Record the 129x129 Ghia cavity results and report steps run per backend

[shaia]

The 129×129 Ghia results table in docs/validation/cavity-backends-validation.md still said TBD, and ROADMAP §6.1 listed it as the main open P0 item for v1.0. No new run was needed: since the Re=400/1000 cases landed, the EC2 workflow (gpu-validation.yml) has built with -DCAVITY_FULL_VALIDATION=ON and run every CavityBackend_* case on each push to master. This PR records those results and closes the Ghia item. It also covers two gaps found along the way: the Explicit Euler cases stop early (recorded, not fixed), and scripts/ec2-validate.sh never ran Re=400/1000 (fixed).

Results

Three EC2 runs gave identical values: 34853615452 (6e1b104), 34815324268 (23b196a) and 30116961115 (#201). AVX2, OpenMP and CUDA agree to four decimals:

Re	RMS_u	RMS_v	Step budget × dt (t)
100	0.0017	0.0024	50000 × 0.0002 (10)
400	0.0096	0.0328	60000 × 0.0005 (30)
1000	0.0299	0.0300	100000 × 0.0005 (50)

Every case is at least 3× under the 0.10 target, so the step budgets stay unchanged.

What changed

Test output

• test_backend_validation() now prints Steps run: N/budget stopped by KE threshold: yes|no KE residual: X for each backend. It only adds output; pass/fail logic is unchanged.
• "Stopped by KE threshold" means the per-step kinetic-energy change fell below 1e-8. It is not a convergence claim: the Euler cases below hit it with an undeveloped flow.
• The line avoids RMS_u:, SKIPPED and [FAILED], the strings the workflow's summary awk keys on. It is named Steps run: so it can't be confused with the Steps: header that main() already prints.

Explicit Euler cases stop early (recorded, not fixed)

• The new line shows the Euler cases ending through the harness's kinetic-energy exit (relative change per step below 1e-8). Their solvers cap dt at 1e-4, so the runs reach only t ≈ 1.1–1.2:
  • 129×129: 11,323 (AVX2) and 11,775 (OpenMP) of 250,000 steps
  • 33×33 CI: 11,300 of 25,000 steps (scalar)
• That is why their RMS_u (0.0957) is the same on both grids. The flow needs about 10–20 time units to reach steady state, so these runs pass the 0.15 target without a developed solution.
• Recorded as a Known Issue in the validation doc and as a new open item in ROADMAP §6.1.

scripts/ec2-validate.sh

• It ran test_cavity_backends with no filter, which skips the Re=400/1000 cases because they need an explicit filter. It now runs ctest -R '^CavityBackend_' --no-tests=error, which covers exactly the 12 entries registered in CMakeLists.txt.
• Under set -euo pipefail, a failing test binary ended the script before EXIT_CODE=$? ran, so the FAILED banner never printed. The status is now captured with || EXIT_CODE=$?.
• It schedules tests like CI: ctest -j 4, with OMP_NUM_THREADS = nproc / 4 unless --threads is given.
• Output is teed to validation_output.txt, and the backend, steps and RMS lines are printed as a summary at the end.
• The quick-start now uses --release, matching CI.

Docs

• docs/validation/cavity-backends-validation.md:
  • results table with run provenance;
  • Known Issue section;
  • the CI/CD section now describes the EC2 workflow;
  • stale projection_jacobi_gpu renamed to projection_gpu;
  • Next Steps rewritten.
• docs/validation/lid-driven-cavity.md:
  • full-validation settings (previously "3000–10000 steps" and a #define);
  • a 129×129 results summary;
  • the 33×33 CI section and Fast Mode step count (previously ~0.10 / ~0.08 RMS, 1500–4000 steps and "reaches steady state"). It now gives the measured 5000-step run: RMS 0.0382 / 0.0440, still far from steady state;
  • the pressure-solver list (previously "Red-Black SOR… GPU: Jacobi").
• docs/reference/solvers.md: measured RMS replaces "RMS error < 0.01", and the broken validation/… link is fixed.
• ROADMAP.md §6.1:
  • Ghia item ticked;
  • section note and CI vs Release table updated;
  • Euler follow-up added.
  • Cross-architecture consistency stays open: CavityBackend_Consistency compares CPU, AVX2 and OpenMP at 33×33 and never includes the GPU.
• CHANGELOG.md: Added and Fixed entries.

Verification

Windows, MSVC with OpenMP.

• Full-validation build (Release, AVX2 on, CUDA off, -DCAVITY_FULL_VALIDATION=ON):
  • euler_avx2, euler_omp and consistency pass and print the new line. Euler RMS matches EC2 exactly (0.0957 / 0.1293 and 0.0957 / 0.1277).
  • ctest -N -R "^CavityBackend_" lists exactly 12 tests.
  • The summary awk from gpu-validation.yml, run on that output, still produces correct rows.
• CI-mode build (Debug):
  • projection_scalar runs 5000/5000 steps with RMS 0.0382 / 0.0440, as documented.
  • euler_scalar passes and prints stopped by KE threshold: yes.
• bash -n scripts/ec2-validate.sh passes, and the new relative doc links resolve.
• Not verified:
  • the projection cases locally (their numbers come from EC2) and the GPU backend;
  • ec2-validate.sh end to end on Linux.
• Still to come: the Steps run: lines for the projection cases, in the first EC2 run after merge (or earlier if this PR is labelled gpu-test).

🤖 Generated with Claude Code

[Copilot]

🟡 Changes recommended

The new output mislabels a KE threshold as convergence, and the updated 33×33 section retains stale results and settings.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Records validated 129×129 Ghia cavity results and improves EC2 validation reporting.

Changes:

• Documents backend results, provenance, and the Explicit Euler limitation.
• Reports actual step counts and KE stopping status.
• Runs all CavityBackend_* tests with improved logging and failure handling.

File summaries

File	Description
tests/validation/test_cavity_backends.c	Adds step and KE-residual output.
scripts/ec2-validate.sh	Runs all backend cases and summarizes results.
ROADMAP.md	Records validation completion and Euler follow-up.
docs/validation/lid-driven-cavity.md	Updates validation configuration and results.
docs/validation/cavity-backends-validation.md	Adds full results, provenance, and known issue.
docs/reference/solvers.md	Updates measured RMS results and links.
CHANGELOG.md	Records validation and script changes.

Review details

• Files reviewed: 7/7 changed files
• Comments generated: 2
• Review effort level: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.