File size: 3,499 Bytes
8bc9528
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5c08c9a
 
 
 
 
 
 
 
 
8bc9528
5c08c9a
8bc9528
 
 
 
 
 
 
 
 
 
 
 
 
 
5c08c9a
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
# Route display orientation

- `display_rotation_deg` is **clockwise screen rotation**, in degrees, applied
  to the basemap and route together. It is NOT a GPS coordinate transform or
  a device/screen orientation setting. Never pass it to Android/Wear orientation.
- Preserve a finite rotation supplied by a route producer. Do not read an
  unversioned `parameters[3]` without verifying its units and axes.
- The main generator now recovers rotation using its existing selected GPX
  example (`shape_example_index`), or its canonical guided outline. The same
  complex correlation handles scale, start point and reversed traversal.
  Only similarity >= 0.75 gets a template-estimated orientation; this threshold
  is presentation-only, not a change to art readiness or recommendation scores.
- `display_rotation_source=template_match` means estimated alignment to the
  available template, not proof that the original placement angle was retained.
  Symmetric shapes may have multiple equivalent orientations. Rotation cannot
  remove distortion, reflection or road detours.
- Generated previews/details default to template orientation. Interactive maps
  offer a labelled button to switch back to north-up. Active running maps default
  to north-up, offer the same switch when metadata exists, and keep manual pan/
  zoom state between live updates. Results use the same angle on both maps.
- All 232 bundled samples now have a checked-in per-ID orientation catalogue
  (`sampleRouteOrientations.json`), attached on load in search, nearby sample
  maps and saved-course lists, and handed off to actual course detail/run views.
  It aligns to upright category outlines; it does not invent a historical
  placement angle that was never stored. The two running-dog variants were
  visually corrected and approved-road near-duplicates inherit source poses.
  `tools/review_sample_orientations.py` regenerates the catalogue and full
  before/after contact sheets without touching the public GPS route files.
  Alignment scores are diagnostic only, not route-quality acceptance scores.
- Angle metadata survives generated detail, course preview, local bookmark and
  run checkpoint storage. Other legacy/shared routes without angle metadata
  remain north-up; they are not guessed from a heading or route start/end.
- Rotation uses pinned `leaflet-rotate@0.2.8` (GPL-3.0), extending Leaflet 1.9.
  Keep its upstream license and account for its license when redistributing
  the web bundle: https://github.com/Raruto/leaflet-rotate . It is loaded only
  in the client-only Leaflet component; controls and marker text stay upright.

## Checks

`python -m pytest api/ai-service/tests/test_gpx_shape_similarity.py api/ai-service/tests/test_route_display_orientation.py`

`cd main; npm run build`

`node tools/test-map-orientation.cjs` against the local app. The browser test
uses a local course fixture, not a real GPS run or quality acceptance test.

`node tools/test-sample-map-orientation.cjs` selects all **232 actual bundled
samples** through the saved-course UI, verifies each displayed angle, unchanged
GPS polyline and fitted viewport, then checks four real route-search galleries
and course-detail handoff. Only user/auth/location state is mocked. Bulk tile
requests are stubbed to avoid downloading hundreds of maps; evidence captures
clear the temporary browser cache and reload real OSM tiles. This is orientation
verification, not a new field-safety or Strava Art quality acceptance decision.