City Walk/WissenZum StadtspaziergangViewer
ImpressumDatenschutz
Leitfaden
  • So funktioniert der Stadtspaziergang
  • Woher die Daten kommen
  • Quellen nach Stadt
  • Wie der Viewer zu einer Angabe kommt
  • Der Weg der Daten
  • Bedienung
  • Glossar
Guide (English)
  • How the city walker works
  • Where the data comes from
  • Sources by city
  • How the viewer comes by what it shows
  • From download to browser
  • Using the viewer
  • Glossary
Entwicklung (English)
  • docs/ — the project knowledge base
  • Rendering — how the data becomes pixels
  • Data pipeline — the bakes, the build step and the artifact contracts
  • Architecture decision records
  • Data → feature provenance
  • Transformation catalog
  • Portability — rendering any location, with whatever data it has
  • Writing a walk
  • Routines: the twin's scheduled agents
  • Implementation plans, backlog and audit history
    • Completed plans — condensed records
    • Plan 086 [rewrite]: The rewrite series — faster on the device, fewer bytes on the wire, less to hold in the head
    • Plan 017: Any German city — the open half: a second Land, NAS input, OSM land cover, a site check
    • Plan 019: The GPU checklist — look at everything built headless, on real GPUs and phones
    • Plan 021: /wissen on Astro Starlight instead of a hand-built Next route
    • Plan 047 (spike): "Problem melden" — the crash report with somewhere to go
    • Plan 048 (spike): Share a view as a link, and commit the reference views the GPU checks need
    • Plan 050: Landmarks and what LoD2 leaves out — measured in DOM1, named by OSM and Wikidata
    • Plan 051: Stand-ins where a Land publishes less, and looks derived from each city's data
    • Plan 052: A queryable twin — identity, semantics and provenance, asked on demand
    • Plan 053: Time and live sources — the twin's fourth dimension
    • Plan 054: Scenarios — what-if studies drawn in the scene's own hand
    • Plan 069: Find what holds the load at 52 %, then take it off the tile path
    • Plan 070: Each tile holds a fraction of the raster memory it holds today
    • Plan 071: A phone renders a lean scene, a desktop a lighter one
    • Plan 074: Bake the static dressing, load the optional features on demand, one store for the HUD
    • 076 — Streaming without holes: the ground first, a plate under everything, coarse buildings and trees from the start
    • Plan 077: One rasters module — a tile's rasters loaded, shared and weighed behind one interface, their files named by the artifact table
    • 078 — One scene environment the dressings bind at birth
    • Plan 079: A page that lost its GPU comes back where it stood, through one recovery session
    • 080 — One row per inquiry kind: the twin's per-kind dispatch in one place
    • 082 — One answer turn, and a code for every way it ends
    • 084 — One session for an answer's lifetime: reading aloud, dictation and the scene's waits in one module
    • 085 — Window niches and shop windows: performance and follow-ups
    • Plan 087 [rewrite]: Plates compared on a real GPU, and one ledger every plan of the series writes its figures into
    • Plan 088 [rewrite]: The tileset is a published product, named by a lock file; the web build bakes nothing and data/ leaves git
    • Plan 089 [rewrite]: The bakes are type-checked in CI against a baseline that only shrinks
    • Plan 090 [rewrite]: A tile's dressing arrives in the layout the GPU draws from
    • Plan 091 [rewrite]: A tile's rasters arrive as the GPU reads them — plain arrays for ids, blocks for soft fields, a few files and bindings
    • Plan 092 [rewrite]: The bakes work on site-wide typed tables, cut tiles last, and no step rewrites another's file
    • Plan 093 [rewrite]: The stream's tile is smaller than the bake's — a quadtree whose levels replace the raster twins, built only if a phone still asks for it
    • Plan 094 [rewrite]: The clay's limits are counted by a test, and every pipeline the scene draws with is known at boot
    • Plan 095 [rewrite]: The viewer is a package with one entrance, a coordinate's frame is its type, and bootApp is owners
    • Plan 096 [rewrite]: Every bake heuristic that guesses has a reference set and a score a bake PR must hold
  • What the data streams carry that the walker does not use yet
