HM ringdown phase-shape gate — 2026-09-27

The bounded hm.phase.ringdown.eq9-c1.v1 gate passed on len, in /home/xangma/tmp/diffgw-hm-eq14-phase-20260927, using a copy of the isolated worktree based on 9cbb2a2. The maintained test is tests/test_imrphenomhm_ringdown_phase_math_reference.py; it calls the production Torch/JAX helpers against the independent mpmath scalar.

Command: PYTHONPATH=$PWD/src JAX_ENABLE_X64=1 JAX_PLATFORMS=cpu /home/xangma/.cache/diffgw-taylorf2-hm-venv/bin/python -m pytest -q tests/test_imrphenomhm_ringdown_phase_math_reference.py --reference-device=cpu -ra

Result: 14 passed in 4.13 s. A final combined run adding tests/test_waveform_definitions.py passed 26 tests in 8.51 s. Environment: Python 3.12.3, Torch 2.14.0+cpu, JAX 0.11.2, mpmath 1.3.0, pytest 9.1.1. The gate covers the paper limit, distinct native D/HM QNM input convention, values, frequency and \(\alpha_4\) gradients, C1 join, and 22 control. The --require-math-reference pytest option belongs to the full CI suite: its focused-file invocation exits at collection because the other required math-reference files are absent from that invocation; it does not indicate a test failure.

No whole-model accuracy claim follows. QNM and alpha fits, gauge/attachment beyond this correction, and LAL/NR comparisons remain separate gates. See the component specification for the source and precise boundary.