# v11 expected-RL63 high-versus-low acoustic association

## Decision

The frozen comparison is mixed and does not validate the common 63-Hz expected-RL proxy as a general acoustic predictor. Four of six pairs had positive high-minus-low 63-Hz contrasts on every required node, including three of five complete-triplet pairs and the one H10S two-node pair. H08S was negative on all three nodes; H10N was mixed and essentially zero at the triplet median. The measured pair medians ranged from -0.82 to +8.41 dB, substantially smaller than every frozen proxy gap (12.60 to 36.09 dB).

Across only six non-independent pairs, the descriptive Spearman rank association between proxy gap and measured median contrast was +0.371. No p-value or independence claim is made. The result supports retaining the common proxy only as a transparent candidate-prioritization covariate for future, context-matched prospective comparisons. It is not a standalone acoustic decision score and supplies no detection or attribution evidence.

## Frozen results

| Pair | Coverage and matching | Proxy gap (dB) | Median 63-Hz contrast (dB) | Positive nodes | Positive-median bands / all-node-positive bands |
|---|---|---:|---:|---:|---:|
| V11-P01 | H01W triplet; same anonymous file-scoped AMSA CRAFT_ID/date | 36.09 | +1.91 | 3/3 | 10/11 / 10/11 |
| V11-P02 | H01W triplet; same anonymous file-scoped AMSA CRAFT_ID/date | 23.20 | +5.38 | 3/3 | 10/11 / 8/11 |
| V11-P03 | H01W triplet; same anonymous file-scoped AMSA CRAFT_ID/date | 20.32 | +7.67 | 3/3 | 8/11 / 5/11 |
| V11-P04 | H08S triplet; different MMSIs, same date/class and length/speed matched | 14.90 | -0.82 | 0/3 | 2/11 / 0/11 |
| V11-P05 | H10N triplet; same retained MMSI/date | 12.60 | -0.08 | 1/3 | 6/11 / 5/11 |
| V11-P06 | H10S two-node; same retained MMSI/date | 17.92 | +8.41 | 2/2 | 7/11 / 5/11 |

All negative and mixed node/band results are retained in the machine-readable tables. H03 was not tested because the completed rank contains no retained H03 waveform overlap; acquisition was outside v11.

## Prospective-style controls and computation

Before acoustic access, v11 pinned the complete 103,851-row common-method rank, 10,923 retained candidates (including 209 sealed rows retained only for audit), 1,681 pair candidates, exact six-pair membership, receive-time bookkeeping, fixed 102.4-second windows, QC, eleven decidecade bands, a 63-Hz primary outcome, summaries, and stopping rules. Protected scenes remained sealed.

H01W and H08S use nominal-pressure products that are response-corrected but not traceably calibrated. H10N and H10S use raw counts because date-valid full complex responses were unavailable. Only within-node high-minus-low contrasts are interpreted; absolute levels and unlike response bases are not pooled. The H01 identity field is an anonymous, file/month-scoped AMSA CRAFT_ID, not MMSI or IMO. ARA snapshots are sparse and incomplete. All controls have unmatched environmental and traffic context, and the H10S high scene was previously acoustically exposed.

The fixed pass read 870,400 selected samples, retained no waveform or PSD arrays, took 2.29 seconds, and peaked at 214,884,352 bytes RSS. There was no network access, acquisition, propagation, directionality, upload, protected access, or canonical repository change.

## Independent reconstruction

Independent replay passed. It reproduced all six selection decisions, confirmed that no same-MMSI H08S pair met the frozen timing gate, reproduced all 34 raw selected-window hashes, and reconstructed all 187 node-band rows with an explicit rFFT Welch implementation. Maximum band-power relative disagreement was 2.16e-15; maximum contrast disagreement was 3.55e-15 dB; no sign count differed. Aggregate hashes for v1 through v10 also matched their pre-outcome pins.

## Claim ceiling and next use

This is openly exploratory event-versus-comparison association only. It does not establish ship detection, contribution, direction, identity, attribution, calibrated prediction, causality, or shipping evidence. Any strengthening should use genuinely held-out cases with preselected, non-overlapping controls matched on environmental state and traffic context; it should not create another acoustic score variant from these outcomes.