fix(reward): bound bonus shaping by ISV[Q_DIR_ABS_REF] to break optimism loop

C.4 timing bonus (experience_kernels.cu) and D.4b regime penalty
multiplied by the trade-cumulative |reward| / |final_pnl| as an
unbounded multiplicand. Across trades this compounds — bonus values
inflate as Q values inflate, then bonus rewards inflate Q values
further (Bellman bootstraps off shaped reward).

Per pearl_one_unbounded_signal_per_reward + feedback_isv_for_adaptive_bounds:
replace |reward|/|final_pnl| with ISV[Q_DIR_ABS_REF_INDEX]-bounded
variant. The unbounded multiplicand becomes the direction-branch
Q-scale EMA (already-tracked, gradient-decoupled), not the
trade-cumulative shaping output. Cap is adaptive (matches Q magnitude
as it evolves) and breaks the multi-trade compounding loop.

Discovered during Plan C Phase 2 smoke diagnosis (researcher report
2026-04-29 a25f669e9df953174). a52d99613 baseline also hits
bonus=235 — pre-existing structural pathology, not Plan-C-specific.

Other shaping sites (D.4a persist, B.2 novelty, D.4c stable) already
have all factors bounded — no fix needed.

Surgical change set:
  - experience_env_step gains final `int q_dir_abs_ref_idx` param.
  - C.4 (line 2521): pnl_unit = |final_pnl|/max(ISV[21], eps);
                     pnl_capped = min(pnl_unit, 1) * ISV[21].
  - D.4b (line 2581): same pattern, reward_unit/reward_capped.
  - gpu_experience_collector.rs:3800 wires Q_DIR_ABS_REF_INDEX as i32.

No new ISV slot, no new producer kernel, no layout-fingerprint shift —
ISV[21] already produced by q_stats_kernel.cu since Plan 1.

Predicted impact:
  - bonus EMA drops O(100-256) -> O(0.05-2.0)
  - rc[5] -> Bellman-target optimism loop broken
  - Plan C smoke: Q-drift kill at F0 ep2 likely no longer fires
  - a52d99613 baseline: same effect; validates pre-existing fix

