geo: acquire the 3DEP 1 m DEM window in EPSG:26911

550x350 m clipped to the block, already in the project CRS, 984 KB, committed
through LFS. Verified against the service's own point query and the earlier
probe: 115.911/119.422/117.593 m, agreement within 6 cm.

Adds Geo/scripts/dem_stats.py, a stdlib tiled-TIFF reader, because GDAL cannot
run on the host and the exported raster is tiled rather than strip-organised.

Geo/manifests/2026-11-06-3dep-dem.md

@@ -0,0 +1,43 @@
+# Elevation source — USGS 3DEP 1 m DEM, block window
+
+| Field | Value |
+| --- | --- |
+| Source | USGS 3DEP Elevation ImageServer, `elevation.nationalmap.gov`, `/arcgis/rest/services/3DEPElevation/ImageServer/exportImage` |
+| Product | 3DEP seamless 1 m (CONUS), raster id 129943 |
+| Fetched | 2026-11-06 22:59 CET |
+| Request | `f=image&bbox=376100,3774050,376650,3774400&bboxSR=26911&imageSR=26911&size=550,350&format=tiff&pixelType=F32&noData=-9999&interpolation=RSP_BilinearInterpolation` |
+| Bounds (EPSG:26911) | E 376 100 – 376 650, N 3 774 050 – 3 774 400 |
+| Bounds (WGS84, approx.) | lat 34.1003 – 34.1032, lon −118.3426 – −118.3370 |
+| Grid | 550 × 350 px, 1 m pixel, float32, tiled TIFF, nodata −9999 |
+| File | `Geo/raw/dem-3dep-1m-26911-2026-11-06.tif`, 984 338 bytes |
+| SHA-256 | `e571c14c7230d9048697db2893095e1700468e45d4fccf258ceb4d4e9291bdc5` |
+| Horizontal CRS | EPSG:26911 requested and returned (GeoKey 3072 = 26911, model type projected) |
+| Vertical datum | NAVD88 orthometric — 3DEP CONUS convention; the service does not restate it in the file, so this is a documented assumption |
+| Licence | US Government work, public domain |
+
+The window is the block plus a ~100 m margin on the east and ~70 m on the other three sides, which is enough to carry the grade through the junctions at both ends without dragging in the hills north of the boulevard.
+
+## Sampling cross-check
+
+The raster is reprojected by the service, not by us, so it was checked against the service's own point query (`getSamples`) and against the independent point service used in `../verification/2026-11-06-elevation.md`:
+
+| Point | Raster pixel | Service `getSamples` | Earlier point probe |
+| --- | ---: | ---: | ---: |
+| Hollywood × Highland (origin) | 115.911 | 115.910 | 115.911 |
+| Hollywood × Orange Dr | 119.422 | 119.418 | 119.418 |
+| Mid-block crossing | 117.593 | — | 117.53 |
+
+Agreement is within 4 cm on the raster side and 6 cm against the point probe, so the file's georeferencing and signage are consistent.
+
+## What the window contains
+
+Min 110.46 m, max 127.73 m, mean 117.43 m over the 550 × 350 m window — a 17.28 m spread. That spread is **not** the street relief: it includes building tops and the rising ground north of the boulevard. The walkable relief measured along the street is 3.94 m (`../verification/2026-11-06-elevation.md`), and that is the number the demo's ground mesh must honour.
+
+## Reading the file
+
+`Geo/scripts/dem_stats.py` parses it with the standard library. GDAL cannot run on the host (its flatpak libraries do not resolve outside the sandbox), and the exported TIFF is **tiled**, not strip-organised — a strip reader silently scrambles the rows. If another tool misreads it, suspect the tile layout before the data.
+
+## Open
+
+- Reprojecting is not needed: the service already returned EPSG:26911. Anything that samples this file must assume NAVD88 and treat −9999 as nodata.
+- The window stops 30 m north of the Chinese Theatre's roof line; if the background massing needs ground beyond it, extend the bbox rather than extrapolating.