Nif00 · Nif00 · Jul 21, 2026
diff --git a/.gitattributes b/.gitattributes
@@ -0,0 +1,3 @@
+data/processed/manhattan_buildings.json filter=lfs diff=lfs merge=lfs -text
+data/processed/manhattan_network.json filter=lfs diff=lfs merge=lfs -text
+data/raw/manhattan_osm_buildings.json filter=lfs diff=lfs merge=lfs -text
diff --git a/PLAN.md b/PLAN.md
@@ -32,7 +32,7 @@ SimulationResult -> ranked hotspots / downloadable JSON / restrained local web U
 - `fire_spread.buildings`: converts preserved direct OSM XML map tiles into projected building footprints and height-confidence records.
 - `fire_spread.propagation`: configurable edge traversal and event-driven spread.
 - `fire_spread.cli`: a small command-line entry point for repeatable runs.
-- `fire_spread.web`: local HTTP API and small static UI for scenario controls, network view, 2.5D building rendering, and hotspot output.
+- `fire_spread.web`: local HTTP API and small static UI for scenario controls, interactive 3D network/building rendering, and hotspot output.
 - `tests`: contract tests for graph validation, directional links, modifiers, horizon handling, deterministic results, OSM conversion, and building provenance.
 - `data/raw`: preserved source city files and provenance records.
 - `data/processed`: normalized network artifacts and derived simulation outputs.
@@ -60,9 +60,18 @@ SimulationResult -> ranked hotspots / downloadable JSON / restrained local web U
 }
 ```
 
-Coordinates are projected/local meters in normalized data. The Polatlı adapter uses a local equirectangular projection centered on the Nominatim town-relation result, with the source WGS84 coordinates retained in node metadata.
+Coordinates are projected/local meters in normalized data. The Manhattan and Polatlı adapters use a local equirectangular projection centered on each study area, with the source WGS84 coordinates retained in node metadata.
 
-## Real starting dataset: Polatlı, Ankara
+## Current starting dataset: Manhattan Island, New York
+
+- Source: OpenStreetMap road and building geometry retrieved from the public Overpass API.
+- Coverage: `40.7000,-74.0200` to `40.8825,-73.9065`, covering Manhattan Island and its immediate shoreline.
+- Normalized graph: 147,121 nodes and 161,601 links from 30,935 highway ways.
+- Building layer: 157,202 OSM building/building-part footprints; 116,917 have explicit height tags, 2,171 have level estimates, and 38,114 are footprint-only.
+- Files: `data/raw/manhattan_osm_roads.json`, `data/raw/manhattan_osm_buildings.json`, `data/raw/manhattan_metadata.json`, `data/raw/manhattan_buildings_metadata.json`, `data/processed/manhattan_network.json`, and `data/processed/manhattan_buildings.json`.
+- License: OpenStreetMap data © OpenStreetMap contributors, ODbL 1.0. Query, coverage, origin, and height policy are preserved in the metadata files and normalized outputs.
+
+## Legacy regression dataset: Polatlı, Ankara
 
 - Source: OpenStreetMap road geometry retrieved from the public Overpass API.
 - Coverage: `39.54,32.08` to `39.63,32.22`, covering the Polatlı urban area and immediate outskirts rather than the full administrative district.
@@ -75,17 +84,17 @@ Coordinates are projected/local meters in normalized data. The Polatlı adapter
 
 ## Current boundary decision
 
-The street graph remains the only spatial input to fire propagation. The OSM building layer is currently presentation-only: it drives the 2.5D viewer but does not alter spread speed, connectivity, or hotspot scores.
+The street graph remains the connectivity input to fire propagation. The web UI now offers an experimental QUIC-URB-inspired urban-wind layer: OSM building footprints are rasterized into a regular obstacle grid, and sampled local wind speed/direction alter edge spread speed. This is a transparent diagnostic heuristic, not a fluid solver. The web result also derives an adjacency layer that marks buildings near an activated road segment or ignition node as burning for 3D visualization.
