Physics release verification
Quantitative use of the periodic2 model requires both portable CI and the HPC release gate:
make test-l2make test-physics-releaseThe HPC gate sequentially runs the three L1 tests needed for convergence rows, L3-heavy tests,
far-correction correctness, MPI ledger, and MPI periodic-cache concurrency. It does not repeat the full L2 suite
owned by portable CI.
Its manifest records elapsed time and GNU time peak RSS for every gate and enforces an 8 GiB default budget,
overridable with BEACH_RELEASE_MAX_RSS_KB. It also regenerates convergence.csv from standardized test output
and fails when a required convergence category is missing.
Per-target Fortran timings are written to test_l3-target-timings.csv and
far_correction-target-timings.csv.
Run periodic far-correction correctness checks with make test-fortran-far-correction.
Runtime comparison is separated from debug correctness and runs under the release profile with
make test-fortran-benchmark.
The reference values and interpretation are maintained in docs/PhysicsReleaseVerification.md.
They are small correctness fixtures and do not replace mesh, time-step, FMM, outer-grid, and sampling studies
for a production simulation.