# H01W 26 March plot-intent reconstruction

## Answer

Juan's track plot reproduces cleanly. Igor's PSD curves reproduce almost
exactly from the project waveforms. His spectrograms show the same broad
time-frequency structure, but not the same fine pixels. That is the result one
would expect when the source images do not include the estimator settings or
the calibration response used to make them.

This gives us useful confidence that both people plotted the intended event
and nodes. It does not independently prove Igor's absolute calibration, and it
does not identify the visible pattern with the selected vessel.

## Juan reconstruction

The exact workbook and its readable CSV export already agree in all 319,590
data cells. This package independently reads all 21,306 one-second rows and
reproduces Juan's reported minima:

| Node | Juan minimum | Juan range | Project WGS84 minimum | Project range |
|---|---:|---:|---:|---:|
| H01W1 | 10:27:47 | 4.372 km | 10:27:50 | 4.376 km |
| H01W2 | 10:28:56 | 6.256 km | 10:29:00 | 6.262 km |
| H01W3 | 10:32:27 | 5.193 km | 10:32:31 | 5.198 km |

The pinned project geometry shifts the minima by only 3 to 4 seconds and 4.3
to 6.1 m. Across the plotted 10:15-10:43 interval, the median absolute range
difference is 10.0 to 10.9 m and the maximum is 14.5 to 17.3 m. The difference
grows to about 142-148 m at the remote ends of Juan's six-hour reconstruction.
That wider discrepancy is recorded, but it does not affect the event-hour
comparison.

The recreated faceted plot keeps Juan's grey-panel layout, node colors,
minimum-time annotations and distance scale. It overlays the project geometry
so the small difference is inspectable.

## Igor reconstruction

The project recreation uses the exact three waveform hours, the literal
250.000015 Hz sample rate, periodic-Hann 4096-sample frames, a 1024-sample hop,
linear detrending, one-sided power spectral density, WID2 `CALIB` at 10 Hz,
and the project's pooled nominal response-magnitude curve. Each node contributes
900,000 samples, 875 complete frames and no gap or clipping sentinel.

### PSD

Igor's blue curves were digitized from the supplied JPEG only to compare their
shape with the independent numerical result. The agreement is very strong:

| Node | Shape correlation | Median Igor-project offset | RMSE after constant offset |
|---|---:|---:|---:|
| H01W1 | 0.9958 | +0.26 dB | 0.44 dB |
| H01W2 | 0.9960 | +0.31 dB | 0.45 dB |
| H01W3 | 0.9957 | +0.36 dB | 0.43 dB |

This strongly supports that Igor used the intended H01W triplet hour and a
very similar spectral calculation. It does not establish the provenance or
uncertainty of his calibration response, because he supplied neither the
response nor a numerical array.

### Spectrogram

The reconstructed plots recover the same broad arches, low-frequency lines,
node ordering and hour-scale rise and fall. Pixel-level agreement is weaker:

| Node | Rendered-field correlation | 35-85 Hz time-profile correlation | Fine morphology correlation |
|---|---:|---:|---:|
| H01W1 | 0.686 | 0.816 | 0.313 |
| H01W2 | 0.658 | 0.783 | 0.267 |
| H01W3 | 0.626 | 0.775 | 0.233 |

The RMSE after one constant level offset is 5.66-5.94 dB. Different frame
lengths, overlap, color rendering and JPEG compression can all change this
fine structure. The result validates the broad plotted intent, not Igor's
unreported processing recipe.

Igor's visual minima occur 17 seconds before, 54 seconds after and 27 seconds
before Juan's three closest approaches. The node order matches. This is a
post-hoc timing correspondence for an exposed event, not an acoustic bearing
or vessel attribution.

## Project TL and waveform comparison

Using the existing direct RAM fields, the triplet modeled received-level curve
correlates with the provisional measured triplet series as follows:

| Frequency | Fixed bottom | Source-aware bottom |
|---|---:|---:|
| 63 Hz | 0.530 | 0.588 |
| 80 Hz | 0.363 | 0.410 |

The source-aware branch is modestly closer at both frequencies. Neither model
matches closely enough to identify the selected vessel. The calculation omits
other ships and a travel-time shift, and the measured series is provisional
nominal pressure rather than traceable calibrated SPL.

## Figures

- `results/juan-distance-cpa-reconstruction.png` recreates Juan's distance and
  closest-approach figure with the project geometry overlaid.
- `results/igor-triplet-psd-reconstruction.png` recreates Igor's PSD layout.
- `results/igor-triplet-spectrogram-reconstruction.png` recreates Igor's
  spectrogram layout.
- `results/igor-source-versus-project-reconstruction-local-only.png` places the
  supplied and recreated views side by side. It is local-only.
- `results/juan-track-tl-waveform-comparison.png` shows the existing fixed and
  source-aware RAM curves against the waveform series at 63 and 80 Hz.

## Claim limit

This is a post-hoc reconstruction of plot intent. It supports the integrity of
the retained track, waveform selection and broad spectral morphology. It does
not establish calibrated ambient SPL, direction, detection, contribution,
vessel identity, attribution, causality or prediction. The earlier simple
Lloyd-mirror result remains indeterminate and unfavorable.