IMRPhenomT dominant inspiral time adapter: len receipt (2026-09-27)

T-INSP-TIME-ADAPTER-v1 passed its strict bounded gate: 16 passed in 60.15 s. No native implementation, defaults, or frozen failure list changed. The six-section component spec was written before the scalar implementation and test.

The run used host len, cwd /home/xangma/tmp/diffgw-phenomt-fit-inputs-20260927, Python 3.12.3, Torch 2.14.0+cpu, JAX 0.11.2, mpmath 1.3.0, and pytest 9.1.1. JAX x64 and CPU were explicitly enabled. The exact command was:

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

The test PID was 3874494 and has exited. Output is preserved at /home/xangma/tmp/diffgw-phenomt-fit-inputs-20260927/phenomt-time-adapter-pytest.log. The stop command while it ran was kill 3874494 on len; no further check is pending. The transferred scalar and test files had matching local/len SHA-256 values 4e2b67c9afd764c1485e10093014658b6f49ac8d215532dc686fc908b6c03225 and e1517db5080c0248bf0b5d429d562f2ca370c052da15bfb09cce9952ee66b489.

Three source-backed intrinsic points and three inspiral frequencies per point were independently inverted in the scalar theta domain and compared with both native backends' time roots and forward frequencies. A separate integrated-frequency phase check covered evolution, reference-frequency rotation, and 2*phic; public checks covered centered and uncentered grids, the lower-frequency mask, and omitted/zero f_ref resolution. Thresholds were frozen in the spec and unchanged after the first passing run.

This is a component-level agreement receipt, not physical calibration or full-waveform qualification. The pinned MIT author internals.py uses a t0 shift when inverting before tEarly; the native code does not. The test targets lie after tEarly and before the inspiral join. The THM paper specifies the frequency series but not the public sampling adapter. Pre-tEarly inversion, merger/ringdown, amplitude, auto-duration, HM/precessing adapters, and physical-error validation remain open.