork whose owner matches the allowlist." Or remove the constraint if it's aspirational.

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 →
← back to all disagreementsview public receiptssee unanimous findings + anatomies →
Tweet ↗