posters_service.render_season_poster() now fetches /dilbert/sports/
formula1/cars/{year}.jpg from the files-api CDN, centre-crops to 2:3,
and composites a vertical dark gradient so the FORMULA 1 / year /
WORLD CHAMPIONSHIP text stays legible over any photo.
Dockerfile:
- python:3.12-slim → python:3.12-alpine (smaller, no fastf1 leftover)
- fonts-dejavu-core → ttf-liberation: same font Arch dev ships, so
posters render byte-identical on dev and in the container.
- Drop fastf1 (unused in the served code path); add numpy explicitly
because models/geo_json/geometry.py imports it at module load.
GET /posters/race/{grand_prix_name}/{season}.png
GET /posters/season/{season}.png
1000x1500 PNGs composed with Pillow + cairosvg:
checkered-flag strips top/bottom, F1 wordmark, round/year ribbon,
GP name, rasterised track outline (white), race date. Season poster
is wordmark + giant year + 'WORLD CHAMPIONSHIP'. Cached under
cache/posters/ so cold renders happen once.
Indianapolis oval layout uses real OSM centerline data for the
2.5-mile oval used in the F1 championship from 1950-1960.
Madring is the new Madrid circuit for the 2026 Spanish Grand Prix,
track path extracted from Wikimedia SVG.
FastF1's get_event() couldn't resolve several grand prix names
(Qatar, Brazilian, Mexican, Barcelona), causing 500/404 errors.
Now uses Jolpica's race schedule directly — slugifies circuitName
to match against circuits.json with ASCII fallback for unicode.
Dataclass models have parent back-references (TrackLayout→Circuit→
Locality→Country) causing infinite recursion in dataclasses.asdict().
Add to_dict() methods that exclude parent refs, use them in API endpoints.