jgrusewski 88abf8185c feat(rl): adaptive gate threshold controller from trade frequency ...

New rl_gate_threshold_controller.cu watches dones EMA and adjusts
confidence + FRD gate thresholds via Schulman bounded step:
- dones_ema < target×0.5 → relax thresholds (×0.95)
- dones_ema > target×2.0 → tighten thresholds (÷0.95)

ISV-driven: target (0.02), conf bounds (0.001-0.50), FRD bounds
(0.05-0.50), adjust rate (0.95). Runs per step after warmup.

Also lowers default thresholds: conf 0.10→0.01, FRD 0.35→0.15.
Post-warmup, the controller adapts these based on actual trade
flow instead of relying on static defaults.

ISV slots 525-531. RL_SLOTS_END → 532.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

2026-05-25 10:26:10 +02:00

..

abs_copy.cu

feat(rl): R7a — lift remaining host work in step_with_lobsim to GPU

2026-05-23 11:30:14 +02:00

actions_to_market_targets.cu

feat(rl): per-batch outcome EMA, vol-adjusted trail, ISV-driven P_MIN

2026-05-24 21:59:34 +02:00

adamw_step.cu

perf(ml-alpha): device-resident AdamW step counter (capture prep)

2026-05-17 14:39:38 +02:00

apply_reward_scale.cu

feat(rl): adaptive RATIO + EWMA V_MIN/V_MAX + λ_distill bump

2026-05-24 14:20:05 +02:00

argmax_expected_q.cu

feat(rl): SP20 P4 — N_ACTIONS 9→11 with HalfFlat actions

2026-05-24 17:17:33 +02:00

attention_pool.cu

refactor(ml-alpha): revert axes B/C/D/E — keep only Kendall σ (axis A)

2026-05-18 18:11:05 +02:00

aux_heads.cu

feat(aux-heads): rewrite for A+B paired (BCE + conditional-Huber) (CB3+CB4)

2026-05-22 11:16:29 +02:00

aux_loss.cu

feat(aux-heads): rewrite for A+B paired (BCE + conditional-Huber) (CB3+CB4)

2026-05-22 11:16:29 +02:00

aux_trunk.cu

feat(aux-supervision): wire AuxTrunk + AuxHeads + Huber loss into PerceptionTrainer (B5)

2026-05-22 09:45:29 +02:00

aux_vec_add.cu

feat(aux-supervision): wire AuxTrunk + AuxHeads + Huber loss into PerceptionTrainer (B5)

2026-05-22 09:45:29 +02:00

bce_loss_multi_horizon.cu

refactor(per-horizon): N_HORIZONS 5→3 — horizon_lambda + bce_loss + smoothness kernels

2026-05-22 01:14:09 +02:00

bellman_target_projection.cu

feat(rl): SP20 P4 — N_ACTIONS 9→11 with HalfFlat actions

2026-05-24 17:17:33 +02:00

bucket_transition_kernels.cu

refactor(per-horizon): N_HORIZONS 5→3 — bucket-coupled CUDA kernels

2026-05-22 01:26:37 +02:00

cfc_step_per_branch.cu

refactor(per-horizon): N_HORIZONS 5→3 — bucket-coupled CUDA kernels

2026-05-22 01:26:37 +02:00

cfc_step.cu

perf(ml-alpha): NVIDIA-grade rewrite of CfC K-loop hot kernels — 2.15× faster

2026-05-18 19:07:29 +02:00

compute_advantage_return.cu

feat(rl): R3 — GPU-resident EMA + advantage/return kernels

2026-05-23 09:56:28 +02:00

dqn_distributional_q.cu

feat(rl): SP20 P4 — N_ACTIONS 9→11 with HalfFlat actions

2026-05-24 17:17:33 +02:00

dqn_target_soft_update.cu

feat(rl): R5 — wire 7 controllers per-step + target-net soft update

2026-05-23 10:19:13 +02:00

ema_update_on_done.cu

feat(rl): R3 — GPU-resident EMA + advantage/return kernels

2026-05-23 09:56:28 +02:00

ema_update_per_step.cu

feat(rl): R3 — GPU-resident EMA + advantage/return kernels

2026-05-23 09:56:28 +02:00

extract_realized_pnl_delta.cu

feat(rl): R6 — replace LobEnv with kernel-driven GPU-pure env step

2026-05-23 10:39:25 +02:00

gpu_log_helpers.cuh

feat(gpu-log): wire smoothness_lambda_controller as first ring producer

2026-05-21 01:36:18 +02:00

gpu_log_ids.h

feat(gpu-log): add gpu_log_ids.h — single source of truth for ring IDs

2026-05-21 01:04:50 +02:00

gpu_log_ring.cu

feat(gpu-log): wire smoothness_lambda_controller as first ring producer

2026-05-21 01:36:18 +02:00

grad_h_accumulate.cu

feat(rl): real Q/π/V forward+backward + per-head Adam (Phase E.2)

