Fly brain plays TrackMania: mushroom-body weights, laps and videos

Artifacts for python/tmnf_fly (see its README): the MaleCNS v1.0 mushroom body learning TrackMania tracks by dopamine reward-prediction error.

• mb_a04_5min_weights.npz, mb_a04_30min_weights.npz, mb_a02_60min_gas_weights.npz: MushroomBody.state_dict() after 5 / 30 / 60 minutes of training (A04-Acrobatic x2, A02-Race gas-only). Load with MushroomBody.load_state_dict.
• mb_*_showcase.npz: per-decision KC code, MBON drives, DAN activity, value and RPE along the video laps, plus the exact per-tick input schedule (ticks_actions). The decision sequences themselves are in the repo under python/tmnf_fly/laps/.
• mb_*_lap.mp4: the three lap videos (game chase view, compound-eye view, MaleCNS skeletons by activity, track map, MBON value / dopamine / LC4 traces). Laps: A04 6.58 s and 6.29 s (author 5.95 s), A02 21.69 s (author 16.25 s).
• mb_a04_30min.csv, mb_a02_60min_gas.csv, mb_a04_5min.log: per-minute training logs.

Game assets are not included; the code rebuilds every connectome-side input from public data (python -m tmnf_fly.fetch_data).