BP+penalty control result: mode 2 (intrinsic credit quality) confirmed REAL

BP + lam=1e-2 ||f||^2 penalty trained for 30 epochs (s42):
  ep 30 final: test_acc 0.5303
  margin vs DFA-shallow 0.349: +18.13 pp

The 2x2 accuracy grid:
                  no penalty       with penalty
  BP             0.609            0.530
  DFA            0.308            0.363

Penalty effect on BP: -8 pp (capacity regularization cost)
Penalty effect on DFA: +5.5 pp (rescue from active harm)

Mode 2 (intrinsic credit quality) is confirmed REAL by this control:
even after the penalty's capacity cost, BP achieves +18 pp depth
utilization. DFA under the same penalty achieves only +1.4 pp. The
difference (~17 pp) cannot be attributed to capacity loss — it is
genuine credit-quality cost of random feedback vs true backprop
gradient.

This validates the round 19 'two distinct failure modes' framing:
mode 2 is not a penalty-induced regularization artifact.

1 files changed, 42 insertions, 0 deletions

diff --git a/results/bp_with_penalty/bp_pen_lam0.01_s42.json b/results/bp_with_penalty/bp_pen_lam0.01_s42.json
new file mode 100644
index 0000000..81836ae
--- /dev/null
+++ b/results/bp_with_penalty/bp_pen_lam0.01_s42.json
@@ -0,0 +1,42 @@
+{
+  "config": {
+    "seed": 42,
+    "epochs": 30,
+    "lr": 0.001,
+    "wd": 0.01,
+    "lam": 0.01,
+    "output_dir": "results/bp_with_penalty"
+  },
+  "final_acc": 0.5303,
+  "log": [
+    {
+      "epoch": 1,
+      "test_acc": 0.3621
+    },
+    {
+      "epoch": 5,
+      "test_acc": 0.4054
+    },
+    {
+      "epoch": 10,
+      "test_acc": 0.4527
+    },
+    {
+      "epoch": 15,
+      "test_acc": 0.4872
+    },
+    {
+      "epoch": 20,
+      "test_acc": 0.5082
+    },
+    {
+      "epoch": 25,
+      "test_acc": 0.5285
+    },
+    {
+      "epoch": 30,
+      "test_acc": 0.5303
+    }
+  ],
+  "margin_pp": 18.130000000000003
+}
\ No newline at end of file