From cb64d220efeea919c3aec676cf79715fd2a6390f Mon Sep 17 00:00:00 2001
From: YurenHao0426 <Blackhao0426@gmail.com>
Date: Wed, 22 Jul 2026 20:46:07 -0500
Subject: test: falsify oral B protocol boundaries

---
 experiments/bci_td_protocol_smoke.py | 122 +++++++++++++++++++++++++++++++++++
 1 file changed, 122 insertions(+)
 create mode 100755 experiments/bci_td_protocol_smoke.py

(limited to 'experiments/bci_td_protocol_smoke.py')

diff --git a/experiments/bci_td_protocol_smoke.py b/experiments/bci_td_protocol_smoke.py
new file mode 100755
index 0000000..ed42db5
--- /dev/null
+++ b/experiments/bci_td_protocol_smoke.py
@@ -0,0 +1,122 @@
+#!/usr/bin/env python3
+"""Endpoint-free falsification checks for the frozen oral-B R1/R2 gates."""
+import copy
+import os
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+from experiments.analyze_bci_td_confirmation import r2_checks, summarize
+from experiments.analyze_bci_td_development import seed_checks
+
+
+def check_r1_boundaries():
+    row = {
+        "conditions": {
+            "intact": {
+                "learning_gain": 0.20, "final_success": 0.80,
+                "role_cosine_after_training": 0.90,
+            },
+            "fixed_vectorizer": {"final_success": 0.50},
+            "plasticity_lesion": {"learning_gain": 0.05},
+            "oracle_role": {"final_success": 0.85},
+        },
+        "signatures": {
+            "causal_role_sign_inversion_index": 0.10,
+            "velocity_minus_error_abs_cv_corr": 0.10,
+            "mean_abs_residual_soma_corr": 0.05,
+            "raw_minus_residual_abs_soma_corr": 0.30,
+        },
+    }
+    assert all(seed_checks(row).values())
+    no_learning = copy.deepcopy(row)
+    no_learning["conditions"]["intact"]["learning_gain"] = 0.0
+    no_learning["conditions"]["plasticity_lesion"]["learning_gain"] = 0.0
+    assert not seed_checks(no_learning)["learning_gain_at_least_0p10"]
+    wrong_sign = copy.deepcopy(row)
+    wrong_sign["signatures"]["causal_role_sign_inversion_index"] = -0.01
+    assert not seed_checks(wrong_sign)["sign_inversion_at_least_0p01"]
+    no_vectorizer_gap = copy.deepcopy(row)
+    no_vectorizer_gap["conditions"]["fixed_vectorizer"]["final_success"] = 0.75
+    assert not seed_checks(no_vectorizer_gap)[
+        "fixed_vectorizer_gap_at_least_0p20"]
+    print("oral-B R1 learning, sign, and vectorizer gates: falsifiable")
+
+
+def passing_r2_inputs():
+    values = {
+        "intact_final": 0.80,
+        "intact_gain": 0.20,
+        "fixed_final_gap": 0.30,
+        "oracle_final_deficit": 0.05,
+        "plasticity_half_margin": 0.05,
+        "role_cosine": 0.90,
+        "residual_soma_corr": 0.05,
+        "raw_residual_corr_gap": 0.30,
+        "surrounding_event_accuracy": 0.60,
+        "decoder_distance_corr": 0.20,
+        "residual_outcome_accuracy": 0.65,
+        "residual_soma_outcome_gap": 0.10,
+        "sign_inversion": 0.10,
+        "velocity_advantage": 0.10,
+        "longitudinal_prediction": 0.40,
+    }
+    clustered = {name: [value] * 6 for name, value in values.items()}
+    metrics = {name: summarize(rows) for name, rows in clustered.items()}
+    return metrics, clustered, [0.90] * 30, [0.10] * 30
+
+
+def flatten(checks):
+    return [value for values in checks.values()
+            for value in ([values] if isinstance(values, bool)
+                          else values.values())]
+
+
+def check_r2_boundaries():
+    metrics, clustered, roles, signs = passing_r2_inputs()
+    checks = r2_checks(metrics, clustered, roles, signs)
+    assert all(flatten(checks))
+
+    no_learning_metrics = copy.deepcopy(metrics)
+    no_learning_clusters = copy.deepcopy(clustered)
+    no_learning_clusters["intact_gain"] = [0.0] * 6
+    no_learning_metrics["intact_gain"] = summarize(
+        no_learning_clusters["intact_gain"])
+    no_learning = r2_checks(
+        no_learning_metrics, no_learning_clusters, roles, signs)
+    assert not no_learning["learning_and_plasticity"][
+        "mean_learning_gain_at_least_0p10"]
+
+    only_24_positive = [0.10] * 24 + [-0.10] * 6
+    sign_failure = r2_checks(metrics, clustered, roles, only_24_positive)
+    assert not sign_failure["outcome_and_causal_role_vectorization"][
+        "positive_sign_inversion_in_at_least_25_of_30"]
+
+    heterogeneous_metrics = copy.deepcopy(metrics)
+    heterogeneous_clusters = copy.deepcopy(clustered)
+    heterogeneous_clusters["fixed_final_gap"] = [0.30] * 5 + [-0.20]
+    heterogeneous_metrics["fixed_final_gap"] = summarize(
+        heterogeneous_clusters["fixed_final_gap"])
+    heterogeneous = r2_checks(
+        heterogeneous_metrics, heterogeneous_clusters, roles, signs)
+    assert heterogeneous["learning_and_plasticity"][
+        "mean_fixed_gap_at_least_0p20"]
+    assert not heterogeneous["learning_and_plasticity"][
+        "fixed_gap_lower_bound_at_least_0p10"]
+
+    no_longitudinal_metrics = copy.deepcopy(metrics)
+    no_longitudinal_clusters = copy.deepcopy(clustered)
+    no_longitudinal_clusters["longitudinal_prediction"] = [0.0] * 6
+    no_longitudinal_metrics["longitudinal_prediction"] = summarize(
+        no_longitudinal_clusters["longitudinal_prediction"])
+    no_longitudinal = r2_checks(
+        no_longitudinal_metrics, no_longitudinal_clusters, roles, signs)
+    assert not no_longitudinal["outcome_and_causal_role_vectorization"][
+        "mean_longitudinal_prediction_at_least_0p30"]
+    print("oral-B R2 learning, clustered robustness, sign, and prediction gates: falsifiable")
+
+
+if __name__ == "__main__":
+    check_r1_boundaries()
+    check_r2_boundaries()
+    print("ALL ORAL-B PROTOCOL BOUNDARY CHECKS PASSED")
-- 
cgit v1.2.3