Runs · teal-sea

Manifests for the runs this hunt reports. The session that did the work on 2026-09-04 wrote none, so the first two were reconstructed at landing (2026-10-21) from the committed artifacts and the commit record. Their ran lists name only commands whose output is committed, and their times are bounded by commit times, not recorded. The third was recorded first-hand at landing. The Lean check that modal_check.py describes was never run, so it has no manifest; RESULTS.md and Symmetry.lean say so.

id: reversal_minimax-2026-09-04-adversary
hunt: reversal_minimax
started: 2026-09-04, time not recorded (reconstructed at landing)
finished: 2026-09-04, before 20:56 -0500 (commit 7415efa1b)
ran:
  - .venv/bin/python hunts/reversal_minimax/probe.py
  - .venv/bin/python hunts/reversal_minimax/verify.py --quadrature
outcome: closing the saved 2,200-cut pool under reversal gave a seven-orbit adversary whose Arb-checked upper bound 182022320819207/23000000000000000 lies below the recorded endpoint 0.007916857810 and the target 19791/2500000
artifacts:
  - hunts/reversal_minimax/adversary.json
  - hunts/reversal_minimax/verification.json

id: reversal_minimax-2026-09-04-counting-audit
hunt: reversal_minimax
started: 2026-09-04, time not recorded; the continuation it answers is recorded after commit 7415efa1b (20:56 -0500)
finished: 2026-09-04, before 22:02 -0500 (commit 1c4625b04)
ran:
  - .venv/bin/python hunts/reversal_minimax/counting_audit.py
outcome: exact incidence counts showed the early span estimate charging an interior gap (n-1)(m-1) times where (n-1)(m-n+1) is exact, with projected candidate gains 400/446400399 and 5/904171
artifacts:
  - hunts/reversal_minimax/counting_audit.json

id: reversal_minimax-2026-10-21-landing-replay
hunt: reversal_minimax
started: 2026-10-21T16:36Z
finished: 2026-10-21T17:10Z
ran:
  - .venv/bin/python hunts/reversal_minimax/verify.py --quadrature
  - counting_audit.report() compared in memory with counting_audit.json, without rewriting it
  - counting_audit.projection(4, Fraction(2330, 10**6), 432, 2500), for main's later four-point build
  - .venv/bin/python -m pytest -q -n0 -p no:cacheprovider hunts/reversal_minimax
outcome: verification.json regenerated byte for byte on the python-flint backend, the counting audit replayed with exact fields identical and float fields within 5e-14 relative apart from one near-singular eigenvalue, the coarse count reproduced main's m = 432 coefficient exactly, and all 18 hunt tests passed
artifacts:
  - hunts/reversal_minimax/RESULTS.md
  - hunts/reversal_minimax/test_counting_audit.py
  - hunts/reversal_minimax/test_probe.py