Kept in this directory rather than in a hunt-level file, because six sessions share one worktree and a shared file invites lost writes. Validated by test_positivity_geometry.py::test_run_manifests_validate with the parser of tests/test_huntspec.py.
Estimate before spending (CLAUDE.md compute rule 3): one brute-force count over F_{13^4} (28561 elements) takes about 0.4 s and the whole exploration about 15 s on one process, so nothing here needed a checkpoint or a detached job.
id: rh_moonshot-positivity_geometry-2026-10-21-construct
hunt: rh_moonshot
started: 2026-10-21T21:40+00:00
finished: 2026-10-21T22:32+00:00
ran:
- .venv/bin/python hunts/rh_moonshot/positivity_geometry/explore.py curves
- .venv/bin/python hunts/rh_moonshot/positivity_geometry/explore.py fakes windows
- .venv/bin/python hunts/rh_moonshot/positivity_geometry/explore.py quadratic
- .venv/bin/python hunts/rh_moonshot/positivity_geometry/explore.py rivals
- .venv/bin/python -m pytest -q -n0 hunts/rh_moonshot/positivity_geometry
outcome: the Frobenius-span Gram matrix has one positive eigenvalue for all 8 test curves and two for every planted fake, and the golden fake meets every arithmetic hypothesis tested while violating RH
artifacts:
- hunts/rh_moonshot/positivity_geometry/frobenius_span.py
- hunts/rh_moonshot/positivity_geometry/explore.py
- hunts/rh_moonshot/positivity_geometry/test_positivity_geometry.py
- hunts/rh_moonshot/positivity_geometry/NOTES.md