# Bounded 63 Hz sediment propagation validation

## Outcome

The frozen 18-path H01/H10 comparison and four-path factor-isolation subset are
complete. All 42 new RAM solves passed their bound input, resource, structural,
serialization, and cleanup checks. The two MAERSK distribution branches were
reused from the validated numerical-depth gate. Accepted proxy-bottom products
remain byte-unchanged; all new fields are additive research artifacts.

This sample does **not** support a universal sediment correction. The mapped or
explicitly completed bottom changes exact 6 m endpoint TL in a strongly
path-dependent way, and H01 and H10 move in opposite directions.

## Primary mapped-minus-proxy result

The primary quantity is exact-range, carrier-aware, 6 m source-depth
`TL_mapped - TL_proxy` at 63 Hz.

| Stratum | n | Minimum | Median | Maximum | Mean | MAD |
|---|---:|---:|---:|---:|---:|---:|
| All frozen paths | 18 | -41.15 dB | -3.04 dB | +25.06 dB | -0.08 dB | 2.17 dB |
| H01 | 14 | -41.15 dB | -3.63 dB | -0.33 dB | -6.75 dB | 1.48 dB |
| H10S | 4 | +20.83 dB | +23.55 dB | +25.06 dB | +23.25 dB | 0.95 dB |

The H01 result is heterogeneous. The -41.15 dB endpoint is a localized exact
point/depth effect: that path's median change across the saved common-wet field
is -1.10 dB. Across all 18 paths, saved-field median changes range from -1.73
to +2.51 dB, while the pathwise 95th percentile absolute field change ranges
from 2.61 to 17.24 dB. Common-wet coverage is 1.0 for every comparison and
whole-field complex-pressure coherence ranges from 0.966 to 0.999. Thus the
large endpoint tails are real within the coherent RAM fields but are not stable
uniform offsets over range and depth.

The three repeated CRETANSEA H10 snapshots are internally consistent at
+20.83, +23.15, and +25.06 dB. The three H01 hydrophone-node paths for the
frozen `1401a9...` event are -0.33, -1.73, and -3.70 dB. These are useful
within-sample consistency checks, not population estimates.

## Factor isolation

| Path | Fully mapped/completed | c/rho branch | attenuation branch | non-additivity |
|---|---:|---:|---:|---:|
| q105 H01W1 | -8.94 dB | +5.99 dB | -6.24 dB | -8.69 dB |
| H01 deep `a1fc63...` | -3.56 dB | +0.81 dB | -2.83 dB | -1.55 dB |
| H01 shoaling `306829...` | -12.86 dB | -3.39 dB | -10.90 dB | +1.44 dB |
| H10 MAERSK | +23.95 dB | +24.15 dB | -0.96 dB | +0.76 dB |

The factors are not generally additive. In particular, q105 has a large
interaction term, so its earlier single-path result cannot be decomposed by
subtracting independent c/rho and attenuation shifts. The H10 MAERSK change is
mostly associated with the c/rho branch under the explicit operational
completion used here.

## H10 interpretation boundary

The H10 source retains the mapped broad Diesing category
`calcareous_sediment`, but the available package chain has no native c, density,
or attenuation values for that broad class. All three acoustic fields are
therefore supplied by the explicitly labelled mud completion. The +20.8 to
+25.1 dB H10 shifts characterize that operational fallback experiment; they
are **not** estimates of native calcareous-sediment acoustics and cannot justify
an H10 production-bottom change.

## Numerical and operational validation

The preceding 12-case numerical-depth gate selected the original baseline
depth for both q105 and MAERSK, proxy and mapped branches. Successive deeper
settings changed mapped endpoint TL by 0.00156 then 0.0000049 dB for H01, and
0.34518 then 0.00425 dB for H10, with full wet-field coverage. The large
sediment effects above are therefore not explained by the tested computational
bottom-depth truncation.

The 42 new runs produced 71,409,011 bytes of JLD2 results (about 70 MiB for the
complete result lineage). Environment-plus-solve time was 15.85/27.22/69.44 s
minimum/median/maximum. Peak worker RSS was 1.65/1.91/2.16 GiB. The minimum
post-serialization available memory was 61.81 GiB; maximum per-case live
swap-out was 1,209 pages and maximum swap growth was 0.00409 GiB, both within
the frozen 35 GiB resource policy. Earlier pre-solve resource holds are retained
and produced no scientific field.

## Claim boundary and next decision

These results establish that bottom specification can materially affect 63 Hz
exact-path TL and that the effect is path-, station-, and property-dependent.
They do not estimate a transferable correction, validate the H10 fallback as
physical calcareous acoustics, or authorize 80 Hz, atlas regeneration, or a
broad production rerun. A production decision needs defensible native H10
calcareous properties and an out-of-sample validation set distinct from these
18 paths.

## Canonical artifacts

- Current-generation validation: `current-generation-validation.json`,
  SHA-256 `492a4ae75b47b9292aa63e3a8282ab716e6ee3a00427b30eee12b4f6dd92eda7`.
- Descriptive statistics: `sediment-multipath-statistics.json`, SHA-256
  `d413553663a626caf476842cf2c0c779a643cbe55c8ca00f9407ebe60d065f58`.
- Execution release: SHA-256
  `5610c226b8c9e064cac035cb525fb57b92f42d2efa29e18607143a37579179c8`.
- Authorization: SHA-256
  `df7b5c0cff385397d71aef63090287b9861723ca2b2c53d9aa63162110992288`.
- Statistical amendment: SHA-256
  `2ed2217b3bb78b43b783592a2c856cf1ac936f2862bb4a8bae46c2e982dafa17`.