IMRPhenomHM SXS:BBH:3619 projected-strain holdout — red receipt
This case is infeasible for qualification under its frozen protocol. Both
Torch and JAX meet the two model/NR projected-match and amplitude gates, but
the Lev2–Lev3 numerical-resolution screen fails at both source angles. The
protocol was committed
as d3a2ef4b313c582efea896a44a512fea55284c15 before either strain file
was downloaded or decoded; its clarification of the same-simulation
resolution screen, 3427ebfa554c16fcd0b33d7de3afc4846694506e, was also
committed before download. Neither gate nor conditioning was altered after
the data were seen. The earlier 2265
and 3631 phase-locked six-mode
gates remain red.
Source and metric audit
The numerical source is the SXS Collaboration's SXS:BBH:3619v3.0 from the official CaltechDATA record. The case was selected from the official version-3 metadata before strain retrieval. The Lev3 metadata give q≈5, negligible spins, and a 2021 run date, after the 2018 London et al. model paper. The six pinned Lev3/Lev2 strain, sidecar, and metadata files passed their official size and MD5 checks before the SXS decoder read them. Their SHA-256 digests, in protocol order, are:
| Source | SHA-256 |
|---|---|
Lev3:Strain_N2.h5 |
274557fb43434ed93c0a5f2c989be152216ec256bb8888a23fbfbfd8f20feaf9 |
Lev3:Strain_N2.json |
eb86644dead5a85a6af12250ad2d4ed1bca2845bae083a20877af57987320bc8 |
Lev3:metadata.json |
9226d3622cfc51e96a4841724168157c5a6b6c22db8d08bfeaa45fb95fbfd085 |
Lev2:Strain_N2.h5 |
251ae2c0e69159d59d94ff9e1dace833b81bdab639525d7eb67d6ab0dec7947f |
Lev2:Strain_N2.json |
5d7ad5314930b12965b6faa4f8ccfabf76609cc813ee38757fe3f46040d75c9c |
Lev2:metadata.json |
8b1305d0c7f1f5fece402b0e1c65f3ac00461b1cc41534ca08ae6202a4901988 |
Both sidecars record space translations, boosts, and memory additions; no
correction was reversed. The memory integration starts differ: Lev3 342.5 M
and Lev2 200.5 M. This version-3 correction context follows the
SXS Catalog 3 description. The SXS
decoder reported inertial spin--2 strain and all required 2≤ell≤5 modes.
The prepared external fixture SHA-256 is
21b6e75f133e4ddc3b3c8e55e52fae6afa99c5e4103c38245ed283e2a587bdcb.
The match follows the paper's detector-projection comparison: fixed intrinsic
parameters and inclination, with template arrival time, orbital phase, and
polarization optimized against full NR. It uses the documented PyCBC aLIGO
design PSD generator at
validation time, rather than copying a LAL array or implementation. The PSD
samples used here have SHA-256
0d20d2092bfaeb6073b0d656f4a3eb495bc3adbeb350ef279420cb4ffd0d67ea.
The committed known-shift check recovers a synthetic arrival exactly. At each
reported optimum, a direct 4 Re Σ a* b/S_n evaluation agrees with the FFT
maximizer to 1.2e-16. All four differential-evolution seeds agree within
4.5e-11 or better. Thus neither the FFT sign/scale nor a missed optimizer
mode explains the red result. This is a paper-like one-case test, not a
reproduction of London's full ensemble.
Frozen-gate result
All values below are Torch; JAX model matches differ by at most 2.9e-15 and
share the same NR resolution statistics. M is the optimized model/Lev3
detector match (gate ≥0.980); A is its PSD-weighted norm ratio (gate
0.85–1.15); R is the Lev2–Lev3 match with frozen time/polarization
optimization but no orbital-phase retuning (gate ≥0.995).
Source (iota,psi) |
M | A | R | Model gates | Resolution gate |
|---|---|---|---|---|---|
(0.7,0.0) |
0.996195851379 | 0.961627051 | 0.991880051 | pass | fail |
(1.2,0.4) |
0.985979162286 | 0.850460134 | 0.980142976 | pass | fail |
The resolution optima have candidate/Lev3 norm ratios 0.983814632 and
0.773284683; their arrival shifts are -0.0106201171875 and
-0.010498046875 s. The resolutions' (2,2) phases agree within about
5e-5 rad at common t=342.5 M, but differ by about 0.60 rad at common
t=7000 M; their peak times are about 21.9 M apart.
