📖About

The all-stories grid loads each card hero as a 1600px WebP (~131KB each); ~55 cards = ~7MB, slow even with lazy-load. Fix: (1) size-key the WebP cache (shared/webp_cache.py) so multiple max_dim variants coexist (default 1600 path preserved, no mass regen); (2) the image-generator /media route accepts an optional ?w= width and serves a small cached WebP; (3) the logbook card thumbnails request ~560px. Originals untouched, fail-soft. Behavior change: cards ~4-6x smaller/faster.

⚖️Decisions

• Caller audit waived by Orion: this is a perf change to the WebP cache + the image-generator media route + the logbook card markup; it leaves the shared/pipeline boundary unchanged and does not alter bypass-runtime behaviour. The default 1600 path is preserved so existing image consumers are unaffected. — thumbnail perf fix; verified live; originals + default cache untouched. (orion)

✅Success criteria

• WebP cache keys by size (variants coexist); /media serves a small WebP for ?w=; logbook cards request small thumbnails; a card thumbnail is materially smaller than 131KB; originals untouched; existing 1600 cache preserved (must_have)