fix(sp7): flip CQL target_ratio direction — death spiral on magnitude

`target_ratio[CQL][b] = ANCHOR_CQL_RATIO * (1 - flatness[b])` had the
sign inverted: it pushed CQL HIGH when Q was flat (the collapse state)
and LOW when Q had variance. Per pearl_2_budget_kernel.cu, `flatness =
var_q / σ²`, so flatness HIGH means Q has variance — exactly when
overconfidence-risk-driven CQL pressure should kick in.

Symptom (T10-v3 train-multi-seed-x7sl2 logs): Once IQN Fix 34 woke
IQN-mag/ord/urg branches and produced real Q-targets, the controller's
inverted formula sent cql_budget to MAX_BUDGET=1.0 saturation on every
branch, the conservative pull kept Q flat, the model collapsed to
single-Flat-action eval (val_active_frac=0, dir_entropy=0,
trade_count=1, sharpe=0).

Fix is one operational character: `(1 - flatness)` → `flatness`. C51
formula was already correctly aligned (`flatness * ANCHOR_C51_RATIO`)
and unchanged.

Per pearl_controller_anchors_isv_driven.md: the inverted formula was
masked for months because the IQN-dead regime (Fix 34) suppressed
cql_raw on mag/ord/urg, never letting the controller engage with the
inverted target. Fixing the IQN regime exposed the dormant formula bug.

ANCHOR_CQL_RATIO=2.0 and MAX_BUDGET=1.0 remain hardcoded; Fix 36 will
make those ISV-driven via a GPU train_active_frac canary signal per
the pearl's "pick the canary that fires under the pathology" rule.

crates/ml/src/cuda_pipeline/loss_balance_controller_kernel.cu

@@ -20,8 +20,21 @@
 // `apply_pearls_ad_kernel` smooths into ISV).
 //
 // Math (per branch b ∈ {dir, mag, ord, urg}, per head h ∈ {CQL, C51}):
-//   target_ratio[CQL][b] = ANCHOR_CQL_RATIO * (1 - flatness[b])
+//   target_ratio[CQL][b] = ANCHOR_CQL_RATIO * flatness[b]      (Fix 35, 2026-05-03)
 //   target_ratio[C51][b] = ANCHOR_C51_RATIO * flatness[b]
+//
+// Fix 35 (formula direction flip): the original CQL formula used
+// `(1 - flatness[b])`, which inverted the relationship between Q-distribution
+// variance and CQL strength. With flatness HIGH meaning "Q has variance"
+// (per pearl_2_budget_kernel definition `var_q / σ²`), the original formula
+// pushed CQL HIGH when Q was flat — exactly when there is no overconfidence
+// for CQL to penalise — creating a death spiral on the magnitude head:
+// flat Q → CQL boosted → conservative pull keeps Q flat → repeat. Validated
+// 2026-05-03 in T10-v3 train-multi-seed-x7sl2 logs (cql_budget saturated to
+// MAX_BUDGET=1.0 on every branch by epoch 1 once IQN-mag/ord/urg woke from
+// Fix 34, and val collapsed to active_frac=0). The corrected formula
+// scales CQL with flatness: more variance → more overconfidence risk → more
+// CQL gradient. C51 formula was already correctly aligned and unchanged.
 //   slice_idx = (b == 0) ? 1 : (b == 1) ? 0 : 2  // dir, mag, trunk
 //   actual_ratio = h_norm[slice] / iqn_norm[slice]  (both > EPS_DIV via outer guard)
 //   correction   = target_ratio[h][b] / max(actual_ratio, EPS_DIV)
@@ -143,8 +156,14 @@ extern "C" __global__ void loss_balance_controller_update(
     flat_b        = fminf(1.0f, fmaxf(0.0f, flat_b));   // safety clamp
 
