Skip to content

Latest commit

 

History

History
28 lines (27 loc) · 2.41 KB

File metadata and controls

28 lines (27 loc) · 2.41 KB

Progress

2026-04-03

  • Inspected the repository and confirmed it is an early-stage workspace with only a README.
  • Added a project-wide workspace instruction file at .github/copilot-instructions.md.
  • Created lightweight planning files to track the bootstrap task.
  • Verified that no competing AGENTS.md or other workspace instruction file was introduced.
  • Added the initial src-layout package, shard assembly, 1-norm condition estimation, and compare CLI.
  • Added real-matrix pytest fixtures and tests that exercise Matrix Market parsing and shard assembly.
  • Fixed a failing condest test assertion after the first real pytest run.
  • Verified the suite with python -m pytest after the fix.
  • Formalized the versioned shard manifest schema and added iterative benchmark/report output.
  • Added a second real Matrix Market sample and a reusable block-diagonal shard manifest fixture.
  • Added a public condest_2 interface and verified it against np.linalg.cond(..., p=2).
  • Removed the hidden fallback path from condest_2 so the 2-norm computation is explicit and single-path.
  • Added Krylov-based condition-number estimators, log-log benchmark trend fitting, and the scl-benchmark CLI.
  • Verified the full suite with python -m pytest after the Krylov benchmark feature.
  • Re-ran the full suite after cleanup; all 12 tests still pass.
  • Generated the final Krylov benchmark report and fitted chart under docs/reports/.
  • Replaced the report benchmark source with a generated anisotropic Poisson family and regenerated the chart/report.
  • Added structured sparse matrix family generators plus stronger correctness-validation cases covering Poisson and coupled complex systems.
  • Verified the full suite after the realistic-family benchmark changes; all 46 tests pass.
  • Added a nonsymmetric convection-diffusion-2d generated family for more realistic non-SPD-style benchmark coverage.
  • Added multi-family Krylov benchmark suite APIs, markdown rendering, and combined figure output.
  • Generated a single suite report covering anisotropic Poisson, coupled diffusion, and convection-diffusion families.
  • Expanded .gitignore so Python caches, egg-info, build outputs, and local environments stay out of version control.
  • Re-ran the full suite after the suite-report and review hardening changes; all 52 tests pass.
  • Completed three consecutive project-wide review rounds with no new findings after the last fixes.