Contents
Leitfaden
  • So funktioniert der Stadtspaziergang
  • Woher die Daten kommen
  • Quellen nach Stadt
  • Wie der Viewer zu einer Angabe kommt
  • Der Weg der Daten
  • Bedienung
  • Glossar
Guide (English)
  • How the city walker works
  • Where the data comes from
  • Sources by city
  • How the viewer comes by what it shows
  • From download to browser
  • Using the viewer
  • Glossary
Entwicklung (English)
  • docs/ — the project knowledge base
  • Rendering — how the data becomes pixels
  • Data pipeline — the bakes, the build step and the artifact contracts
  • Architecture decision records
  • Data → feature provenance
  • Transformation catalog
  • Portability — rendering any location, with whatever data it has
  • Writing a walk
  • Routines: the twin's scheduled agents
  • Implementation plans, backlog and audit history
    • Completed plans — condensed records
    • Plan 086 [rewrite]: The rewrite series — faster on the device, fewer bytes on the wire, less to hold in the head
    • Plan 017: Any German city — the open half: a second Land, NAS input, OSM land cover, a site check
    • Plan 019: The GPU checklist — look at everything built headless, on real GPUs and phones
    • Plan 021: /wissen on Astro Starlight instead of a hand-built Next route
    • Plan 047 (spike): "Problem melden" — the crash report with somewhere to go
    • Plan 048 (spike): Share a view as a link, and commit the reference views the GPU checks need
    • Plan 050: Landmarks and what LoD2 leaves out — measured in DOM1, named by OSM and Wikidata
    • Plan 051: Stand-ins where a Land publishes less, and looks derived from each city's data
    • Plan 052: A queryable twin — identity, semantics and provenance, asked on demand
    • Plan 053: Time and live sources — the twin's fourth dimension
    • Plan 054: Scenarios — what-if studies drawn in the scene's own hand
    • Plan 069: Find what holds the load at 52 %, then take it off the tile path
    • Plan 070: Each tile holds a fraction of the raster memory it holds today
    • Plan 071: A phone renders a lean scene, a desktop a lighter one
    • Plan 074: Bake the static dressing, load the optional features on demand, one store for the HUD
    • 076 — Streaming without holes: the ground first, a plate under everything, coarse buildings and trees from the start
    • Plan 077: One rasters module — a tile's rasters loaded, shared and weighed behind one interface, their files named by the artifact table
    • 078 — One scene environment the dressings bind at birth
    • Plan 079: A page that lost its GPU comes back where it stood, through one recovery session
    • 080 — One row per inquiry kind: the twin's per-kind dispatch in one place
    • 082 — One answer turn, and a code for every way it ends
    • 084 — One session for an answer's lifetime: reading aloud, dictation and the scene's waits in one module
    • 085 — Window niches and shop windows: performance and follow-ups
    • Plan 087 [rewrite]: Plates compared on a real GPU, and one ledger every plan of the series writes its figures into
    • Plan 088 [rewrite]: The tileset is a published product, named by a lock file; the web build bakes nothing and data/ leaves git
    • Plan 089 [rewrite]: The bakes are type-checked in CI against a baseline that only shrinks
    • Plan 090 [rewrite]: A tile's dressing arrives in the layout the GPU draws from
    • Plan 091 [rewrite]: A tile's rasters arrive as the GPU reads them — plain arrays for ids, blocks for soft fields, a few files and bindings
    • Plan 092 [rewrite]: The bakes work on site-wide typed tables, cut tiles last, and no step rewrites another's file
    • Plan 093 [rewrite]: The stream's tile is smaller than the bake's — a quadtree whose levels replace the raster twins, built only if a phone still asks for it
    • Plan 094 [rewrite]: The clay's limits are counted by a test, and every pipeline the scene draws with is known at boot
    • Plan 095 [rewrite]: The viewer is a package with one entrance, a coordinate's frame is its type, and bootApp is owners
    • Plan 096 [rewrite]: Every bake heuristic that guesses has a reference set and a score a bake PR must hold
  • What the data streams carry that the walker does not use yet
Entwicklung (English)

