quick fixes

quick fixes

add policy-demo upload + submit upload UI

quick fixes

HF mirror: only bootstrap leaderboard JSONs at boot (was timing out on ckpt download); lazy-fetch ckpts on demand

checkpoint_loader: patch Box.shape to tuple before initialize_*_agent (Hanabi exposes scalar shape)

API: simultaneous-env demo emits per-player frames; nicer Play game log (LBF/OC narration, mini-hanabi uses Hanabi log); hide empty difficulty pill; uploaded_partners endpoint + upload: kind in demo

API: HF dataset mirror for leaderboard + ckpts; fix ego_fn obs.get bug

API: fix partner not acting (current_player unwrap + simultaneous-step handler), -w 1 to keep sessions, drop difficulty labels

fix OBL + BC weight_file paths; Flawed description; flag Prolific as LBF-only

Flawed partner: use play_threshold kwarg (not mistake_prob), drop bogus second entry

API: refresh hanabi heuristics (drop weighted rule_based, add cautious/internal, fix flawed_60 key); About/Home wording + card alignment

UI polish: LBF caps, About page, GitHub nav link, UT orange accent

re-deploy: study.tsx full play.tsx parity + runtime weight download

deploy: demo viz, prolific study sessions, mini-hanabi, client-eval submit

Initial jaxaht-benchmark deployment