MineChoice bucket training checkpoint 118
This model has not beaten Minecraft. No held-out full-game success rate or server-verified Ender Dragon kill exists for this checkpoint.
This immutable checkpoint is a 73-input, 45-action categorical MLP trained on real local Minecraft Java 1.20.4 reward transitions. It uses direct structured state, no pixels, and no teacher fallback. Coded game skills execute selected actions; these weights alone are not a standalone bot.
Seed 384455844 began naturally with an empty inventory and no
privileged setup in 20260928-061257-train-384455844. The bucket was crafted with three iron
ingots in later continued-world episode 20260928-072932-train-persistent-001. Its reward transition
shows bucket 0 to 1 and iron ingots 3 to 0. This was not an uninterrupted
fresh-world completion trial. The episode was interrupted for a code repair
after the bucket milestone and its 35 real reward
transitions were trained into update 118 (6175 cumulative
optimized transitions). There is no measured repetition rate.
model.safetensors and policy.json contain identical neural weights.
config.json gives the exact input and action order. milestone_evidence.json
records the lineage and scope. The bundled policy schema is the episode's
source snapshot. Training ran locally on an RTX 3090. No world save,
credential, or raw gameplay log is included.
Diamond acquisition, Nether entry, End entry, and game completion remain unverified for this checkpoint.