Status and claims

claims/claims.yaml is the machine-audited source of truth. The manuscript may explain a claim, but it cannot promote one by rhetoric.

Claim	Status	Promotion gate
Outgoing complex-pole operator	validated	Retained residual, passivity, reciprocity and branch regressions pass.
Differentiable geometry	validated	Production-discretization directional and coordinate checks pass.
Quan–Lončar FDTD gain	provisional	Exact frozen endpoints pass the publication convergence panel, with final Q above initial in every matched variant.
Design23 FDTD gain	provisional	Exact step 0 and 105 pass identical mesh/runtime/boundary/mode gates, with final Q above initial in every matched variant.
Two-device generality	provisional	Both device-gain claims become validated.
Direction transfer without magnitude transfer	provisional	Both device-gain claims validated under the panel, and the rank correlation recomputed on the panel-qualified endpoints.

Evidence boundary

The method claims certify numerical self-consistency of the retained analytical operator and derivative. They do not certify absolute agreement between analytical and FDTD Q. The two observables remain distinct in prose, plots, and machine-readable results.

The headline transfer claim is deliberately bounded. It rests on the rank correlation over a trajectory under one fixed FDTD protocol (Spearman 0.911, n = 23) together with the two matched endpoint comparisons. It does not rest on per-step agreement: consecutive diagnostics move in the same direction 19 times out of 31, a two-sided binomial p of 0.28, which is not distinguishable from chance and is not claimed.

Ringdown adequacy is now audited for every retained FDTD observation. Three of the four quoted endpoints were fitted over less than one amplitude lifetime, the worst being the optimized Quan endpoint at 0.24 lifetimes. This is the precise reason both gains remain provisional, and it forces an amendment to the predeclared panel: the run-time axis must be extended until at least three lifetimes are observed at the optimized endpoint.

The source migration records that cavitygrad had uncommitted changes at the snapshot commit. Relevant implementation files are preserved in this repo, and the full original status is recorded in provenance/cavitygrad-source.json so that this fact is not hidden.