task_plan.md

Task Plan

Goal

Bootstrap the initial SparseCondLab package, validate it against real matrix fixtures, and add Krylov benchmark trend prediction for 1e6 DOF extrapolation.

Phases

• Discover existing conventions and documentation
• Read workspace-customization guidance
• Draft and add workspace instructions
• Review the generated instruction file for clarity and overlap
• Confirm no extra workspace instruction file was added
• Add package scaffold and project metadata
• Implement Matrix Market and NPZ loading
• Implement shard assembly and condest wrapper
• Add CLI compare command
• Formalize shard manifest schema
• Add iterative benchmark runner and flat report output
• Add extra real matrix sample and shard manifest fixture
• Add condest_2 interface and test coverage
• Remove hidden fallback from condest_2
• Run tests against the real matrix fixture
• Triage and fix any test failures
• Add Krylov condition-number estimators
• Add log-log benchmark trend fitting
• Add scl-benchmark CLI entry point
• Document benchmark extrapolation workflow
• Verify the full test suite after the Krylov benchmark feature
• Replace tiny block benchmark samples with realistic generated sparse families
• Strengthen correctness validation with PDE-like and coupled sparse matrices

Notes

• Tests should use the checked-in Matrix Market fixture rather than only synthetic arrays.
• Keep the initial implementation small and deterministic.
• The compare command now emits flat CSV/JSON records with per-solver benchmark metrics.
• condest_2 now mirrors condest_1 as a public API and compare output field.
• condest_2 now uses a single exact dense path instead of switching algorithms behind the scenes.
• The Krylov benchmark flow measures smaller FEM samples and extrapolates runtime to 1e6 DOF instead of running that size directly.
• The benchmark flow now supports generated PDE-like matrix families (anisotropic-poisson-2d, coupled-diffusion-2d) in addition to explicit input files.
• Correctness validation now mixes analytical Poisson references with exact dense-reference checks on moderate structured sparse systems.

Errors Encountered

• test_condest.py used an incorrect boolean assertion against the numeric result; corrected to check finiteness separately.