Measured waveform method-stability comparison · sealed current evidence

Broad patterns recur. Fine spectral detail does not.

Across six close H01W passages, periodic-Hann and five-taper DPSS spectrograms have a median broad-structure correlation of 0.784, but only 0.445 after removing the local 33-bin frequency median. The fine-detail median absolute disagreement is 3.0 dB and its 95th percentile is 10.6 dB. This is estimator sensitivity—not evidence for or against a ship or propagation mechanism.

Broad correlation

0.784

median; 0.731–0.910

Broad RMSE

5.3 dB

offset removed; 2.9–5.4

Fine correlation

0.445

median; 0.389–0.664

Fine p95 difference

10.6 dB

5.6–10.7

PassageEstimator

Display-only decimation of the fixed-triplet linear-power combined field. All reported metrics use every jointly finite cell, separately per passage and subject; pixels are never pooled across passages.

Passage-by-passage metrics

Passage / subject	Broad r	Broad RMSE	Fine r	Fine median \|Δ\|	Fine p95 \|Δ\|

Interpretation boundary

The six passages were frozen prospectively from geometry and calibrated support. Agreement does not validate a Lloyd-mirror explanation, identify a ship, estimate source contribution, or change the unfavorable simple Lloyd-mirror result.