README: correct root cause (dynamics NMSE ~0.85 at aug onset), add three tuning arms

README.md

@@ -39,8 +39,14 @@ in the e2e arm the spread between 1st and 5th place in the top-k heap is ~7%.
 | path | arm | best epoch | val action MSE |
 |---|---|---|---|
 | `baseline/` | BC baseline | 60 | **0.003426** |
-| `seq_be/` | sequential CCIL (backward-Euler aug) | 10 | 0.003440 |
+| `seq_be/` | sequential CCIL, clean dynamics | 10 | 0.003401 |
 | `e2e/` | end-to-end CCIL | 10 | 0.003215 |
+| `e2e_fix_stepscale/` | e2e, `aug_step_scale=1.6` | 10 | 0.003260 |
+| `e2e_fix_lossweight/` | e2e, `loss_normalization=fixed`, `lambda_aug=0.3` | 10 | 0.003258 |
+| `e2e_fix_combined/` | e2e, all of the above + `aug_q_threshold=0.8`, `phase1=120` | 10 | 0.003258 |
+
+Every arm carries `*_best.pt` (validation-selected) **and** `final_epoch300.pt`
+(last epoch). They differ: best is typically epoch 10, final is epoch 300.
 
 Each directory holds `*_best.pt`, `val_top_k.json` (the full top-k selection
 record), and where available the hydra `overrides.yaml`.
@@ -88,16 +94,57 @@ than halved (0.0239 -> 0.0112) — an overfitting signature on 44 training
 episodes. `final_epoch300.pt` is included alongside `best.pt` so the gap between
 last-epoch and val-selected weights is inspectable.
 
-Measured causes, from the run's own diagnostics: the internal dynamics head
-under-predicts the latent displacement (`delta_pred_norm/delta_norm` peaks at
-0.645), so backward-Euler augmented states sit too close to the expert states;
-the relative-quality mask keeps exactly the least-displaced 50% by construction,
-which selects *against* novelty; and `loss_normalization=adaptive` rescales the
-augmentation loss to ~1.0 regardless of how uninformative it is, giving those
-near-duplicates the same gradient weight as real data. Three of the advertised
-quality gates are inert in this code path: `aug_cycle_threshold` (its
-`cycle_error_mean` is assigned to zero and never computed), `aug_hardness_min`
-(`hard_ratio` is hardcoded to 1.0), and the mask's fixed 50% quantile.
+### Why augmentation never helped — the measured cause
+
+**The latent dynamics head barely models the dynamics.** `fwd_image_nmse` is the
+normalized one-step prediction error, where **1.0 means no better than predicting
+"no change"**:
+
+| epoch | 10 | 50 | 100 | 150 | 200 | 299 |
+|---|---|---|---|---|---|---|
+| e2e | 0.968 | **0.852** | 0.779 | 0.696 | 0.631 | 0.549 |
+
+At epoch 50, when augmentation switches on, the head explains roughly **15% of
+the latent displacement variance**; even after 300 epochs it reaches only ~45%.
+Backward-Euler synthesizes `z_g = z_prev - step * fwd_head(z_prev, a_t)` and then
+trains the policy to emit `a_t` at `z_g`. When `fwd_head` is that wrong, `z_g` is
+largely noise and the policy is taught incorrect state-action associations. CCIL
+assumes the dynamics model is good enough to place a corrective label; on 44
+training episodes in a 6921-dim latent space, it is not.
+
+**Three tuning arms confirm this and rule out hyperparameters as the cause.**
+Each was run for a full 300 epochs from the same baseline checkpoint:
+
+| arm | best | top-5 epochs | post-augmentation checkpoints in top-5 |
+|---|---|---|---|
+| original | 0.003215 | 10, 150, 50, 140, 160 | 150, 140, 160 |
+| stepscale | 0.003260 | 10, 20, 40, 50, 30 | **none** |
+| lossweight | 0.003258 | 10, 20, 40, 50, 30 | **none** |
+| combined | 0.003258 | 10, 20, 40, 50, 70 | **none** |
+
+The arm with the *weakest* augmentation — the original — is the only one whose
+post-augmentation checkpoints placed at all. Correcting the displacement
+(`aug_step_scale=1.6`, bringing effective displacement from 0.645 to ~0.98 of
+true), relaxing the mask (`aug_q_threshold=0.8`, verified: `mask_ratio` 0.500 ->
+0.797), reducing the augmentation weight, and delaying warmup to epoch 120 each
+made post-augmentation checkpoints *less* competitive, not more. Amplifying a
+mostly-wrong displacement amplifies the error.
+
+Meanwhile training BC loss more than halved in every arm — the policy is
+overfitting 44 episodes, and augmentation does not counteract it.
+
+**Secondary issues, real but downstream.** Three advertised quality gates are
+inert in this code path: `aug_cycle_threshold` (its `cycle_error_mean` is
+assigned to zero and never computed), `aug_hardness_min` (`hard_ratio` is
+hardcoded to 1.0), and the mask's quantile filter, which keeps exactly 50% by
+construction regardless of absolute quality — and since `label_err` *is* the
+novelty metric, it keeps the **least**-displaced half, selecting against novelty.
+Worth fixing, but they are not why the method fails here.
+
+**What would actually change the result:** gate augmentation on measured
+`fwd_image_nmse` rather than an epoch count (e.g. require < 0.3), and give the
+dynamics head materially more data. A 100-episode session exists and is being
+evaluated separately.
 
 ## Reproducing