2026-05-22 23:25:42 +02:00

grad_norm.cu

perf(ml-alpha): eliminate all dtoh from training hot path

2026-05-17 13:57:02 +02:00

heads_block_diagonal_fwd.cu

refactor(per-horizon): N_HORIZONS 5→3 — bucket-coupled CUDA kernels

2026-05-22 01:26:37 +02:00

horizon_lambda.cu

refactor(per-horizon): N_HORIZONS 5→3 — horizon_lambda + bce_loss + smoothness kernels

2026-05-22 01:14:09 +02:00

layer_norm.cu

feat(ml-alpha): LayerNorm backward + per-row param-grad reducer kernels

2026-05-17 21:22:44 +02:00

log_pi_at_action.cu

feat(rl): SP20 P4 — N_ACTIONS 9→11 with HalfFlat actions

2026-05-24 17:17:33 +02:00

mamba2_alpha_kernel.cu

arch(crt-a): forward_step incremental SSM state — enables event-rate trunk forward

2026-05-20 18:05:44 +02:00

multi_horizon_heads.cu

refactor(per-horizon): N_HORIZONS 5→3 — bucket-coupled CUDA kernels

2026-05-22 01:26:37 +02:00

output_smoothness.cu

refactor(per-horizon): N_HORIZONS 5→3 — bucket-coupled CUDA kernels

2026-05-22 01:26:37 +02:00

ppo_clipped_surrogate.cu

feat(rl): SP20 P4 — N_ACTIONS 9→11 with HalfFlat actions

2026-05-24 17:17:33 +02:00

ppo_log_ratio_abs_max_b.cu

feat(rl): ISV-driven PPO importance-ratio clamp + log-ratio diagnostic

2026-05-23 20:55:00 +02:00

projection.cu

feat(ml-alpha): projection kernel (128->8 + layer-norm)

2026-05-16 21:52:57 +02:00

reduce_axis0.cu

perf(ml-alpha): NVIDIA-grade rewrite of CfC K-loop hot kernels — 2.15× faster

2026-05-18 19:07:29 +02:00

rl_action_kernel.cu

feat(rl): SP20 P4 — N_ACTIONS 9→11 with HalfFlat actions

2026-05-24 17:17:33 +02:00

rl_atom_support_update.cu

feat(rl): adaptive C51 atom span ratchet to lift Q learning ceiling

2026-05-24 13:22:31 +02:00

rl_confidence_gate.cu

feat(rl): gate warmup — confidence + FRD gates inactive for first 10k steps

2026-05-25 10:16:05 +02:00

rl_encoder_context_broadcast.cu

feat(rl): encoder input expansion 40 → 56 dims (trade-context + multires)

2026-05-24 22:35:26 +02:00

rl_entropy_coef_controller.cu

feat(rl): SP20 P4 — N_ACTIONS 9→11 with HalfFlat actions

2026-05-24 17:17:33 +02:00

rl_frd_fwd.cu

feat(rl): FRD head forward pass + GPU-oracle tests (F.2)

2026-05-24 18:13:35 +02:00

rl_frd_gate.cu

feat(rl): gate warmup — confidence + FRD gates inactive for first 10k steps

2026-05-25 10:16:05 +02:00

rl_frd_layer1_bwd.cu

feat(rl): FRD layer-1 backward (dW1, db1, dh_t with ReLU mask) — F.3c

2026-05-24 18:40:30 +02:00

rl_frd_layer2_bwd.cu

feat(rl): FRD layer-2 backward (dW2, db2, dhidden) — F.3b

2026-05-24 18:36:29 +02:00

rl_frd_softmax_ce_grad.cu

feat(rl): FRD softmax + CE + dL/dlogits backward stage 1 (F.3a)

2026-05-24 18:31:03 +02:00

rl_gamma_controller.cu

fix(rl): apply derive-from-input bootstrap to γ + coef controllers

2026-05-23 13:56:40 +02:00

rl_gate_threshold_controller.cu

feat(rl): adaptive gate threshold controller from trade frequency

2026-05-25 10:26:10 +02:00

rl_isv_write.cu

fix(rl): ISV-ify ALL remaining tunable design constants (10 new slots)

2026-05-24 01:10:53 +02:00

rl_kl_approx_b.cu

feat(rl): wire remaining 3 EMA inputs (kl_pi, q_divergence, trade_duration)

2026-05-23 16:37:34 +02:00

rl_kurtosis_streaming.cu

audit: ISV-ify 10 more design constants — Schulman + bootstraps + streaming α

2026-05-24 01:47:18 +02:00

rl_l2_diff_norm.cu

feat(rl): wire remaining 3 EMA inputs (kl_pi, q_divergence, trade_duration)

