HM outer-map selection gate — 27 September 2026

The six-section specification freezes the London et al. (2018) Eq. (5)/(8) cutoffs: 0.018 R/Q for amplitude and 0.014 R/Q for phase. The new public london_2018 option selects these outer-map joins for the D carrier; the default phenomd_runtime retains the established reverse pair. The scalar reference receives QNM frequencies as inputs and imports neither frontend.

On len, CPU FP64, the integrated code at 833ba03 passed 67/67 in 17.44 s for tests/test_imrphenomhm_frequency_map_math_reference.py and tests/test_imrphenomhm_phase_offsets_math_reference.py, using JAX_ENABLE_X64=1 JAX_PLATFORMS=cpu PYTHONPATH=src:tests OMP_NUM_THREADS=1 and --reference-device=cpu -ra. The gate checks public (3,3) Torch/JAX class and dispatcher calls, map values and one-sided slopes at both joins, and identical complex mode values when the prior default is selected explicitly. The later integrated NSBH/HM focused run passed 142/142.

The expanded selection gate on len, CPU FP64, passed 84/84 map-reference tests in 32.89 s. The map and neighboring phase-offset reference files passed 111/111 in 34.19 s with the same environment and --reference-device=cpu -ra. The new public checks sample all six positive mode labels, two mass/spin inputs, and both cutoff choices through Torch and JAX. They compare captured map values and one-sided slopes to the scalar reference supplied with native QNM frequencies and compare omitted-option complex modes with explicit phenomd_runtime. The default remains unchanged.

This evidence qualifies the named outer-map selection component at the sampled points. It does not qualify QNM fits, carrier joins, phase offsets, amplitude calibration, projection, or the whole HM waveform.