results: freeze predictor timescale choice

3 files changed, 36 insertions, 4 deletions

diff --git a/RESULTS.md b/RESULTS.md
index da8c957..1f5ca5c 100644
--- a/RESULTS.md
+++ b/RESULTS.md
@@ -131,6 +131,33 @@ SDIL matches DFA on easy MNIST and **beats DFA on FashionMNIST (+1–2 pts) and
 On CIFAR the SDIL−DFA gap GROWS with depth (+1.8 → +2.7 → +5.0 at d5/10/20): DFA degrades with depth,
 SDIL holds.
 
+## Endogenous top-down traffic timescale screen (validation only)
+
+This development screen used one model seed, a fixed 5,000-example stratified MNIST validation
+split drawn only from training data, and never evaluated the test set. All 18 runs share validation
+index hash `c4cec216...2e3d`, clean source revision `17bcd97`, and the same 15-epoch training
+protocol. The grid crossed `eta_P in {0.002, 0.01, 0.05}`, neutral warmup steps in
+`{0, 200, 1000}`, and endogenous top-down traffic scales in `{0.2, 0.5}`.
+
+| eta_P | warmup steps | rho=0.2 last val (%) | rho=0.5 last val (%) | mean (%) | worst (%) |
+|------:|-------------:|---------------------:|---------------------:|---------:|----------:|
+| 0.002 | 0 | 91.50 | 77.24 | 84.37 | 77.24 |
+| 0.002 | 200 | 91.02 | 89.10 | 90.06 | 89.10 |
+| 0.002 | 1000 | 93.10 | 88.82 | 90.96 | 88.82 |
+| 0.010 | 0 | 94.94 | 92.40 | 93.67 | 92.40 |
+| 0.010 | 200 | 92.88 | 91.12 | 92.00 | 91.12 |
+| 0.010 | 1000 | 95.04 | 91.06 | 93.05 | 91.06 |
+| 0.050 | 0 | 96.52 | 91.40 | 93.96 | 91.40 |
+| **0.050** | **200** | **94.20** | **95.26** | **94.73** | **94.20** |
+| 0.050 | 1000 | 94.70 | 93.62 | 94.16 | 93.62 |
+
+The selected development setting is `eta_P=0.05`, 200 warmup steps: it has the highest mean
+last-epoch validation accuracy and the highest worst-traffic endpoint. Its final mean traffic R2 is
+0.802 at rho=0.2 and 0.999 at rho=0.5. This reverses the initial assumption that `P` must be much
+slower than forward learning: once teaching and traffic are separated by neutral periods, fast
+tracking is beneficial, while a long warmup is not consistently useful. This is a frozen
+validation choice for the multi-seed C1 experiment, not evidence that C1 has passed.
+
 ## Legacy credit assignment / depth utility
 Per-layer alignment cos(r_l, −∇_{h_l}L), early-third average, CIFAR no-BN residual:
 | depth | DFA early-align | SDIL early-align |
diff --git a/ROADMAP.md b/ROADMAP.md
index d093fc0..d6f5588 100644
--- a/ROADMAP.md
+++ b/ROADMAP.md
@@ -36,6 +36,11 @@ held-out accuracy by at least 2 percentage points over norm-matched raw feedback
 traffic level, without harming the no-traffic setting by more than 0.5 points. The predictor must
 remove measurable soma-predictable traffic while retaining causal teaching alignment.
 
+The validation-only timescale screen froze `eta_P=0.05` and 200 neutral warmup steps by the highest
+mean last-epoch accuracy across top-down traffic scales 0.2 and 0.5 (and the best worst-scale
+endpoint). C1 must use this choice without further test-set tuning. Traffic projection seeds remain
+separate from model/data seeds and must be recorded explicitly.
+
 ### C2. SDIL scales when depth is useful
 
 Flattened CIFAR is retained as a preservation result, not as evidence that added depth is used.
diff --git a/sdil/core.py b/sdil/core.py
index cece194..8b3f923 100644
--- a/sdil/core.py
+++ b/sdil/core.py
@@ -18,7 +18,7 @@ For hidden layer l:
 
 Three learning rules, all local (no weight transport, no reverse-mode graph):
     dW_l = eta   * (r_l ⊙ phi'(u_l)) h_{l-1}^T      (three-factor: pre * gain * innovation)
-    dP_l = eta_P * (a_l - P_l h_l) h_l^T            (slow; only on neutral periods)  eta_P << eta
+    dP_l = eta_P * (a_l - P_l h_l) h_l^T            (identified only on neutral periods)
     dA_l = eta_A * (q_l - r_l) c_l^T                (on perturbation trials; q_l = causal node-pert estimate)
 
 q_l is a node-perturbation estimate of the causal descent direction
@@ -513,9 +513,9 @@ def sdil_step(net, x, y, y_onehot, cfg, step, prev_error=None):
                 dA = (qs[l] - r_list[l]).t() @ c / B   # (n_l, n_classes)
                 net.A[l] += cfg.eta_A * dA
 
-        # ================= predictor P (slow) =============================
-        # KEY timescale-separation condition. P must learn the soma->apical
-        # coupling WITHOUT absorbing the teaching signal. On "neutral periods"
+        # ================= predictor P (neutral) ==========================
+        # KEY identification condition. P must learn the soma->apical coupling
+        # WITHOUT absorbing the teaching signal. On "neutral periods"
         # the apical compartment carries only the soma-predictable nuisance drive
         # (top-down teaching c is absent), so we fit P to the c=0 apical. This
         # makes P -> nuisance map (rho*Bnuis) and leaves the innovation intact.