Docs / Brainstorm stack — complete implementation map

Brainstorm stack — complete implementation map

This chat’s brainstorms are implemented and wired to Minecraft DQN + MathNN.

One command

minecraft dqn pack stack enable
# or start training immediately:
minecraft dqn pack stack enable train
minecraft dqn pack stack doctor

Checklist

Brainstorm	Status	Entry
Jump/place when stuck	Done	`unstuck on` (situation tiers 5/10/20/21/25)
Epsilon profiles + packs	Done	`pack apply default\\|scout\\|sniper\\|…`
Live ε schedules	Done	`anneal_fast\\|anneal_slow\\|stuck_boost\\|goal_boost` via pack or `epsilon schedule`
Train/eval chat	Done	`eval-chat`, `train-chat`, `speak`, narrate
MySQL chat/runs/feedback	Done	`pack mysql init\\|run\\|feedback`
Infinity WM hub	Done	`pack wm status\\|mode\\|imagine`
Synthetic infinity envs (AWM-lite)	Done	`pack synth run 32`
Neural learn methods	Done	`nn learn method pack learn_all`
Fine-tune harvest	Done	`pack ft harvest\\|export`
MathNN multi-task goal/stuck features	Done	goal type + stuck in state features
Dyna + offline + DAgger + nightly	Done	`nn learn method …`
Access map + recipes	Done	`nn access recipe brainstorm_stack`
Papers + HTML	Done	`docs/agent-packs/`
Full stack enable	Done	`pack stack enable [train]`

Architecture

pack stack enable
  → mysql tables + push packs
  → net_curiosity knobs + stuck_boost schedule
  → MathNN dueling + WM + RM
  → learn_all (dyna/per/nstep/dagger/ewc/ssl/safe)
  → wm dyna imagine 8
  → ft harvest on
  → mysql run begin
  → optional train mode

Synthetic envs

In-process tool tasks (gather_wood, gather_stone, craft_table, food_run, reach_camp) train agent minecraft_synth and inject synthetic transitions into minecraft experience for offline mix.

minecraft dqn pack synth run 64
nn learn method offline minecraft 128

MathNN inclusion

Signal	Feature / path
Goal type / progress	state features (goal + biome mix + [16])
Stuck intensity	temporal feature + unstuck override
Exploration	EffectiveEpsilon() schedules
Imagination	NeuralLearnMethods Dyna on Observe
Human	reward / rlaif / prefer → RM
Language	FT harvest (parallel to motor net)

Docs

Generated from the project markdown docs on 2026-07-24. This is a static, self-contained site.