results: establish KP teaching-signal advantage

1 files changed, 28 insertions, 0 deletions

diff --git a/results/kp_teaching_signal_ablation_gate.json b/results/kp_teaching_signal_ablation_gate.json
new file mode 100644
index 0000000..75a2809
--- /dev/null
+++ b/results/kp_teaching_signal_ablation_gate.json
@@ -0,0 +1,28 @@
+{
+  "checks": {
+    "all_control_invariants_audited": true,
+    "dynamic_reference_is_finite": true,
+    "matched_degrades_or_collapses_by_at_least_3_points": true,
+    "raw_and_matched_use_identical_source": true,
+    "raw_degrades_or_collapses_by_at_least_5_points": true,
+    "two_frozen_controls_present": true
+  },
+  "claim": "Under four-RMS soma-predictable apical traffic, innovation subtraction is necessary relative to both raw and norm-matched KP controls.",
+  "metrics": {
+    "analysis_source_commit": "ff0069f8d0dabbb8819a726bd899381e8a95462c",
+    "clean_kp_accuracy": 0.9126,
+    "control_mean_wall_s": 8237.62871158123,
+    "dynamic_accuracy": 0.9118,
+    "dynamic_minus_matched_points": 81.18,
+    "dynamic_minus_raw_points": 81.18,
+    "matched_accuracy": 0.1,
+    "matched_finite": false,
+    "prerequisite_sha256": "48957573e6feb18e5c2dd0c6785bbd20f37126dd33e8bb7cb357190d236b4937",
+    "raw_accuracy": 0.1,
+    "raw_finite": false,
+    "training_source_commit": "31b70e1b1f2e7fd26c9c3228f525fc0b5192cd81"
+  },
+  "protocol": "kp_teaching_signal_ablation_validation_v1",
+  "scope_limit": "This is a conditional robustness result on one validation seed; it is not clean-setting or strong-baseline superiority.",
+  "status": "passed"
+}