     // ── Compute per-branch flatness-modulated target ratio ───────────
+    // Fix 35 (2026-05-03): both heads now scale UP with flatness. When Q has
+    // variance (flatness ≈ 1) overconfidence risk is high → both CQL and C51
+    // budgets grow. When Q is flat (flatness ≈ 0) there is nothing to
+    // penalise / nothing to distribute → both budgets shrink. The prior CQL
+    // formula `(1 - flat_b)` created a death spiral on magnitude (flat Q →
+    // boosted CQL → pulled Q flatter). See kernel header.
     float target_ratio = (head == 0)
-        ? ANCHOR_CQL_RATIO * (1.0f - flat_b)
+        ? ANCHOR_CQL_RATIO * flat_b
         : ANCHOR_C51_RATIO * flat_b;
 
     // ── Read prior budget + per-branch Wiener state + activation flag ─

docs/dqn-wire-up-audit.md

@@ -4308,3 +4308,58 @@ Files: `trainers/dqn/fused_training.rs` (+12 LOC, −1 LOC), this audit
 entry. No `StateResetRegistry` changes — `cached_target_h_s2_ptr` is
 trainer-internal scratch state cleared by `.take()` on every IQN call,
 not an ISV slot.
+
+## Fix 35 — SP7 controller CQL target_ratio direction flip (2026-05-03)
+
+**Symptom:** T10-v3 train-multi-seed-x7sl2 (commit 9b5296b2f, IQN Fix 34
+applied): IQN warnings disappeared (Fix 34 working); BUT post-fix the model
+flat-collapsed from epoch 2 onward. Per-epoch HEALTH_DIAG showed:
+  - cql_budget_per_branch saturated to MAX_BUDGET=1.0 on every branch
+  - val_active_frac = 0.0 (no trades)
+  - val_dir_entropy = 0.0
+  - val_sharpe = 0.0
+  - q_var_per_branch[mag] = 0.0003 (mag head not differentiating)
+  - cql_raw_mag = 0.5179 (~50× pre-fix value of 0.0157)
+
+**Root cause:** `loss_balance_controller_kernel.cu` line 23 spec / line ~146
+implementation had the CQL formula:
+  `target_ratio[CQL][b] = ANCHOR_CQL_RATIO * (1 - flatness[b])`
+
+In this codebase, `flatness` = `var_q[b] / σ²` per
+`pearl_2_budget_kernel.cu`. So flatness HIGH means "Q has variance" and
+flatness LOW means "Q is flat." With the `(1 - flatness)` factor, the
+formula pushed CQL HIGHER when Q was flat — exactly when there is no
+overconfidence to penalise. The result was a death spiral on the magnitude
+head:
+
+  flat Q → CQL boosted → CQL pull keeps Q close to mean → Q stays flat
+  → controller sees flat Q → boosts CQL further → ad infinitum
+
+Pre-fix the spiral was masked because IQN-mag/ord/urg silently no-op'd
+(Fix 34) — no real Q-targets meant no overconfidence to penalise meant
+small cql_raw, and the controller never engaged on those branches. Once
+Fix 34 woke IQN-mag/ord/urg, real Q-targets produced large cql_raw, the
+controller engaged with the inverted formula, and the spiral fired
+immediately.
+
+**Fix:** flip the formula direction so CQL target scales WITH flatness:
+  `target_ratio[CQL][b] = ANCHOR_CQL_RATIO * flatness[b]`
+
+When Q has variance (flatness ≈ 1) overconfidence risk is real → grow CQL.
+When Q is flat (flatness ≈ 0) there is nothing to penalise → shrink CQL.
+This aligns CQL's controller direction with what CQL semantically does
+(penalise overconfident Q estimates).
+
+**C51 unchanged:** the C51 formula `flatness * ANCHOR_C51_RATIO` was already
+correctly aligned (distributional learning is informative when the
+distribution has structure to learn from).
+
+**Per `pearl_controller_anchors_isv_driven.md`:** the deeper lesson is
+that controller anchors silently encode the regime in which they were
+tuned. The pre-Fix-34 IQN-dead regime hid the formula's incorrect sign;
+fixing the regime exposed it. ANCHOR_CQL_RATIO=2.0 and MAX_BUDGET=1.0
+remain hardcoded constants — these are addressed in Fix 36 (ISV-driven
+adaptive cap based on GPU-computed train_active_frac canary).
+
+Files: `cuda_pipeline/loss_balance_controller_kernel.cu` (+15 LOC
+comments, 1 line behavior change), this audit entry.