After seeing the red result, a separate exploratory diagnostic also
optimized one common orbital phase for all Lev2 NR modes, in addition to
arrival time and polarization. Its full-mode basis reconstructed the frozen
Lev2 plus/cross fixture to relative maximum errors below 4.2e-13. With
that added freedom the best Lev2–Lev3 matches become 0.999884070 and
0.999842333, with norm ratios 0.999932 and 0.995088; the best common
orbital phases are about -0.408 and -0.399 rad modulo 2π. Three of four
seeds at the second angle found the best value; the fourth was lower by
2.56e-5. This strongly supports an accumulated orbital phase difference as
the main reason the frozen no-orbital-retuning screen fails. It also shows
that the NR shapes can agree closely under a standard alignment. It is not a
predeclared gate and cannot turn this red holdout into a qualified pass.
The complete external exploratory record is
data/hm_sxs_3619_resolution_orbital_diagnostic.json, with basis SHA-256
25a4271ff5c4442a2e19c97178c56e54960d4344c330026d25b31a5a1329b170.
Reproduction and disposition
Host len; cwd /tmp/diffgw-hm-sxs-3619-20260927/repo; external data in
/tmp/diffgw-hm-sxs-3619-20260927/data. Preparation used Python 3.12.3,
NumPy 2.5.3, SciPy 1.18.1, SXS 2026.0.6, and SymPy 1.14.0. Evaluation used
NumPy 2.3.5, SciPy 1.16.3, PyCBC 2.11.0, Torch 2.14.0, and JAX 0.11.2.
The raw files and derived arrays stay outside Git. Preparation command:
PYTHONPATH=src:tests /tmp/diffgw-hm-sxs-2265-20260927/.venv/bin/python \
scripts/prepare_imrphenomhm_sxs_3619_projection.py \
--data-dir /tmp/diffgw-hm-sxs-3619-20260927/data
Preparation PID 688517 completed successfully; its log is
/tmp/diffgw-hm-sxs-3619-20260927/prepare.log. The final strict command was:
env JAX_ENABLE_X64=1 JAX_PLATFORMS=cpu OMP_NUM_THREADS=1 \
OPENBLAS_NUM_THREADS=1 DIFFGW_REQUIRE_HM_SXS_3619_PROJECTION=1 \
DIFFGW_HM_SXS_3619_PROJECTION_DIR=/tmp/diffgw-hm-sxs-3619-20260927/data \
PYTHONPATH=src:tests /tmp/diffgw-waveform-library-1315f6e/venv-latest/bin/python \
-m pytest -q -s tests/test_imrphenomhm_sxs_3619_projection.py -ra
Runner PID 810339 finished with exit 1: 2 failed, 1 passed in 17.69 s.
The log and exit marker are holdout-test.log and holdout-test.exit beside
the remote repo; complete machine-readable metrics are the external
data/hm_sxs_3619_{torch,jax}_projection_metrics.json files. No job remains
active; there is no next check or stop command. During a rerun, stop an active
PID with kill <PID> on len. The source-only invocation yielded 1 pass and
2 expected skips. Ruff passed for the new Python files.
The historical exploratory resolution diagnostic used
/tmp/diffgw-hm-sxs-3619-20260927/data, first preparing alternatives in
the SXS environment and then scoring them in the PyCBC environment, each
with PYTHONPATH=.:src:tests. PIDs 931927 and 939587 both completed
with exit 0; logs and exit markers are
diagnostic-{prepare,score}.{log,exit} beside the remote repo. Its
one-off replay script is not maintained. It did not alter the fixture or
strict test; the maintained preparer and holdout test above remain the
frozen workflow.
This red receipt is retained without gate changes. A future numerical reference or resolution-screen design must be predeclared against an unseen case before any new qualification claim; the 3619 result cannot be rescored as an independent pass after inspecting it.