Plan 051: Stand-ins where a Land publishes less, and looks derived from each city's data

Executor instructions: Read fully first. The code is built and baked where noted; what is open is the look on a real GPU (bun run shots --headed, full profile, SHOTS_SITE=<id>), the re-bakes listed under Open items, and the items a stand-in cannot fix. Update the status row in docs/plans/README.md when one lands.

Builds on ADR 0037 (sites, providers, per-tile steps read from the site config), plan 050 (landmarks, the roof relief, OSM materials) and ADR 0041 (bridges measured in DOM1). The decisions are ADR 0012.

Drift check (run first): git log --oneline -5 -- pipeline/bake/rail_osm.py pipeline/bake/ndvi.py pipeline/bake/osm_buildings.py lib/city/valley-fog.ts lib/city/source-matrix.ts

Status

  • Priority: P2 (every city but Dresden looked poorer than its data allows)
  • Effort: M (eight small items, two of them by other threads)
  • Risk: LOW–MED — a stand-in weaker than what it replaces; a derived look drifting from what was tuned on Dresden
  • Planned at: 2026-10-01
  • Status: BUILT (2026-10-01), looks unverified on a real GPU. Baked: Hamburg's rails and decks, Hamburg's and Leipzig's registers, Munich's vegetation index, Hamburg's facade context (brick on all four tiles), Unna's relief floor. Not re-baked: osm-buildings on the other sites (their files carry no context, which builds as plaster — the old default), the landmarks (the floor) and the relief floor on the other sites.

Goal

Generalise the one-off fixes so that every city benefits. A city added under ADR 0037 got exactly what its Land publishes, and a step whose source was missing wrote an empty file; some looks were tuned on Dresden and applied everywhere, or set per site by hand. Two rules came out of it (ADR 0012):

  1. Where a Land or a city lacks the best source, read a named stand-in that measures or maps the same thing — and mark it per city, generated from the configs, so the difference is visible.
  2. No per-site look switches: derive a look that differs between cities from the city's own data, in the bake or the build.

What was done

  1. OSM rails, ballast and bridge decks without a Basis-DLM (Hamburg, Berlin). pipeline/bake/rail_osm.py reads the rail ways (rail/light_rail/subway/narrow_gauge, out of tunnels, with tracks and electrified), buffers their beds into ballast, and hands the bridge-tagged road, path and rail ways and the man_made=bridge outlines to rail.py's deck code (decks(..., merge=True)): a way on an outline already claimed adds nothing, the loose buffers of one bridge on one layer merge into one deck (merged_decks) — OSM draws a carriageway, its pavements and a cycle track as separate ways. Ground(tile, networks) classifies an unclaimed outline from OSM's networks. Deck heights, ribs, the fairway and Wikidata are unchanged. The rail step now runs before canopy, and canopy.bridge_mask reads the baked bridge_<tile>.geojson without a DLM, so no crown grows on an OSM deck.
  2. The landmark relief stands on the roof under it (structures.py relief: grid.floor per cell, the lowest LoD2 roof in the 3 × 3 around it, roof cells only, sunk 0.3 m; fill_voids for DOM cells > 1 m below the LoD2 roof; lib/city/structures.ts reliefMesh drops each edge wall to its cell's floor). Found on Unna's Evangelische Stadtkirche (plan 050, Findings).
  3. Facade material from the neighbourhood (osm_buildings.py apply_contexts): mapped wall materials within 300 m vote, distance weighted (brick for, plaster/stone/concrete/wood against, glass and metal abstain); ≥ 6 votes locally, else the tile and ≈ 350 m around it with ≥ 20, else plaster. The file's top-level context is the tile's, an object's own only where it differs. Site.facades is gone; prepare-data.ts passes the tile's context, bake-city-mesh.ts uses look.context ?? facades, a part inherits its root's (inheritedLook).
  4. A vegetation index from the visible bands for a DOP without infrared (Bavaria): ndvi.py gli_raster, GLI = (2G − R − B) / (2G + R + B), mapped by 3.3 · GLI + 0.14 onto the NDVI's scale.
  5. Tram gauge from OSM: tram.py gauge_of → the track's g (m); TRAM_GAUGE is the standard-gauge fallback 1.435 m; tram-layer.ts addTrack lays the rails at ± half the track's own gauge.
  6. Valley-haze depth from the site's relief (lib/city/valley-fog.ts): prepare-data.ts samples each tile's coarse ground on a 32² grid, the tileset's extras.ground carries the site's 2nd and 90th height percentile, create-app.ts sets the fade height to 0.6 × that relief, clamped to 8–28 m.
  7. Landmarks by relative notability (landmarks.py): ≥ 8 % of the tile's top sitelinks (NOTABLE_SHARE, never under 2), 40 a tile as a safety cap instead of a fixed 12 — plan 050's Dense tiles item.
  8. The generated Sources by city page: lib/city/source-matrix.ts (pure: a row per drawn layer, a column per configured site, each cell 🟢 best / 🔵 OSM only / 🟡 stand-in / ⚪ none, with the reason), scripts/docs-matrix.ts (bun run docs:matrix) writes docs/guide/{en,de}/sources-by-city.md, scripts/docs-matrix.test.ts fails when they differ.

