Strict qualification-route successor

The frozen research and native checkpoints contain real trajectories with correct refinement behavior. Independent review found that their qualification readers also accepted malformed substitutes. This additive successor preserves every old packet, failed oracle and original numerical/physical/Lean limitation. It repairs the claim-verification route, not the simulation equations.

The research reader now requires the declared .1 endpoint and distinct ordered .05,.025,.0125,.00625,.003125 schedule, 2/4/8/16/32 actual steps, complete physical fields, unclamped actual clock/version and the same exact real endpoint in every case. It regenerates both DOP853 references from the declared initial field, canonical settings and common endpoint, and checks the supplied reference fields. It independently recomputes final velocity/geometry, lumped errors, reference separation and velocity/geometry ratios from the related replays. Original physical work/GCL/momentum and convergence bands remain unchanged. Report comparison uses a separate floating representation allowance, never an expanded physical acceptance tolerance. Every returned convergence quantity is computed, not copied from the claimed summary.

The native reader requires actual accepted mass, raw PS positions, topology, physical pressure and coefficient fields on every row. Missing mass cannot turn these checks off. It requires the actual owner ID/version, canonical distinct case schedule, monotone per-case version and actual clock, and obtains reference and host endpoints only through the repaired full research reader. The new additive native export example obtains ID/version and its step version directly from state.stamp; the frozen loop-counter export remains preserved. Numerical fields must match the old real native capture exactly after removing the newly added stamp field. Simulation Rust and existing tests/examples/Cargo stay unchanged.

Actual frozen-oracle reproductions are retained: ten cells in five duplicate coarse cases with duration999, empty references and fabricated error/ratio claims; 124 native cells with publication geometry/mass/topology/physical pressure omitted. Focused regressions and full normal/optimized strict replays reject those gaps while accepting the real stored trajectory corpus and a fresh actual-state export. No fresh rollback, continuum or governing-equation claim is inferred from these reader checks. The governing-equation/pressure-interpretation audit remains a separate source/evidence lane; no new material variable is introduced here.