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

crates/ml/src/cuda_pipeline/experience_kernels.cu

@@ -1696,7 +1696,19 @@ extern "C" __global__ void experience_env_step(
      * Kernel reads isv_signals_ptr[trade_attempt_rate_ema_idx] and
      * isv_signals_ptr[trade_target_rate_idx] at Flat→Positioned transitions. */
     int trade_attempt_rate_ema_idx,
-    int trade_target_rate_idx
+    int trade_target_rate_idx,
+    /* C.4 / D.4b bonus optimism break (2026-04-29): direction-branch
+     * Q-scale EMA slot index (Q_DIR_ABS_REF_INDEX = 21). The C.4 timing
+     * bonus and D.4b regime penalty multiply by |final_pnl| / |reward|
+     * which is itself the trade-cumulative shaped reward — across trades
+     * this compounds (bonus inflates Q which inflates next-trade reward
+     * which inflates bonus, …). Replacing the unbounded multiplicand
+     * with a Q-scale-bounded variant breaks the loop. ISV[21] is a
+     * gradient-decoupled EMA of the direction-branch max(|Q_mean|),
+     * already produced by q_stats_kernel.cu — no new producer needed.
+     * Per pearl_one_unbounded_signal_per_reward.md +
+     * feedback_isv_for_adaptive_bounds.md. */
+    int q_dir_abs_ref_idx
 ) {
     int i = blockIdx.x * blockDim.x + threadIdx.x;
     if (i >= N) return;
@@ -2518,9 +2530,23 @@ extern "C" __global__ void experience_env_step(
             const float peak_bar   = ps[PS_PEAK_PNL_BAR];
             const float bars_early = fmaxf(0.0f, segment_hold_time - peak_bar);
             const float final_pnl  = reward;
+            /* pearl_one_unbounded_signal_per_reward + feedback_isv_for_adaptive_bounds
+             * (2026-04-29): |final_pnl| compounds across trades through Layer 4-9
+             * credits, creating a multi-trade optimism loop (bonus → Q → reward →
+             * bonus). Bound it to the direction-branch Q-scale EMA
+             * (ISV[Q_DIR_ABS_REF_INDEX=21]) so the term's unbounded multiplicand
+             * is the stable, gradient-decoupled Q-scale, not the trade-cumulative
+             * shaped reward. ISV[21] is already produced by q_stats_kernel.cu.
+             * NOTE: ISV slot index passed as kernel arg (no magic numbers per
+             * feedback_no_quickfixes / feedback_isv_for_adaptive_bounds). */
+            const float pnl_norm   = (isv_signals_ptr != NULL)
+                                   ? isv_signals_ptr[q_dir_abs_ref_idx]
+                                   : 0.0f;
+            const float pnl_unit   = fabsf(final_pnl) / fmaxf(pnl_norm, 1e-6f);
+            const float pnl_capped = fminf(pnl_unit, 1.0f) * pnl_norm; /* ≤ Q-scale */
             const float timing_bonus = shaping_scale
                                      * (bars_early / segment_hold_time)
-                                     * fabsf(final_pnl)
+                                     * pnl_capped
                                      * conviction_core;
             reward += timing_bonus;
             /* C.2 attribution — accumulate into rc[5] bonus slot (see above). */
@@ -2559,12 +2585,27 @@ extern "C" __global__ void experience_env_step(
          * absolute Q-magnitude) AND |reward| simultaneously — produced
          * penalties 5–50× larger than the reward, flipping trade outcomes
          * wholesale and destabilising training (smoke showed Return swings
-         * −400% / +170% across epochs). Fix: pick exactly ONE unbounded
-         * scaling signal per reward term — here, |reward| itself.
+         * −400% / +170% across epochs). The follow-on fix used |reward| as
+         * the single unbounded multiplicand — but |reward| at this point IS
+         * the trade-cumulative shaped reward (Layer 2 + Layer 4-9 credits +
+         * C.4 timing bonus + D.4a persistence bonus accumulated above). Across
+         * trades this compounds: penalty inflates with reward inflation,
+         * which the Bellman target then bootstraps off, which inflates the
+         * NEXT trade's reward, etc. (Plan C Phase 2 smoke diagnosis 2026-04-29:
+         * bonus EMA hit 235 by F2 ep2 in the a52d99613 baseline as well —
+         * pre-existing structural pathology.)
+         *
+         * Final fix: bound |reward| by ISV[Q_DIR_ABS_REF_INDEX=21], the
+         * direction-branch Q-scale EMA. The unbounded multiplicand becomes
+         * the stable, gradient-decoupled Q-scale rather than trade-cumulative
+         * shaping output. Same surgical pattern as C.4 above — pearl
+         * pearl_one_unbounded_signal_per_reward.md +
+         * feedback_isv_for_adaptive_bounds.md.
          *
          * bars_late ∈ [0, hold_time]. When shift fires late in the trade, penalty
          * is small; when shift fires early and we held until exit, penalty is
-         * up to |reward| — flipping a positive reward to zero in the worst case.
+         * up to ISV[21] — bounded by the Q-scale instead of the (compounding)
+         * trade-cumulative reward.
          *
          * Attribution: subtracts from rc[5] via -= (co-exists with the three positive
          * contributions B.2/C.4/D.4a at other (i,t) slots; cancellation within the
@@ -2572,8 +2613,13 @@ extern "C" __global__ void experience_env_step(
         if (segment_complete && ps[PS_REGIME_SHIFT_BAR] > 0.0f && saved_hold_time > 0.0f) {
             const float bars_late_frac = fmaxf(0.0f,
                 fminf(1.0f, ((float)saved_hold_time - ps[PS_REGIME_SHIFT_BAR]) / (float)saved_hold_time));
+            const float reward_norm   = (isv_signals_ptr != NULL)
+                                      ? isv_signals_ptr[q_dir_abs_ref_idx]
+                                      : 0.0f;
+            const float reward_unit   = fabsf(reward) / fmaxf(reward_norm, 1e-6f);
+            const float reward_capped = fminf(reward_unit, 1.0f) * reward_norm;
             const float penalty = shaping_scale * conviction_core
-                                * bars_late_frac * fabsf(reward);
+                                * bars_late_frac * reward_capped;
             reward                                         -= penalty;
             reward_components_per_sample[out_off * 6 + 5]  -= penalty;
         }

crates/ml/src/cuda_pipeline/gpu_experience_collector.rs

@@ -3797,6 +3797,18 @@ impl GpuExperienceCollector {
                         use crate::cuda_pipeline::gpu_dqn_trainer::TRADE_TARGET_RATE_INDEX;
                         TRADE_TARGET_RATE_INDEX as i32
                     })
+                    // C.4 / D.4b bonus optimism break (2026-04-29):
+                    // direction-branch Q-scale EMA slot index. The C.4 timing
+                    // bonus and D.4b regime penalty bound their |reward|/|pnl|
+                    // multiplicands by ISV[Q_DIR_ABS_REF_INDEX=21] to break
+                    // the cross-trade compounding optimism loop. ISV[21] is
+                    // produced by q_stats_kernel.cu — no new producer needed.
+                    // Per pearl_one_unbounded_signal_per_reward.md +
+                    // feedback_isv_for_adaptive_bounds.md.
+                    .arg(&{
+                        use crate::cuda_pipeline::gpu_dqn_trainer::Q_DIR_ABS_REF_INDEX;
+                        Q_DIR_ABS_REF_INDEX as i32
+                    })
                     .launch(launch_cfg)
                     .map_err(|e| MLError::ModelError(format!(
                         "experience_env_step t={t}: {e}"

