MineChoice fresh-world house and stone-pickaxe training checkpoint
This model has not beaten Minecraft. No held-out full-game evaluation or server-verified Ender Dragon kill has been demonstrated.
This is immutable reward-training checkpoint 43 from local Minecraft Java 1.20.4 survival play. The 73-input, 45-action MLP uses direct structured state, not pixels. Coded skills execute its categorical decisions; there is no teacher fallback at inference. The classifier is not a standalone game bot.
Episode 20260928-011705-train-353570031 began in a fresh naturally generated world, seed 353570031,
with empty inventory and no privileged gameplay setup. Its trace verified one
complete 52/52-block house and later a stone pickaxe, both with gameplay reward
transitions. The episode was interrupted after repeated unreachable descents;
its 152 recorded reward transitions then trained this checkpoint. These are
training milestones, not a measured fresh-world success rate or proof this
checkpoint will repeat them. milestone_evidence.json gives exact inventory
and reward transitions.
model.safetensors and policy.json contain the same verified neural weights.
config.json records the input and action order. The bundled JavaScript policy
schema is the episode's source snapshot. The model was trained on a local RTX
3090; inference does not require a hosted service.
This checkpoint has not demonstrated an iron pickaxe, diamond acquisition, Nether entry, End entry, or game completion. No world save, private credential, or raw training log is included.