 
-The viewer exposes orthographic and perspective projections plus a clearly labeled vertical exaggeration control. Unknown building heights remain flat in source-only mode; the optional 3 m massing fallback is a visualization assumption, not a measured height.
+The viewer exposes orthographic and perspective 3D cameras plus a clearly labeled vertical exaggeration control. Unknown building heights remain flat in source-only mode; the optional 3 m massing fallback is a visualization assumption, not a measured height. OSM `height`, `building:levels`, `min_height`, and roof metadata are preserved for rendering; sparse OSM node `ele` values are also carried when present.
 
 ## Simulation phases
 
 1. **Baseline graph:** load and validate a normalized network; reject missing endpoints, duplicate IDs, non-positive lengths, and invalid parameter ranges.
 2. **Deterministic propagation:** simulate one or more ignition nodes over a finite time horizon; preserve earliest arrival per node and link interval.
 3. **Hotspot ranking:** rank links by normalized advancement score and nodes by arrival/exposure metrics.
-4. **City ingestion:** retrieve and preserve the Polatlı OSM extract, reproject to local meters, split ways at geometry vertices, preserve one-way/surface/width tags, and validate topology.
-5. **Current web UI:** run real scenarios against the Polatlı graph, show the full network and activated links, and allow JSON result download.
+4. **City ingestion:** retrieve and preserve a city OSM extract, reproject to local meters, split ways at geometry vertices, preserve one-way/surface/width tags, and validate topology.
+5. **Current web UI:** run real scenarios against the Manhattan graph, show the detailed 3D building layer, activated links, and local flow-vector fields, and allow JSON result download.
 6. **Calibration:** compare modeled travel times and spread patterns against historical incidents or expert-labelled scenarios; keep calibrated coefficients versioned.
 7. **Scenario analysis:** support multiple ignitions, barriers/closures, time-varying wind, fuel/moisture rasters, and repeated stochastic runs only when validated data exists. Building footprints and 3D tags may become simulation inputs only after their effect is specified and validated.
 8. **Presentation:** export GeoJSON and extend the current map view with a timestamp slider and richer city layers after the baseline is validated.
@@ -100,9 +109,9 @@ The viewer exposes orthographic and perspective projections plus a clearly label
 
 ## Definition of done for the current baseline
 
-- The preserved Polatlı OSM road and building extracts load from disk through source-specific adapters.
-- A normalized Polatlı graph and building layer are available with provenance, local-meter coordinates, and height-source labels.
+- The preserved Manhattan OSM road and building extracts load from disk through source-specific adapters.
+- A normalized Manhattan graph and detailed building layer are available with provenance, local-meter coordinates, and height-source labels.
 - A CLI and local web UI run real propagation scenarios against the road graph.
-- The UI offers orthographic and perspective 2.5D views without feeding unvalidated building heights into propagation.
+- The UI offers orthographic and perspective 3D views without feeding unvalidated building heights into propagation.
 - Outputs include arrival times, activated links, ranked advancement scores, and dataset/scenario provenance.
 - Tests cover the observable baseline contracts, OSM conversion, and building conversion boundaries.
diff --git a/README.md b/README.md
@@ -4,21 +4,21 @@ A small, reproducible baseline for simulating an advancing fire front over urban
 
 ## Status
 
-The repository contains a runnable baseline against a preserved real OpenStreetMap extract for Polatlı, Ankara:
+The repository contains a runnable baseline against a preserved real OpenStreetMap extract for Manhattan Island, New York. The earlier Polatlı extract remains available as a smaller regression dataset:
 
 - validated normalized city-network model;
-- 1,837 downloaded OSM highway ways with geometry;
-- 12,266 normalized nodes and 13,920 street/alley links;
-- 12,151 OSM building/building-part footprints for the 2.5D viewer;
+- 30,935 downloaded OSM highway ways with geometry;
+- 147,121 normalized nodes and 161,601 street/alley links;
+- 157,202 OSM building/building-part footprints with height provenance for the 3D viewer;
 - configurable heuristic spread modifiers;
 - deterministic event-driven propagation;
 - JSON input/output CLI;
-- local web UI with orthographic/perspective 2.5D views, hotspot ranking, and result download;
+- local web UI with interactive WebGL 3D views, hotspot ranking, and result download;
 - contract tests and a small hand-authored network used only for unit tests.