Done in the same round by other threads, recorded here for the whole picture:

  • Tree registers for Hamburg, Leipzig and Berlin: cadastre.py holds each register as data — the WFS (service, feature types, output format, CRS) and a field mapping onto one tree record (REGISTERS; lib/city/site.ts TREE_REGISTERS). Hamburg's Straßenbaumkataster (BUKEA, dl-de/by-2-0, street trees, no heights: measured in DOM1 − DGM1 around the trunk where that fits the crown), Leipzig's Baumkataster (Amt für Stadtgrün und Gewässer, dl-de/by-2-0, street and park trees; felled ones and stands skipped), Berlin's Baumbestand (Geoportal Berlin, dl-de/zero-2-0, street and park trees). Sites hamburg, leipzig, berlin name them (treeCadastre).
  • Bavaria's laser scan: providers/by.py lsc fetches the LDBV's classified laser points (CC BY 4.0, four 1 km LAZ a tile, merged) and maps their classes into the AdV scheme (LSC_CLASSES: 6 building and 20 object points → 20; 9, 23, 24 → DTM only; class 22 by the file's year); every scan's low-return intensities are normalised by its own ground median (lsc.rasterise). See the ledger and data-pipeline.md.

Results

  • Hamburg (four tiles): 348 bridge decks — the Busanbrücke, the Brooktorkaibrücke, the Kuhmühlenbrücke, the Berlin–Hamburg and Lübeck–Hamburg lines' bridges among them —, 81 km of track and 0.27 km² of ballast, where there were none. 10 418 register trees.
  • Leipzig: 25 498 register trees.
  • Munich: ndvi_<tile>.png on its four tiles from the GLI. The line was fitted on Unna's two summer DOP tiles (NRW's RGBI): r ≈ 0.7 against the NDVI, 85 % agreement on NDVI > 0.3; on Hamburg's tiles r 0.75–0.86.
  • Facades — mapped wall materials in the baked files, the votes the context counts: Hamburg 106 brick / 11 plaster (its four tiles vote brick, so the clinker look Hamburg had by hand now comes from its data), Unna 45 plaster / 0 brick, Munich 272 plaster / 38 brick, Dresden 866 plaster / 246 brick.
  • Tram gauge: Dresden 1450, Leipzig 1458, Munich 1435 mm, each from OSM.
  • Valley haze: Dresden, Grimma, Meißen and Unna keep 28 m (the cap); Hamburg ≈ 9.6 m, Leipzig ≈ 12 m, Munich ≈ 15 m.
  • Unna's Stadtkirche: the tower's relief stands on the nave's slope (floor down to −10.1 m under the 136.3 m roof) instead of hovering at the ridge, without the "legs" the sound openings tore into it.

Open items

  • GPU plates not judged: the OSM decks and track in Hamburg (the U3 viaduct, the Hauptbahnhof's approaches, from the vantage and on foot), Munich's crowns and meadows under the GLI, a brick quarter beside a plaster one, the shallower haze in Hamburg and Munich, Unna's tower.
  • Re-bakes: osm-buildings on every site but Hamburg (so Dresden's 246 mapped brick walls can turn their quarters brick), landmarks (with the Wikidata cache; the committed files still hold the fixed 12) and then structures — only Unna's structures carry the relief floor yet; a file without it builds the old walls.
  • Hamburg's register has no heights: 82 % are measured in DOM1, the rest come from the crown — fine for street trees, weaker under a closed canopy where the highest texel is a neighbour's.
  • Berlin is configured, not built: its adapter is untested (the portal was unreachable from the agent's container); its register, OSM rails and decks wait for a first fetch and bake.
  • Hamburg publishes no laser scan (it declined to, citing privacy): no sheds, no extra crowns, OSM hedges at their tag height. No stand-in: DOM1 alone cannot tell a shed from a crown.
  • The GLI is weaker than the NDVI: it tells green from grey well and vigour less well; a dark conifer and a dry lawn read alike. If Bavaria opens its CIR orthophoto, read it instead.
  • OSM rails per track: OSM maps each track as its own way where the DLM bundles them (tracks), so a yard is many single pairs rather than one bundled line, and the ballast is the beds' union, not the DLM's surveyed yard area. Acceptable from eye height; check from the air.

Docs

Ledger (Rails, ballast and decks from OSM, Vegetation index from the visible bands, Facade context, the haze depth under Height-term fog, the tram gauge, the relief floor, the landmark floor, five 🗃️ rows), data-flow.md (both diagrams), the rendering.md codebook, data-pipeline.md, portability.md, the guide's data sources (en + de) and the generated Sources by city, ADR 0012, AGENTS.md.

Previous
Plan 050: Landmarks and what LoD2 leaves out — measured in DOM1, named by OSM and Wikidata
Next
Plan 052: A queryable twin — identity, semantics and provenance, asked on demand
Source on GitHub: docs/plans/051-stand-ins-and-derived-looks.md
On this page
  • Status
  • Goal
  • What was done
  • Results
  • Open items
  • Docs