Module skill/luck re-evaluation (points + go-out)

After teaching AI module play (spool candidates, abort risk, go-out forks), re-run the 285k-class module matrix for both objectives and refresh the paper’s Section 15 / module taxonomy.

One-shot (recommended)

Runs points → analyze → go-out → analyze. Log under tools/nn/logs/.

# Aggressive on M4 Max 64GB (~14 workers = ncpu − 2)
MODULE_WORKERS=14 nohup bash tools/nn/run-points-then-goout-matrix.sh &
# or: MODULE_WORKERS=14 MODULE_GAMES=500 nohup bash tools/nn/run-points-then-goout-matrix.sh &

Run this in Terminal.app (not Cursor) so the farm isn’t killed with the agent shell. Resume is built in — existing cell JSON is skipped.

Prerequisites

yarn build:engine
yarn test:engine -- module-aware-ai.spec.ts spool-strategy.spec.ts hand-exchange-ai.spec.ts

Points (baseline refresh)

Keep prior JSON if you want a side-by-side; otherwise overwrite tools/nn/data/.

# Full points matrix (~285k games at MODULE_GAMES=500)
WARP12_ANALYSIS_DATA_DIR=tools/nn/data/points-modules-rerun \
  MODULE_OBJECTIVE=points MODULE_GAMES=500 MODULE_WORKERS=14 \
  bash tools/nn/run-module-analysis-parallel.sh

WARP12_ANALYSIS_DATA_DIR=tools/nn/data/points-modules-rerun \
  npx tsx tools/nn/analyze-module-results.ts

Summaries write under the data dir (module-analysis-summary.json, MODULE-ANALYSIS.md). Optionally copy the MD into docs/ for the paper pass.

Go-out (first full matrix)

Epsilon is omitted automatically when MODULE_OBJECTIVE=go-out.

WARP12_ANALYSIS_DATA_DIR=tools/nn/data/go-out-modules \
  MODULE_OBJECTIVE=go-out MODULE_GAMES=500 MODULE_WORKERS=14 \
  bash tools/nn/run-module-analysis-parallel.sh

WARP12_ANALYSIS_DATA_DIR=tools/nn/data/go-out-modules \
  npx tsx tools/nn/analyze-module-results.ts

Quick smoke (100 games/config):

WARP12_ANALYSIS_DATA_DIR=tools/nn/data/go-out-modules-smoke \
  MODULE_OBJECTIVE=go-out MODULE_GAMES=100 MODULE_WORKERS=14 \
  bash tools/nn/run-module-analysis-parallel.sh

# Points figs 11–20 + table4
WARP12_ANALYSIS_DATA_DIR=tools/nn/data/points-modules-rerun \
  MODULE_FIGURE_OBJECTIVE=points python3 tools/nn/create-module-figures.py

# Go-out figs 21–28 + table5 (separate assets; go-out labels)
WARP12_ANALYSIS_DATA_DIR=tools/nn/data/go-out-modules \
  MODULE_FIGURE_OBJECTIVE=go-out python3 tools/nn/create-module-figures.py

# Optional: shared-ID compare/contrast only (figure29)
MODULE_FIGURE_OBJECTIVE=contrast python3 tools/nn/create-module-figures.py

AI teaching shipped with this re-run

Gap	Fix
Spool not in candidates	Re-enabled via `getSpoolOptions`
Abort / double mismatch	`estimateSpoolValue` abort risk (+ Fracture penalty)
Hot Potato pass	`hotPotatoPass` heuristic
Salamander Surge	`salamanderSurge` (go-out)
Trail Momentum	`trailMomentum` (go-out)
Hand Exchange	give-back heuristic + pending actor
Desperation Dig	candidate + mild draw preference
Module skill profiles	`getWarpSkillProfile(..., modules)` + local/host/collector