# H01W dense-track direct TL comparison

## Result

The 26 March 2026 comparison is complete for the 60 UTC minute centres from
10:10:30 through 11:09:30, all three H01W nodes, 63 and 80 Hz, and both bottom
treatments. That is 720 current-generation RAM cases. The 696 already valid
cases were preserved. The only missing cases were 24 short H01W1
source-aware paths; all 24 passed after the registered RAMTools 0.1.49
short-path fix and were merged by the combined validator without rerunning
long paths.

Across the 360 paired node/frequency/minute comparisons, source-aware minus
fixed-bottom TL has median **+1.88 dB**, 5th–95th percentile **−2.22 to
+5.86 dB**, and full range **−14.91 to +12.59 dB**. Positive values mean
greater propagation loss and therefore lower modeled received level for the
same source level. The node/frequency medians are:

| Frequency | H01W1 | H01W2 | H01W3 |
|---|---:|---:|---:|
| 63 Hz | +1.72 dB | +1.79 dB | +1.99 dB |
| 80 Hz | +1.99 dB | +1.91 dB | +1.56 dB |

The updated bottom modestly improves the Pearson morphology correlation with
the provisional one-minute measured series in every reported comparison. For
the triplet linear-power mean, correlation changes from 0.530 to 0.588 at
63 Hz and from 0.363 to 0.410 at 80 Hz. These are descriptive comparisons,
not calibrated SPL validation or ship attribution.

## Scientific boundary

- Source longitude/latitude and the pinned hydrophone-node coordinates define
  the WGS84 RAM geometry identically for both bottom branches. Juan's supplied
  distance columns remain provenance only: median signed difference −10.88 m,
  median absolute difference 15.13 m, maximum absolute difference 36.70 m.
- Source depth is 6.16 m. The water field, bathymetry, grid, extraction,
  frequency, source level, and time sampling are held fixed between branches;
  the intended difference is the bottom treatment.
- Measured curves are provisional nominal-band levels, not traceably
  calibrated SPL. The modeled series uses one fixed JOMOPANS-ECHO source level
  per frequency, omits other ships, and applies no acoustic travel-time shift.
- The result does not reopen the already closed/indeterminate Lloyd-mirror
  conclusion.

## Package and execution lineage

The preserved long-route results use OceanAcoustics 0.1.25, OceanData 0.1.27,
RAMTools 0.1.48, and RAM 0.1.21 through the public
`buildenv_source_aware` route. RAMTools 0.1.48 had a confirmed sampling defect
for nonzero paths short enough to collapse to one sample. The additive
recovery uses registered RAMTools 0.1.49, release commit
`8cbbe9da7a7ae95b042d3d4202f6f4927508353f`, tree
`82440da29efc3ae61448a478ef5ca066bd785816`, and LocalRegistry commit
`7f04a30f1907dcb4c664ed81678fb3aaa49612da`. Every recovered receipt contains
both path endpoints and zero endpoint error. No manual bottom tuple or
attenuation override was used.

The combined validation passed 18/18 checks. Peak case RSS was 1.84 GiB,
median solve time 1.79 s, maximum solve time 10.16 s, and minimum observed
MemAvailable was 59.47 GiB. The largest per-case swap-out delta was 259 pages
and the largest swap-use increase was 0.00145 GiB.

## Scope B: monthly/fortnightly ship-type derivative

The January 2025 4 km × 4 km grid and AIS eligibility ledger are frozen. Of
1,469 rows inside the 1,976-cell grid, 1,411 are eligible and 58 are excluded
(35 missing/nonpositive speed, 19 unmapped type, 4 missing/nonpositive
length; counts are overlapping by design). The intended derivative assigns
each eligible row to its nearest grid cell, computes corrected JOMOPANS-ECHO
source level, subtracts the labelled RAM TL layer, and sums unrelated rows in
linear power by fortnight and ship class.

The fixed-bottom January 2025 63 Hz base is complete. The matched January 2025
source-aware 63 Hz base is a separate continuation stage and is not yet a
completed local derivative input. No matched January 2025 80 Hz fixed or
source-aware base exists. Therefore no honest two-bottom monthly/fortnightly
comparison is published yet; the eligibility/grid work is complete and the
missing TL layers are stated rather than substituted with February 2026 or an
approximate legacy field. Juan's five revised monthly PDFs are retained as
external comparison/provenance only.

## Main artifacts

- `validated/validation-report.json` — combined numerical validation.
- `validated/direct-node-series.csv` — 720 node-level rows.
- `validated/direct-paired-differences.csv` — 360 fixed/source-aware pairs.
- `validated/direct-triplet-series.csv` — 240 triplet series rows.
- `validated/h01w-dense-track-tl-rl-morphology-63hz.{png,svg}`.
- `validated/h01w-dense-track-tl-rl-morphology-80hz.{png,svg}`.
- `validated/figure-visual-validation.json` — full-render visual QA.
- `short-path-recovery-v1/package-resolution.json` — registered 0.1.49 receipt.