Update README.md

README.md

@@ -18,13 +18,13 @@ inspection budget. A zone-equivariant GRU policy supports curriculum phases
 with changing `n_zones`. Deterministic crop-risk scoring grounds rewards
 and product alerts. An ONNX → MNN path is provided for edge inference.
 
-This is currently a research codebase, not a production alerting service.
+Research codebase, not a production alerting service.
 
 ## What is included
 
 - `weather_forecast_env.py` — Gymnasium env (belief map, action masks, triage/scarce/full budget modes, zone-order shuffle, real EpisodeContext injection)
-- `gru_weather_policy.py` — Zone-equivariant GRU extractor (`ZoneEquivariantMaskablePolicy`; shared per-zone weights; precip input scaling)
-- `train_curriculum.py` / `train_kaggle.py` — Curriculum and standalone training (`--budget-mode triage|scarce|full`)
+- `gru_weather_policy.py` — Zone-equivariant GRU extractor (`ZoneEquivariantMaskablePolicy`; shared per-zone weights; precip input scaling, `precip_scale=40.0`)
+- `train_curriculum.py` / `train_kaggle.py` — Curriculum and standalone training (`--budget-mode triage|scarce|full`, `--precip-scale`)
 - `crop_risk_scorer.py` — Deterministic, economics-calibrated risk scoring
 - `climatology.py` + `indonesia_zones.py` — Per-zone climatology and Indonesian agricultural zones
 - `evaluate_checkpoint_real.py` — Real-trajectory eval (L1 labels, belief Δ, zero_inspect control)
@@ -47,23 +47,25 @@ python train_kaggle.py \
   --out ./run_nz2_triage \
   --n-zones 2 \
   --budget-mode triage \
-  --steps 150000 \
+  --steps 2400000 \
   --clean-episode-ratio 0.80 \
   --event-spatial-correlation 0.50 \
+  --precip-scale 40.0 \
   --device auto
 ```
 
-Requires `ZoneEquivariantMaskablePolicy` from `gru_weather_policy.py`.
-Also requires `GRUWeatherFeaturesExtractor(..., precip_scale=40.0)` — not
-yet exposed as a `train_kaggle.py` CLI flag. If the installed
-`train_kaggle.py` does not already default to this, running the command
-above as-is will not reproduce the accuracy figures in Validation status
-below.
+Requires `ZoneEquivariantMaskablePolicy` from `gru_weather_policy.py` (not
+the string `"MultiInputPolicy"`). `--precip-scale` is a real CLI flag
+(default `40.0`, matching the validated recipe). **Step count matters**:
+150k steps is enough for drought/flood selection but not fungi; fungi
+selection was confirmed solved by 2.4M steps on the same recipe — see
+Validation status.
 
 ### Load
 
 ```python
 from sb3_contrib import MaskablePPO
+
 model = MaskablePPO.load("run_nz2_triage/final_model.zip")
 action, _ = model.predict(obs, action_masks=masks, deterministic=True)
 ```
@@ -83,12 +85,13 @@ python mnn_export.py \
 | Claim | Status |
 |-------|--------|
 | Weight transfer across `n_zones` changes | Supported by design |
-| Zone-order shuffle avoids fixed-slot lock-in | Observed |
-| Risk-conditioned zone choice under triage — drought/flood (synthetic, single-dirty) | Observed (~100% accuracy; holds when evaluated at higher spatial correlation than used in training) |
-| Risk-conditioned zone choice under triage — fungi/disease pressure (synthetic, single-dirty) | Open (~57% accuracy, not confidently above chance) |
+| Zone-order shuffle avoids fixed-slot lock-in | Observed, confirmed at 250k and 2.4M steps (raw per-slot action counts stay ~50/50) |
+| Risk-conditioned zone choice — drought (synthetic, single-dirty) | **Confirmed 100%** at 2.4M steps (n=76 combined, two seed blocks + ρ=0.85 transfer) |
+| Risk-conditioned zone choice — flood (synthetic, single-dirty) | **Confirmed 100%** at 2.4M steps (n=97 combined, two seed blocks + ρ=0.85 transfer) |
+| Risk-conditioned zone choice — fungi/disease pressure (synthetic, single-dirty) | **Confirmed 100%** at 2.4M steps (n=95 combined, two seed blocks + ρ=0.85 transfer) — lagged at ~57–67% through 250k on the same run; needed more steps, not a different mechanism |
 | Triage agent evaluated against real L1 trajectories | Open — all accuracy figures above are synthetic; `evaluate_checkpoint_real.py` has not yet been run on a triage-trained checkpoint |
 | Product skill vs curated L1 labels | Scorer/product path available; L1 density limits apply |
-| Multi-seed replication of the triage recipe | Open |
+| Multi-seed replication of the triage recipe | Open — the 2.4M result is one training run (`--seed 7`, confirmed via the training command; this pins the environment's episode generation but not PyTorch's weight-init RNG, which `train_kaggle.py` doesn't currently seed separately). A second seed reaching the same result has not been tried. |
 | Held-out real product / emit metrics | Open |
 
 ## Limitations
@@ -96,13 +99,12 @@ python mnn_export.py \
 - Research stack only: no on-call, feed SLA, or regulatory certification.
 - Training is largely synthetic for the moment.
 - Full-tour belief or episode-length metrics are not zone-selection skill.
-- Risk-conditioned zone selection is hazard-type-dependent: strong for
-  precipitation-driven events (drought, flood), not yet demonstrated for
-  fungi/disease-pressure risk.
 - All zone-selection accuracy figures to date are from synthetic episodes;
   the triage-trained policy has not been evaluated against real L1
   trajectories (see Validation status).
-- MNN/C++ path is export-oriented, not yet a validated production edge deployment.
+- The 2.4M-step selection-accuracy result comes from a single training run;
+  not yet replicated on a second seed.
+- MNN/C++ path is export-oriented, not a validated production edge deployment.
 
 ## Dependencies