A sampled witness.
Explicit timing and contamination costs.

Sample timing enters the Gram bound.

The proposed independent monitor has 1024 complex samples on a nominal one-second cyclic grid and finite modes 100–199 Hz. The first-to-last nominal sample span is 1023/1024 seconds.

Per-mode frequency error ≤5 mHz, affine clock error ≤10 ppb and per-sample residual ≤10 ns give a sampled-Gram eigenvalue lower bound above 0.9096. Epoch and common frequency shifts leave that bound invariant but do not certify absolute phase.

Raw-monitor contamination is charged directly.

The outside-model contribution appears inside the sample-norm bound and in the full tail reference. With the previous 200 µV outside cap, this direct route’s sufficient bound would miss the target. The new 10 µV cap is a prospective stronger requirement, not an observed result.

The independent-monitor error allocation is 10 µV acquisition/quantization, 5 µV quadrature calibration, 3 µV baseline and 2 µV numerical error. Applicable full-path calibration remains necessary; these requirements are separate from the original acquisition budget.

Exact digital arithmetic cannot exclude unknown aliases.

The synthetic fixture is stored as binary64 hex words. Rational mean-squared norm and an integer ceiling square root verify its digital RMS; five corrupted/non-fixture controls return HOLD. No physical source accuracy is inferred.

Modes at 100 and 1124 Hz alias exactly on the nominal grid. Opposite source coefficients can therefore vanish in the sampled record while their spectral variation remains large. Independent source completeness and outside-model evidence are indispensable.

Next bind real monitor timestamps, I/Q calibration, source preparation, outside-model evidence and computation hashes into a reviewable receipt contract.