Skip to content

QF Solver 0.2.5a0 Owner review: 025-G04

Decision identity

Field Value
Gate 025-G04
Owner decision REQUIRED
Qualified source SHA recorded in results/vnv_0_2_5/g04_latest/evidence_manifest.json
Qualified source worktree CLEAN at evidence start
Evidence pack results/vnv_0_2_5/g04_latest/
Evidence manifest results/vnv_0_2_5/g04_latest/evidence_manifest.json
Contract lowered NO

This review records the available evidence and the reasons the gate cannot be closed. It is not a release decision and does not authorize a tag, push or package publication.

Contract audit

The G04 contract requires all of the following in the same bounded scope:

  • a true FEM branch crossing a limit point and continuing post-limit;
  • complete load-factor/displacement and reaction histories;
  • a coarse/medium/fine/refined branch study, or an explicit Owner-approved bounded refinement decision;
  • restart before or near the turn, plus controlled rollback and retry;
  • analytical or published branch reference;
  • the mandatory Code_Aster external correlation of the complete branch;
  • clean-source provenance and digest-controlled evidence.

The contract does not accept a reduced shallow-arch model as a FEM proof, does not accept Newton convergence alone, and does not allow a monotone external path to stand in for a missing post-limit correlation.

Evidence assessment

QF Solver common-driver FEM

The common-driver run is a genuine sparse two-element TET4 path:

Quantity Observed value
Nodes / global DOF 5 / 15
Continuation steps 80
Arc-length radius 0.02
Control DOF 14
Load-factor range [-0.0373265418, -0.0007272447]
Control-displacement range [-0.9419440185, -0.0115649488]
Turning points 1, around step 75
Maximum relative residual 2.8114e-11
Minimum det(F) 0.4414340033
Evidence status PASS_INTERNAL_RESEARCH

The signed load factor reverses between steps 76 and 77 while the control displacement continues. The targeted tangent diagnostic finds the smallest reduced-tangent eigenvalue crossing from positive to negative at the turn, with a reassembled free residual below 5e-13 and positive det(F). This demonstrates the intended internal branch diagnostic, but it is still a minimal research path and not a qualified FEM release scope.

Restart and rollback

The controlled internal helpers pass restart checks before and after the observed turn. They reproduce the continuation suffix, displacement endpoint and material state. The adversarial run injects a failure after two Newton corrections at step 76 and records a radius reduction from 0.02 to 0.01 with a clean retry.

These results are PASS_INTERNAL_RESEARCH. They do not provide the missing external correlation or mesh study.

Reduced analytical reference

The reduced shallow-arch equation remains useful for algorithmic verification and crosses its analytical limit point. It is not a finite-element reference for closing G04.

Code_Aster external branch diagnostic

The historical Code_Aster result was an executed configuration mismatch, not a solver disagreement. QF applies a positive reference force with a negative load factor, hence the physical loading is downward; the old Code_Aster deck applied FZ=+1/3 and followed the upward branch. It also measured the mean CROWN/DZ, while QF controls apex node N5 UZ (global DOF 14), and used an unmatched continuation window.

The corrected pinned Docker replay uses the same unperturbed two-element TET4 geometry, FZ=-1/3, APEX/DZ, and a 0..0.96 window sampled at 160 intervals. It produces a complete branch and one turning point. QF-versus-Code_Aster is compared by apex displacement, because the solvers do not use the same arc-length parameterization:

Quantity Observed value
Image simvia/code_aster@sha256:4629a21a109309bb97fbdc27d750445cc869e151e2e2ed6290f69539614e4435
External path complete
External turning points 1
Maximum branch load-factor difference 4.8719e-07
Peak-normalized maximum difference 1.3052e-05
Turning-load difference 6.3485e-08
External diagnostic status RESOLVED_CONFIGURATION_MATCH

This is bounded code-to-code diagnostic evidence. It removes the historical external-deviation blocker but is neither physical validation nor sufficient to close G04 on its own. The detailed parameter audit is recorded in 0_2_5_g04_external_branch_diagnostic.md.

Gate decision

025-G04 = OPEN
OWNER DECISION = REQUIRED
MESH DECISION = OPEN_MISSING_REQUIRED_LEVELS
CODE_ASTER = RESOLVED_CONFIGURATION_MATCH
CONTRACT LOWERED = NO

Exact remaining blockers

  1. Link and reproduce a published FEM snap-through branch reference. The current two-element configuration is custom and no exact public reference has been identified.
  2. Execute and archive the required coarse/medium/fine/refined arc-length branch study, or obtain an explicit bounded-refinement Owner decision.
  3. Regenerate a final controlled G04 pack only after the preceding evidence is available.

Until these blockers are addressed, no qualified arc-length production claim is made. G03, G05 and G06 remain unchanged.