0.2.6 G07 Step B1 — Arc-Length Targeted Evidence

Evidence ID: 026-G07-B1-ARC-LENGTH-001
Baseline start SHA: eb105657f1d5f1a0994f5598327a290358f37e7e
Execution source SHA: 84087a22dd51fe1445a5dad51fc690bfe8b7c85e
Gate status preserved: 026-G07 = NOT_STARTED
Functional source changed: NO

The complete machine-readable record is g07_b1_arc_length_evidence.json. The runner is run_g07_b1_arc_length.py. This step is targeted evidence only; it does not modify TL, change a formulation, or close G07.

Predeclared ARC-002 design

The same TET4 snap-through benchmark and fixed solver regime were evaluated with two arc-length radii (0.01 and 0.02) on two conforming meshes:

Mesh	Definition	R_SMALL	R_NOMINAL
M_COARSE	reference two-TET4 bipyramid	160 steps	80 steps
M_REFINED	one-to-eight conforming refinement, 16 TET4	160 steps	80 steps

Before execution, case validity required completion of the declared step window, a detected branch turn, finite load/displacement/residual/determinant fields, non-duplicate path points, and positive det(F). The bounded sensitivity report uses relative turning-point changes when both turning points exist; no universal numerical stability threshold is introduced.

Result

The two coarse-mesh settings completed finite continuous paths with one turning point each. Their relative turning-point changes were:

• load factor: 3.6021e-12;
• control displacement: 1.7822e-08.

Both refined-mesh settings also completed with solver status PASS, finite fields, positive minimum det(F) (0.8886481) and continuous paths, but no turning point was observed in either predeclared window. They are therefore classified DEFER, not PASS or solver failure. The mesh turning-point sensitivity comparison is not available for this refined series.

ARC-002 = DEFER: the run provides discriminating bounded evidence, but the turning-point sensitivity gap remains open. A future case must use an equivalent refined boundary/load discretization or an Owner-approved compatible continuation window before this gap can be closed.

ARC-003 restart and rollback

Three route-native cases were executed on the nominal coarse benchmark:

Case	Result	Evidence
restart before turn	PASS_BOUNDED	checkpoint step 75, suffix error 0, final state digest equal
restart after turn	PASS_BOUNDED	checkpoint step 76, suffix error 0, final state digest equal
controlled rollback near turn	PASS_BOUNDED	explicit MAX_ITERATIONS, rollback before retry, clean retry

The before-turn restart was replayed. Classification, checkpoint step, state digest and final digest were identical. No ghost state was detected.

Step B1 conclusion

Runtime assertions passed for all eight recorded runtime rows (four ARC-002 cases, three ARC-003 cases and one restart replay): deterministic replay, finite runtime fields, no silent PASS and state integrity. External calculation was not run by policy. No numerical regression was observed and no functional source changed.

G07_B1 = PARTIAL
ARC-003 = PASS_BOUNDED
ARC-002 = DEFER
ARC_LENGTH_OWNER_CANDIDATE = PASS_WITH_LIMITATIONS / ARC-002_DEFERRED

The remaining blocking gap is the refined-mesh turning-point comparability in ARC-002. G07 remains NOT_STARTED; Owner closeout is not yet ready.