Workstream B: add a neighbouring country, and verify `road_class` across a border

art-bot commented

2026-08-01 19:12:55 +00:00

Collaborator

No two of Iceland, Sweden and Poland share a land border. Iceland is an island;
Sweden and Poland do not touch. So the central claim of the shared road ladder — that
road_class means the same thing in every country — is argued (each country's mapping is
defensible on its own terms, with the reasoning next to it in each config) and cannot be
tested. This is the highest-value next step for the project.

Roads must meet at a border with matching classification, matching width, and no seam or
duplication. Treat "roads meet correctly at Tornio–Haparanda" as the acceptance test.

Researched 2026-07-29; the choice is still pending. Norway (Kartverket) and Finland
(NLS MTK / Digiroad) both clear every bar — open licence, no-CAPTCHA scriptable access over
plain HTTPS that fits the existing downloads: + fetch.go model, real classification
schemes, long land border with Sweden.

Finland is the recommendation: GeoPackage throughout (matching the project's hard
no-GDAL constraint), Digiroad answers roads best including surface type, EPSG:3067
everywhere. Its cost is displaced not absent — the general forest/open-land fabric lives
in MTK-maastokuvionreuna as terrain-parcel edges (29 GB), so cover needs polygon
reconstruction in SQL.
Norway would cost a materially more general GML reader first (SOSI-derived
application-schema GML: gml:Surface/PolygonPatch, mixed feature types per file, deeply
nested attributes; our readGMLTable is geometry-only and polygon-only). Against that:
the Sweden–Norway border is ~1 630 km with many crossings near populated areas versus
~600 km of mostly sparse Lapland, so it is the broader test — and the GML work would pay
off beyond Norway, since SOSI-derived GML is a common Nordic pattern.

Verify before committing: download one Finnish GeoPackage and confirm the Digiroad file
URL (the portal is an SPA over an object store — light reverse-engineering, no auth), and
check the maastokuvionreuna edge schema for left/right class attributes, which decides
whether landcover reconstruction is cheap or painful.

Expect the existing road mappings to move when the first neighbour lands. Full analysis:
docs/unification.md, "Workstream B".

**No two of Iceland, Sweden and Poland share a land border.** Iceland is an island; Sweden and Poland do not touch. So the central claim of the shared road ladder — that `road_class` means the same thing in every country — is *argued* (each country's mapping is defensible on its own terms, with the reasoning next to it in each config) and cannot be *tested*. This is the highest-value next step for the project. Roads must meet at a border with matching classification, matching width, and no seam or duplication. Treat **"roads meet correctly at Tornio–Haparanda"** as the acceptance test. **Researched 2026-07-29; the choice is still pending.** Norway (Kartverket) and Finland (NLS MTK / Digiroad) both clear every bar — open licence, no-CAPTCHA scriptable access over plain HTTPS that fits the existing `downloads:` + `fetch.go` model, real classification schemes, long land border with Sweden. - **Finland** is the recommendation: GeoPackage throughout (matching the project's hard no-GDAL constraint), Digiroad answers roads best *including surface type*, EPSG:3067 everywhere. Its cost is displaced not absent — the general forest/open-land fabric lives in `MTK-maastokuvionreuna` as terrain-parcel **edges** (29 GB), so `cover` needs polygon reconstruction in SQL. - **Norway** would cost a materially more general GML reader first (SOSI-derived application-schema GML: `gml:Surface`/`PolygonPatch`, mixed feature types per file, deeply nested attributes; our `readGMLTable` is geometry-only and polygon-only). Against that: the Sweden–Norway border is ~1 630 km with many crossings near populated areas versus ~600 km of mostly sparse Lapland, so it is the broader test — and the GML work would pay off beyond Norway, since SOSI-derived GML is a common Nordic pattern. **Verify before committing:** download one Finnish GeoPackage and confirm the Digiroad file URL (the portal is an SPA over an object store — light reverse-engineering, no auth), and check the `maastokuvionreuna` edge schema for left/right class attributes, which decides whether landcover reconstruction is cheap or painful. Expect the existing road mappings to move when the first neighbour lands. Full analysis: `docs/unification.md`, "Workstream B".

art-bot referenced this issue

2026-08-01 19:15:32 +00:00

CLAUDE.md is too long #7

art commented

2026-08-03 08:40:35 +00:00

Owner

Let's try Finland. Try reaching feature parity with Sweden if possible. While researching Finland, please make a full catalogue of the data that's available to us in the country specific file in docs (the thing that we need to do for the already existing countries in issue #16).

art-bot was assigned by art

2026-08-03 08:40:42 +00:00

art-bot referenced this issue from a commit

2026-08-03 16:10:51 +00:00

finland: add the fourth country, and test the road ladder across a border

art-bot closed this issue

2026-08-03 16:10:51 +00:00

Rows
Columns

Workstream B: add a neighbouring country, and verify road_class across a border #17

Workstream B: add a neighbouring country, and verify `road_class` across a border #17