Repair the three layers a walking score has to stand on

[richardkfm]

Summary

Groundwork for the walkability view (slice 3a of three). While planning the walk score, three of its four inputs turned out to be broken or missing — and each one would have made the score lie rather than merely fall short. They are repaired here, on their own, before anything is built on top of them.

Pedestrian crossings synced nothing. The Overpass query ended out tags, whose verbosity prints ids and tags without node coordinates, so every crossing was discarded by the element converter before it could be stored. The layer reported success and held an empty set. It now asks for geometry, and a new normaliser gives each crossing a stable reading of whether it is signalised, marked, dropped-kerb and tactile — three-state throughout, so a crossing surveyed as unmarked stays distinct from one nobody has looked at.

Speed limits were unreadable outside the countries that tag a bare number in km/h. maxspeed had no normaliser at all, so 20 mph and DE:urban reached every consumer as raw strings and drew as unknown. They are now parsed once into a comparable figure. Two cases deliberately carry no number: an unrestricted road, because 0 km/h would rank an autobahn as the calmest street in town, and a national zone, because what urban means is a question of local law — the zone is preserved and the workspace supplies the value, rather than a table of one country's defaults being baked into core code.

A street whose pavement is mapped separately was recorded as having none. sidewalk=separate says the footway exists and is drawn as its own line; it was sitting in the set of values meaning "surveyed, nothing here". Left alone, the walking score would have condemned exactly the cities that map pedestrian space best. It now reads as a pointer — footway_present: null with sides: "separate" — and outranks a no on the opposite side, because one side mapped separately still means there is a pavement to find.

Also: kissing gates and stiles now report affects: "walking", a value docs/AREA_TARGETS.md has described since it was written but which nothing ever emitted; and the map's speed colouring reads the new numeric field, keeping the old string match behind a coalesce guard so workspaces synced before this change render exactly as they do today.

No model changes, so no migration.

Type of change

• Bug fix
• New feature / connector / measure rule
• Refactor or cleanup
• Documentation
• CI / infrastructure

Checklist (CLAUDE.md governance)

• I have read CLAUDE.md and my changes comply with all core principles
• CHANGELOG.md has a new entry under [Unreleased]
• README.md is updated if user-facing behavior changed
• VERSION is bumped if warranted (see versioning policy in CLAUDE.md) — 0.51.0 → 0.52.0
• No logic is hard-wired to a specific city, country, or language
• All new user-facing strings use gettext / {% trans %} — no new user-facing strings in this change
• docker compose up --build runs cleanly and core flows work
• python manage.py test passes

The last two are unticked because they could not be run. This container has Django but no GDAL, so every GeoDjango import fails and neither manage.py test nor docker compose up can start. CI runs both. Rather than tick a box on a command I did not execute, they are left open.

Testing notes

34 new connector tests, all green, plus the existing suite unchanged:

• MaxspeedParserTests — km/h, mph (30 mph and 30mph), walk, none, four national zones, nonsense, absence. One test asserts DE:urban and NZ:urban parse identically, so that if either ever resolves to a number this feature has acquired a home country and the test says so.
• StreetsWithSpeedNormalizerTests — the comparable number, the mph city, the preserved zone, and that the raw value always survives whatever we failed to parse.
• PedestrianCrossingTemplateTests — the template asks for geometry, a crossing node survives the element converter, and a crossing without coordinates is dropped. The last one records the failure mode so a future out tags cannot creep back in unnoticed.
• PedestrianCrossingNormalizerTests — signalised, marked, unmarked, level crossing, unknown; tactile paving three-state; an unrecognised kerb value is not invented; crossing:markings=surface is a marking (a change of paving, not the absence of paint); signals outrank missing paint.
• ObstacleAffectsTests — step-over barriers affect walking, ordinary barriers still affect both, tram tracks are still a cycling problem, and affects never leaves its documented domain.
• FootwayNormalizerTests — the sidewalk=separate test that asserted the old behaviour is replaced by four: separate is a pointer not an absence; separate stays distinct from silence; separate on one side outranks no on the other; a surveyed side still wins over separate.

Verification run:

• ruff check backend/ — clean.
• python3 -m unittest connectors.tests — 232 tests, with the same 1 pre-existing failure and 16 pre-existing "settings are not configured" errors as the base commit. Confirmed by running the identical command in a clean git worktree at 28e06d4: 198 tests, identical failure counts. None of them touch changed code.
• python3 -m unittest measures.test_geo measures.test_street_space measures.test_parking_estimate — 102 tests green.
• map.html compiles as a Django template, and its extracted JavaScript passes node --check.

Not verified, and it is the first thing worth checking in review: that the crossings layer now stores a non-zero record count. That needs a live Overpass call, and outbound access to overpass-api.de is blocked from this container. The out tags diagnosis is from the Overpass specification and from _osm_element_to_feature returning None for a coordinate-less node — out geom tags is the correct form for a node query either way, so the fix is safe even if the diagnosis is wrong.

Suggested manual check in docker compose up --build:

1. Sync pedestrian_crossings in a demo workspace; confirm a non-zero record count and that /api/v1/workspaces/<slug>/features/pedestrian_crossings/ returns Point geometries.
2. Re-sync streets_with_speed; confirm maxspeed_kmh is populated, maxspeed_source is honest, and at least one feature keeps its maxspeed_zone.
3. Confirm the speed colouring still renders for a workspace synced before this change (the coalesce fallback).
4. Confirm a sidewalk=separate street reads footway_present: null, sides: "separate".

Related issues

Follow-up worth filing separately, not fixed here: transit_stops and trees use the same out tags query and may be silently empty for the same reason. That is a wider blast radius than this change and deserves its own issue rather than a quietly widened pull request.

Next in this series: 3b adds measures/walkability.py and the /walkability/ endpoint; 3c draws the coloured line and the parking-vs-walking story view.

🤖 Generated with Claude Code

https://claude.ai/code/session_01CKiSPbDwCocoqdk3zoty9w

---

Generated by Claude Code