2026-05-23 16:37:34 +02:00

rl_l2_norm.cu

feat(rl): signal-driven LR controller from per-head grad-norm EMAs

2026-05-23 18:11:58 +02:00

rl_lr_controller.cu

fix(rl): ISV-ify ALL remaining tunable design constants (10 new slots)

2026-05-24 01:10:53 +02:00

rl_multires_features_update.cu

feat(rl): trade-context + multires features + P14 validation tests

2026-05-24 22:21:30 +02:00

rl_per_alpha_controller.cu

audit: ISV-ify 10 more design constants — Schulman + bootstraps + streaming α

2026-05-24 01:47:18 +02:00

rl_pi_action_kernel.cu

feat(rl): per-batch outcome EMA, vol-adjusted trail, ISV-driven P_MIN

2026-05-24 21:59:34 +02:00

rl_position_heat_check.cu

feat(rl): SP20 P6 position heat cap — force-flat on over-leverage

2026-05-24 20:36:05 +02:00

rl_ppo_clip_controller.cu

audit: ISV-ify 10 more design constants — Schulman + bootstraps + streaming α

2026-05-24 01:47:18 +02:00

rl_ppo_ratio_clamp_controller.cu

audit: ISV-ify 10 more design constants — Schulman + bootstraps + streaming α

2026-05-24 01:47:18 +02:00

rl_q_distill_lambda_controller.cu

feat(rl): adaptive λ_distill controller + reward_scale MIN ISV

2026-05-24 14:55:55 +02:00

rl_q_pi_agree_b.cu

feat(rl): SP20 P4 — N_ACTIONS 9→11 with HalfFlat actions

2026-05-24 17:17:33 +02:00

rl_q_pi_distill_grad.cu

feat(rl): SP20 P4 — N_ACTIONS 9→11 with HalfFlat actions

2026-05-24 17:17:33 +02:00

rl_recent_outcome_update.cu

feat(rl): per-batch outcome EMA, vol-adjusted trail, ISV-driven P_MIN

2026-05-24 21:59:34 +02:00

rl_reward_clamp_controller.cu

fix(rl): WIN/LOSS + C51 atom span are STRUCTURAL, not adaptive (G.2)

2026-05-24 19:56:16 +02:00

rl_reward_scale_controller.cu

feat(rl): adaptive λ_distill controller + reward_scale MIN ISV

2026-05-24 14:55:55 +02:00

rl_rollout_steps_controller.cu

audit: ISV-ify 10 more design constants — Schulman + bootstraps + streaming α

2026-05-24 01:47:18 +02:00

rl_step_counter_update.cu

feat(rl): wire remaining 3 EMA inputs (kl_pi, q_divergence, trade_duration)

2026-05-23 16:37:34 +02:00

rl_streaming_clamp_init.cu

fix(rl): ISV-driven K-loop divisor + max ceiling (slot 450, 451)

2026-05-24 00:23:38 +02:00

rl_target_tau_controller.cu

audit: ISV-ify 10 more design constants — Schulman + bootstraps + streaming α

2026-05-24 01:47:18 +02:00

rl_trade_context_update.cu

feat(rl): trade-context + multires features + P14 validation tests

2026-05-24 22:21:30 +02:00

rl_trail_mutate.cu

fix(rl): wire TrailTighten/TrailLoosen + SP20 P1+P5 foundation

2026-05-24 16:47:53 +02:00

rl_trail_stop_check.cu

feat(rl): pyramiding semantics — add/partial-flat/anti-martingale sizing

2026-05-24 21:15:04 +02:00

rl_unit_state_update.cu

feat(rl): per-batch outcome EMA, vol-adjusted trail, ISV-driven P_MIN

2026-05-24 21:59:34 +02:00

rl_var_over_abs_mean_streaming.cu

audit: ISV-ify 10 more design constants — Schulman + bootstraps + streaming α

2026-05-24 01:47:18 +02:00

smoothness_lambda_controller.cu

refactor(per-horizon): N_HORIZONS 5→3 — horizon_lambda + bce_loss + smoothness kernels

2026-05-22 01:14:09 +02:00

snap_feature_assemble.cu

feat(rl): encoder input expansion 40 → 56 dims (trade-context + multires)

2026-05-24 22:35:26 +02:00

v_head_fwd_bwd.cu

feat(rl): V_pred structurally clamped to C51 atom span (G.1)

2026-05-24 19:47:25 +02:00

variable_selection.cu

perf(ml-alpha): block-per-row VSN bwd refactor (Phase B commit 3)

2026-05-17 23:57:32 +02:00