bci: audit v2 outcome-surprise signatures

1 files changed, 68 insertions, 0 deletions

diff --git a/experiments/bci_v2_smoke.py b/experiments/bci_v2_smoke.py
index fc6509b..6b7a900 100644
--- a/experiments/bci_v2_smoke.py
+++ b/experiments/bci_v2_smoke.py
@@ -8,6 +8,7 @@ import torch
 sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 from sdil.bci import generate_trajectories
 from sdil.bci_v2 import BCIV2, BCIV2Config, run_day_v2
+from sdil.bci_v2_metrics import annotate_events, signature_metrics_v2
 
 
 def check_role_estimator():
@@ -284,10 +285,77 @@ def check_horizon_and_pairing():
     print("v2 fixed-horizon terminal accounting and paired replay: exact")
 
 
+def check_signature_metric_mechanics():
+    cfg = BCIV2Config(
+        days=2, episodes_per_day=128, steps_per_episode=5,
+        target=0.15, forward_eta=0.03,
+        feedback="performance_velocity",
+    )
+    trajectories = generate_trajectories(cfg, 9)
+    model = BCIV2(cfg, model_seed=10)
+    model.P.copy_(model.coupling)
+    model.A.copy_(model.role)
+    training = run_day_v2(
+        model,
+        trajectories,
+        0,
+        collect=True,
+        learn_role=False,
+        probe_role=False,
+        learn_predictor=False,
+    )
+    annotate_events(
+        training["events"], 0, training["success"], 0,
+        cfg.steps_per_episode,
+    )
+    modes = {}
+    settings = {
+        "intact": {},
+        "acute_critic_lesion": {"critic_enabled": False},
+        "acute_outcome_lesion": {
+            "terminal_outcome_enabled": False
+        },
+    }
+    for name, setting in settings.items():
+        report = run_day_v2(
+            model.clone(),
+            trajectories,
+            1,
+            collect=True,
+            plasticity_gain=0.0,
+            learn_role=False,
+            probe_role=False,
+            learn_predictor=False,
+            learn_critic=False,
+            **setting,
+        )
+        annotate_events(
+            report["events"], 0, report["success"], 10_000,
+            cfg.steps_per_episode,
+        )
+        modes[name] = report["events"]
+    metrics = signature_metrics_v2(
+        training["events"], modes, cfg, model.role
+    )
+    assert metrics["challenge_episodes"] == cfg.episodes_per_day
+    assert (
+        metrics["challenge_success_count"]
+        + metrics["challenge_failure_count"]
+        == cfg.episodes_per_day
+    )
+    assert all(
+        torch.isfinite(torch.tensor(value))
+        for value in metrics.values()
+        if isinstance(value, (int, float))
+    )
+    print("v2 paired terminal-outcome signature metrics: finite")
+
+
 if __name__ == "__main__":
     check_role_estimator()
     check_predictor()
     check_td_and_eligibility_updates()
     check_terminal_outcomes_and_lesions()
     check_horizon_and_pairing()
+    check_signature_metric_mechanics()
     print("ALL ORAL-B-V2 MECHANICS CHECKS PASSED")