51 lines
2.3 KiB
Markdown
51 lines
2.3 KiB
Markdown
# Two-Area Native Road Comparison
|
|
|
|
## Runs
|
|
|
|
2026-08-14:
|
|
|
|
```bash
|
|
npm run road:compile -- --config config/areas/nantaizi-lake-innovation-valley.json
|
|
npm run build:area -- --config config/areas/hanyang-block.json --stages intermediates
|
|
npm run road:compile -- --config config/areas/hanyang-block.json
|
|
```
|
|
|
|
`comparison.json` is a coverage and diagnostic record. Feature counts are not
|
|
a geometry-quality score: osm2streets and the native compiler segment roads at
|
|
different levels.
|
|
|
|
| Area | Directional roads | Native surfaces | Native junctions | Movements | Internal ends | Manual candidates | osm2streets road surfaces |
|
|
| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
|
|
| nantaizi-lake-innovation-valley | 34 | 19 | 6 | 46 | 4 | 2 | 50 |
|
|
| hanyang-block | 475 | 352 | 20 | 369 | 161 | 83 | 1826 |
|
|
|
|
## Observed Failure Modes
|
|
|
|
The hanyang osm2streets/QGIS run completed, but its log reported repeated:
|
|
|
|
- roads trimmed into oblivion;
|
|
- degenerate intersections that could not be collapsed because layers, names,
|
|
highway types, or lane specifications differ;
|
|
- intersection polygon requests with no roads.
|
|
|
|
The native compiler did not hide the related uncertainty. It reported 161
|
|
internal road ends. Eighty-three have one or more nearby, direction-compatible
|
|
candidate departures within 35 metres; these remain suggestions for explicit
|
|
review rather than automatic topology edits. The initial five one-way junctions
|
|
that lacked connector curves now publish road-surface envelopes based on their
|
|
semantic movements. Across hanyang, 369 movements are identified: 243 require
|
|
a connector curve, while 126 are continuous at their OSM node and intentionally
|
|
have no separate geometry.
|
|
|
|
## Resulting Priorities
|
|
|
|
1. Keep manual candidate suggestions and semantic overrides as the correction
|
|
path for near-miss topology. Do not bulk-connect candidates.
|
|
2. Expand ordinary junction support from 3/4 physical approaches only after
|
|
identifying a repeated unsupported topology; one-way movements that are
|
|
continuous at a node are already supported without fake connector geometry.
|
|
3. Add an inspectable movement artifact so turn geometry is not the only
|
|
representation of a road-to-road movement.
|
|
4. Use a visual review of a few explicit hanyang diagnostics before changing
|
|
connection-distance or road-class rules.
|