 
 The architecture and staged roadmap are in [`PLAN.md`](PLAN.md). The current model is for exploration and ranking only; it is not a validated fire-behavior or life-safety system.
 
-The current milestone is simulation-first: the real street/alley graph drives propagation. OSM building footprints and height tags are currently presentation-only and do not alter spread calculations.
+The current milestone is simulation-first: the real street/alley graph drives propagation. The web UI now includes an explicitly labeled, experimental QUIC-URB-inspired urban-wind layer that rasterizes OSM footprints and heights, then changes local wind speed and direction along graph links. It is a diagnostic approximation, not the QUIC-URB solver.
 
 ## Run the sample
 
@@ -39,29 +39,37 @@ python -m fire_spread --network examples/sample_city.json --ignition A --horizon
 
 Write a result file with `--output path/to/result.json`.
 
-## Run the real Polatlı dataset
+## Run the Manhattan dataset
+
+```text
+python -m fire_spread --network data/processed/manhattan_network.json --ignition osm:40.7911145:-73.9631218 --horizon-minutes 60 --output data/processed/manhattan_result.json
+```
+
+The default web dataset is Manhattan Island. Its study area is `40.7000,-74.0200` to `40.8825,-73.9065`; the normalized origin is the center of that bbox. The current extract contains 147,121 nodes, 161,601 links, and 157,202 building/building-part footprints.
+
+## Run the legacy Polatlı dataset
 
 ```text
 python -m fire_spread --network data/processed/polatli_network.json --ignition osm:39.5852148:32.1436842 --horizon-minutes 60 --output data/processed/polatli_result.json
 ```
 
-The default ignition node used by the web UI is the normalized node nearest the Polatlı town-center origin. The exact graph node ID is reported by the UI and `/api/status`.
+The legacy dataset's default ignition node is the normalized node nearest the Polatlı town-center origin. The exact graph node ID is reported by the UI and `/api/status`.
 
 The raw Overpass road response is preserved at `data/raw/polatli_osm_roads.json`. The building layer is preserved as direct OSM API XML tiles under `data/raw/polatli_osm_map_tiles/`. Queries, bounding boxes, source, license, coordinate origin, and height policy are recorded in `data/raw/polatli_metadata.json` and `data/raw/polatli_buildings_metadata.json`.
 
 ## Run the local web UI
 
 ```text
-python -m fire_spread.web --dataset data/processed/polatli_network.json --buildings data/processed/polatli_buildings.json --host 127.0.0.1 --port 8000
+python -m fire_spread.web --dataset data/processed/manhattan_network.json --buildings data/processed/manhattan_buildings.json --host 127.0.0.1 --port 8000
 ```
 
-On Windows, double-click [`launch_polatli_ui.bat`](launch_polatli_ui.bat). It creates `.venv` on the first run, installs the project, starts the server with the real Polatlı network and building layer, and opens the browser automatically. Press `Ctrl+C` in the launcher window to stop it.
+On Windows, double-click [`launch_manhattan_ui.bat`](launch_manhattan_ui.bat). It creates `.venv` on the first run, installs the project, starts the server with the Manhattan network and detailed building layer, and opens the browser automatically. Press `Ctrl+C` in the launcher window to stop it. The older launcher now also points at Manhattan for compatibility.
 
 The launcher intentionally prefers the `python` command on `PATH`; `py -3` can select an older installed interpreter even when `python --version` reports 3.11.9.
 
-Open `http://127.0.0.1:8000`. The UI reads the real local graph and building layer, allows a node ignition and parameterized run, supports orthographic and perspective projections, offers source-only or explicitly labeled 3 m massing fallback heights, shows activated links, ranks advancement scores, and downloads the exact JSON result returned by the simulation.