docs/dqn-wire-up-audit.md

@@ -2192,3 +2192,5 @@ Predicted impact: Plan C fold 0 ep2 with `q_mean(t)=0.82, q_mean(t-1)=-0.018, IS
 Per `pearl_adaptive_moe_lambda.md` — canonical "EMA-tracked diagnostic drives a controller" pattern: kernel + ISV slot + bootstrap (0.0 cold-start) + reset (0.0 fold boundary) + observability. Per `pearl_cold_path_no_exception_to_gpu_drives.md` — even cold-path scalar arithmetic stays on GPU when its reduction inputs already live on GPU (ISV[16,21] both produced by `q_stats_kernel.cu`). Per `feedback_adaptive_not_tuned.md` — tau decay is ISV-driven, not a constant. Per `feedback_isv_for_adaptive_bounds.md` — the `4.0` upper clip and `1e-6` denom floor are structural bounds (drift saturation matching the kill-criterion's 3.0× ratio threshold; numerical-stability floor) carried in ISV slot semantics.
 
 Plan C A.2 wire-up (2026-04-29, second commit): `tau_update_kernel.cu`, `state_reset_registry.rs`, `training_loop.rs::reset_named_state` + per-epoch launch site, and `trainer/mod.rs::reset_for_fold` comment all updated in lockstep with the kernel+slot landed by the previous commit. Producer launch lives between the existing Q-drift kill criterion and the `prev_epoch_q_mean = q_mean` update — both prev and curr q_means are in scope, and the `prev_epoch_q_mean.abs() > 1e-6` cold-start gate matches the kill check's gate. Tau consumer applies `1/(1+clip(ISV[129],0,4))` AFTER the existing `[tau_final, tau_base]` clamp so dampening can drive tau below tau_final on runaway drift (intent: tau_final is the design floor for healthy runs, but a runaway requires further suppression). Defensive `fminf(...,4.0f)` in the consumer guards against producer-side bugs even though the producer already clips (Invariant 1 carve-out, belt-and-braces).
+
+Plan C Phase 2 follow-up (2026-04-29): bonus reward optimism-coupling break in `experience_kernels.cu`. The C.4 timing bonus (~line 2521) and D.4b regime penalty (~line 2581) both used the trade-cumulative `|final_pnl|` / `|reward|` as their unbounded multiplicand. By the time bonus shaping runs in `experience_env_step`, `reward` is the Layer 2 vol-normalised return PLUS Layer 4-9 credits PLUS the C.4 timing bonus PLUS the D.4a persistence bonus — i.e. it is itself a function of all prior shaping inflation. Across trades this compounds into a multi-trade optimism feedback loop: bonus inflates → Q-target inflates (Bellman bootstraps off shaped reward) → next trade's reward inflates → bonus inflates further. Researcher reproduction (a25f669e9df953174, 2026-04-29) found the a52d99613 baseline ALSO hits bonus EMA=235 by F2 ep2 — the pathology is structural and pre-dates Plan C Phase 2. Per `pearl_one_unbounded_signal_per_reward.md` the rule is exactly ONE unbounded multiplicand per reward term, and per `feedback_isv_for_adaptive_bounds.md` adaptive bounds live in the ISV signal bus. **Surgical fix**: replace `|final_pnl|` (C.4) and `|reward|` (D.4b) with `min(|x|/ISV[Q_DIR_ABS_REF_INDEX=21], 1) × ISV[21]` — the multiplicand is now the direction-branch Q-scale EMA (already produced gradient-decoupled by `q_stats_kernel.cu` since Plan 1) rather than the trade-cumulative shaping output. Cap is adaptive (matches Q magnitude as it evolves) and breaks cross-trade compounding; |reward| can still drive the cap to its max (= ISV[21]) so directional information is preserved, just bounded. **Kernel signature**: `experience_env_step` gains a final `int q_dir_abs_ref_idx` parameter (after the existing `trade_target_rate_idx`); single Rust call site `gpu_experience_collector.rs:3800` passes `Q_DIR_ABS_REF_INDEX as i32`. `experience_action_select` not affected (it never read `|reward|`/`|final_pnl|` as a multiplicand). **Other shaping sites NOT touched** because they were already bounded — D.4a persist (multiplicand `drawdown_depth ≤ ~0.1` gates the term, `tanh(reward/dd)` is bounded), B.2 novelty (vol_proxy ≤ 0.01 gates, kl_amp ≤ 1.0), D.4c stable (vol_proxy gates + bounded stability/conviction). Per `feedback_no_partial_refactor.md` consistency: both fixed sites use the SAME ISV[21] bound and the SAME `(unit, capped)` two-step formula — symmetric migration. Predicted impact: bonus EMA O(100-256) → O(0.05-2.0); rc[5] → Bellman-target optimism loop broken; Plan C smoke F0 ep2 Q-drift kill should no longer fire on this pathway; a52d99613 baseline gets the same fix automatically (validates pre-existing pathology, not Plan-C-specific). Layout fingerprint unchanged — kernel signature change but no ISV slot add and no param-tensor change.