A directorate record, not a hunt result: six defects in recorded claims were reproduced, including proven_sign returning a wrong nonzero sign on numpy.float32, a strategic coefficient-blindness claim was corrected, and two positives held under attack.
hunts/director_run (Record 6 of 98 in chronological sequence)
Guiding Question
What genuine mathematical information, including defects in recorded conclusions, can the laboratory extract when it treats this repository as an untrusted scientific artifact?
Method & Verification
Nine investigators with conflicting mandates ran in parallel, the generator of a claim never judged it, and each named defect was reproduced by the director before being written down.
Lineage & Relationships
Primary Sources (at pin 8fa46e134)
Editorial Notes
No HuntSpec, HANDBACK, or RESULTS.md at the pin. Question filled from MISSION.md. Disposition is qualified because the case log states this is not a hunt in the usual sense and not a result. Evidence grade is measured for the reproduced defects; the Lean rebuild (8715 jobs, zero sorrys) is a separate kernel-checked process check and is not treated as the composite grade.
Date Provenance
commit 5eed14344b651422229d542c6272916c030ae599, hunts/director_run/, author 2026-08-11T00:22:21Z