+Open `http://127.0.0.1:8000`. The UI reads the real local graph and building layer, extrudes every OSM footprint into a 3D scene, applies tagged heights and `building:levels` estimates, highlights buildings adjacent to the advancing front in bright red, offers source-only or explicitly labeled 3 m massing fallback heights, applies the optional QUIC-URB-inspired urban-wind grid, and can draw a local flow-vector graph around the selected node. It also shows activated links, ranks advancement scores, and downloads the exact JSON result returned by the simulation.
 
-Map navigation: use the mouse wheel to zoom, drag with the left button to pan, and hold `Shift` while dragging (or use the right mouse button) to rotate. `Reset view` restores the fitted scene.
+Map navigation: use the mouse wheel or `E`/`Q` to zoom, drag with the left button to orbit, hold `Shift` while dragging (or use the right mouse button) to pan, use `W`/`A`/`S`/`D` to pan, and use the arrow keys to rotate. The scene has no distance fog; `Reset view` restores the fitted scene.
 
 
 Vertical exaggeration is a visual control only. It changes how height is displayed, not the stored OSM heights or fire propagation calculations.

diff --git a/data/processed/manhattan_buildings.json b/data/processed/manhattan_buildings.json
diff --git a/data/processed/manhattan_network.json b/data/processed/manhattan_network.json
diff --git a/data/raw/manhattan_buildings_metadata.json b/data/raw/manhattan_buildings_metadata.json
@@ -0,0 +1,21 @@
+{
+  "dataset": "Manhattan Island OSM building footprints and 3D tags",
+  "provider": "OpenStreetMap contributors",
+  "source_api": "Overpass API",
+  "source_url": "https://overpass.kumi.systems/api/interpreter",
+  "license": "OpenStreetMap data © OpenStreetMap contributors, ODbL 1.0",
+  "bbox": {
+    "south": 40.7,
+    "west": -74.02,
+    "north": 40.8825,
+    "east": -73.9065
+  },
+  "bbox_description": "Same Manhattan Island study area as the street graph",
+  "query": "[out:json][timeout:600];(way[building](40.7000,-74.0200,40.8825,-73.9065);way[\"building:part\"](40.7000,-74.0200,40.8825,-73.9065););out body geom;",
+  "origin": {
+    "latitude": 40.79125,
+    "longitude": -73.96325,
+    "source": "center of the Manhattan Island study bbox"
+  },
+  "height_policy": "Use height tag when present; otherwise estimate building:levels at 3 metres per level; otherwise retain footprint with height 0 and mark footprint_only"
+}
diff --git a/data/raw/manhattan_metadata.json b/data/raw/manhattan_metadata.json
@@ -0,0 +1,21 @@
+{
+  "dataset": "Manhattan Island urban street network",
+  "provider": "OpenStreetMap contributors",
+  "source_api": "Overpass API",
+  "source_url": "https://overpass.kumi.systems/api/interpreter",
+  "license": "OpenStreetMap data © OpenStreetMap contributors, ODbL 1.0",
+  "bbox": {
+    "south": 40.7,
+    "west": -74.02,
+    "north": 40.8825,
+    "east": -73.9065
+  },
+  "bbox_description": "Manhattan Island and immediate shoreline study area",
+  "query": "[out:json][timeout:300];way[highway~\"^(motorway|trunk|primary|secondary|tertiary|unclassified|residential|living_street|service|road|track|pedestrian)$\"](40.7000,-74.0200,40.8825,-73.9065);out body geom;",
+  "origin": {
+    "latitude": 40.79125,
+    "longitude": -73.96325,
+    "source": "center of the Manhattan Island study bbox"
+  },
+  "downloaded_for": "urban-fire-spread detailed city replacement"
+}
diff --git a/data/raw/manhattan_osm_buildings.json b/data/raw/manhattan_osm_buildings.json