0.2.7 Large-Scale and 1M-DOF Plan

The active Level-Up record is WP16 / LUP-027-G16, now PASS on the official PETSc retry. The historical WP12 readiness evidence below remains unchanged and is not a 1M qualification. WP16 required a complete iterative solve on at least 1,000,000 real FEM DOF, including loads, boundary conditions, reactions, residual, equilibrium, energy, matrix-free/assembled subscale comparison, two replays and complete resource/provenance capture. A timeout or resource-limited attempt is not PASS.

This plan characterizes resource boundaries. It does not promise that every route can solve one million degrees of freedom.

Size ladder

Level	Purpose	Required interpretation
100k DOF	repeatable baseline	establish timing, memory and residual method
300k DOF	large full-solve checkpoint where feasible	report full solve or resource limitation
500k DOF	ambitious intermediate target	characterize bottleneck and resource context
1M DOF	completed bounded qualification	full solve, iterative/HPC result or characterized resource limit

Measurements

Record model creation/import, assembly, factorization or preconditioner setup, solve/iterations, post-processing, peak RSS and Python allocation when available, total wall time, nnz, residual, hardware/software and topology. Separate warm-up, I/O and compute timings.

Allowed verdicts

• PASS_FULL_SOLVE: the declared route completes with residual and resource evidence;
• PASS_ITERATIVE_OR_HPC: an explicitly supported iterative or HPC route completes with its own limitations;
• RESOURCE_LIMITED_CHARACTERIZED: the probe cannot complete, but the reason, resource boundary and partial measurements are reproducible.

The last verdict is not an automatic release blocker when no public full-solve claim is made. It is also not a PASS for the missing capability.

WP12 evidence checkpoint

The bounded campaign is recorded in qualification/0_2_7/wp12_scaling_evidence.json and the separate 300k assembly probe in qualification/0_2_7/wp12_assembly_probe_300k.json. Matrix-free CG completed the declared structured TET4 route through 750141 DOF. The 300k probe reached 311469 DOF for assembly only. The 1M attempt was RESOURCE_LIMITED_TIME; SciPy retained its explicit 200000-DOF guard and the direct route was memory-limited at 107811 DOF. PETSc/SLEPc were unavailable in the measured environment. WP12 status is PASS_WITH_LIMITATIONS with PROPOSED_OWNER_REVIEW pending. The later WP16 retry used the pinned PETSc route and completed two reproducible 1,029,000-DOF replays under the frozen WP14 acceptance contract; this is a bounded TET4 linear-static result, not a universal 1M or multi-million-DOF claim.