docs(SAC_LSTM_Bot): campaign-v1 notebook — locked config + story so far (#56)

2026-08-21 23:26:31 +02:00
parent 6a294ad7ad
commit 2f49cb243f
1 changed files with 72 additions and 0 deletions
@@ -0,0 +1,72 @@
+# Campaign Notebook — campaign-v1 (SAC_LSTM_Bot)
+
+Overnight training campaign on branch `research/goto-controller`.
+Companion tickets: config+launch = **#56**, morning verdict = **#57**, map = **#53**.
+Monitoring contract: observability inventory in **#55 comment 461** (9 signals, thresholds, four-way discrimination).
+
+## Locked config (campaign-v1)
+
+Launch command (tmux session `sac_campaign`, stdout teed to `campaign_stdout.log`):
+
+```bash
+SAC_OPPONENTS='Corners:3,Crazy:2,RamFire:1,Target:1,SacTwin:1' \
+SAC_EVAL_OPPONENT=Corners \
+SAC_TOTAL_ROUNDS=2000 \
+SAC_CHUNK_SIZE=10 \
+SAC_EVAL_INTERVAL=2 \
+SAC_EVAL_ROUNDS=10 \
+SAC_MAX_CRASHES=5 \
+SACLSTM_HIDDEN_SIZE=256 \
+SACLSTM_BATCH_SIZE=16 \
+SACLSTM_UTD_RATIO=1 \
+SACLSTM_SAVE_INTERVAL=20 \
+./sac_train.sh 2>&1 | tee -a campaign_stdout.log
+```
+
+| Knob | Value | Source / rationale |
+|------|-------|--------------------|
+| `SAC_OPPONENTS` | `Corners:3,Crazy:2,RamFire:1,Target:1,SacTwin:1` | Working recommendation kept. Twin pinned at **1 not 2**: #54 showed mirror battles end early ⇒ fewer transitions per chunk; weight 2 would starve the replay buffer. |
+| `SAC_EVAL_OPPONENT` | `Corners` | Pinned explicitly (= first pool entry default, #54 note) — removes reorder footgun. |
+| `SAC_TOTAL_ROUNDS` | `2000` | Sized so the **wall-clock ceiling binds first** (~100–150 rounds/h observed incl. evals ⇒ ~13–20 h of headroom). |
+| `SAC_CHUNK_SIZE` | `10` | Harness default. |
+| `SAC_EVAL_INTERVAL` | `2` | Harness default — eval every ~20 rounds ≈ every ~10 min. |
+| `SAC_EVAL_ROUNDS` | `10` | Harness default. |
+| `SAC_MAX_CRASHES` | `5` | Harness self-abort; monitor intervenes earlier at ≥3 consecutive crashes (#55). |
+| `SACLSTM_HIDDEN_SIZE` | `256` | Module default (`network.nim`); real capacity vs #49 smoke's 32; under `MaxHidden`=512 cap. |
+| `SACLSTM_BATCH_SIZE` | `16` | Module default (`integration.nim`). |
+| `SACLSTM_UTD_RATIO` | `1` | Module default. |
+| `SACLSTM_SAVE_INTERVAL` | `20` | **Deviation** from default 500: #54 proved the shutdown save never reaches disk in harness context — only mid-battle interval saves persist; 10–20 fired reliably, 50 did not in twin chunks. 20 = top of proven range, least IO. |
+| *(not pinned)* | module defaults | `LR_ACTOR/LR_CRITIC/LR_ALPHA=3e-4`, `GAMMA=0.99`, `TAU=0.005`, `TARGET_ENTROPY=-4.0`, `BUFFER_CAPACITY=500000`, `BURN_IN=8`, `TRAIN_WINDOW=16`. |
+
+**Budget**: generous wall-clock **ceiling, not a deadline** — **T+12 h** from launch. While HEALTHY per #55 discrimination rules the run continues; a monitor kills the tmux session at the ceiling or on an intervene threshold.
+
+**Fresh start**: pre-campaign `weights/` held #49-smoke 32-hidden checkpoints, incompatible with hidden=256. Archived to `weights_smoke49_backup/`; baseline re-established by a 10-round bootstrap battle vs Corners (also created the seed for `make_twin.sh`).
+
+## Phase log
+
+- **2026-08-21 ~23:30** — Config locked (this file), claim posted on #56 (comment 471). Smoke weights archived.
+- **2026-08-21 ~23:4x** — Bootstrap battle vs Corners done; baseline checkpoint + `best_score.txt` written; twin regenerated via `./make_twin.sh` from the campaign-v1 baseline.
+- **2026-08-21 ~23:5x** — Launched tmux `sac_campaign`. First-health-check: see *Incidents & checks* below.
+
+## Decision-issue index
+
+| Issue | What it decided |
+|-------|-----------------|
+| #37–#48 | Bot built: skeleton, state, actions, rewards, LSTM network, weights, SAC+LSTM training, integration. |
+| #49 | Training harness + smoke run (toy hyperparams: hidden 32). |
+| #54 | Mirror-twin sparring partner; SAVE_INTERVAL ≤20 rule; eval opponent = first pool entry. |
+| #55 | 9-signal observability inventory; CRASHED/STALLED/SLOW-LEARNER/HEALTHY discriminators; monitor thresholds. |
+| #56 | This campaign: locked config + launch (this notebook). |
+| #57 | Morning verdict — consumes this notebook + logs. |
+
+## Incidents & checks
+
+*(appended during the run)*
+
+### Launch health check (first chunks)
+
+Pending — filled right after launch.
+
+## Results
+
+*(filled by #57 / morning session)*