Both Discover rec rows already exclude what you own/watch, so novelty is baked in; the missing lever
is a popularity penalty. apply_adventurousness(items, level) re-ranks dicts (score + optional 0-100
popularity) so globally-popular candidates sink as the dial rises. Pure + reusable across both rec
rows. level<=0 returns the input order unchanged (a copy) — fully additive, no regression; 1.0 applies
the full penalty (a popularity-100 pick loses 70% of its score). Missing popularity is never penalised.
5 seam tests (no-op+copy, demotion, proportional penalty, missing-pop, clamping). 37 pass. Wiring
(scan stores popularity -> routes re-rank live -> Settings slider) is the next increment.
f0a6d5e696