segment_pnl depended on raw_pnl (now zero after raw_next removal).
The trade-closing P&L was missing the final bar's unrealized gain/loss.
Fix: compute unrealized_at_exit = position * (raw_close - entry_price)
directly. No future price needed — just current close vs entry.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
E1 enrichment was computing q_corr = mean(predicted_q - pnl) but
predicted_q was set to avg_pnl → bias always ~0. Now uses actual
avg_q_value from the training step, producing meaningful corrections
when Q-values drift away from realized returns.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
portfolio_sim_kernel also used next_close_raw (future price) for
mark-to-market and reward computation. Fixed: use current price only.
Updated stale comments referencing per-bar reward in trade-level section.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The experience collector used raw_next (NEXT bar's close) for P&L and
equity, creating a 1-bar misalignment between actions and rewards.
The model couldn't learn which actions produce which outcomes.
val_Sharpe=24 (backtest uses raw_close correctly) but training Sharpe=0
(experience collector used raw_next, shifting reward by 1 bar).
Fix: ALL portfolio computations use raw_close only. raw_next removed
from reward path entirely. Rewards are per-trade only (position change
= realized P&L, holding = tiny cost, flat = zero).
This is the root cause of training Sharpe being capped at ~0 across
ALL runs regardless of model complexity or component changes.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Post-validation enrichment: extract eval trades from WindowMetrics,
run all 8 enrichments, apply adaptive epsilon + gamma + agreement
threshold. Smoke test verifies enrichment runs at least once.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Risk branch weights trained via gentle decay toward initial values
(prevents R collapse to 0 or 1). Full BPTT backward deferred —
the risk branch learns its initial representation from trunk gradients
flowing through shared weights.
compute-sanitizer: 0 errors. Smoke test passes.
NUM_WEIGHT_TENSORS: 68. Total risk params: ~33K.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
c51_loss_kernel: reads cvar_alpha_buf[sample_id] when available (NULL = iqn_readiness fallback).
env_step: reads commit_lambda_buf[i] when available (NULL = 0.01 fallback).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
1. Zero-target normalization: scale=max(|target|,|observed|,1.0) instead of
max(|target|,1e-6). Prevents Q-mean (target=0) from producing infinite
error that steals entire budget from other observables.
2. Per-observable budget cap: each observable gets budget/N_OBS instead of
competing for a global pool. One runaway can't starve the others.
3. Readiness-driven alpha: alpha = 0.3*(1-readiness) + 0.01*readiness.
Model readiness drives target adaptation speed, not epoch number.
Exploring → fast targets. Converged → slow targets. Never frozen.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Per-bar reward (next_close - close) has SNR ~0.01 — 99% random walk noise.
Trade-level P&L has SNR ~0.1-0.5 — the atomic unit of trading signal.
Position change + had old trade → reward = realized_pnl (trade outcome)
Holding (no change) → reward = -0.0001 * |position| (holding cost)
Flat → reward = 0
C51 atoms now model distribution of TRADE OUTCOMES instead of
distribution of per-bar noise. 10-50× signal improvement.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Adds G16 homeostatic regularization that penalizes training observables
drifting from calibrated set-points. All 6 scalar signals use pinned
device-mapped memory (zero memcpy). Targets self-calibrate via EMA
during epochs 1-5, then freeze.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Q-mean drift: linear penalty (0.01 * q_mean) → quadratic
(0.01 * q_mean * |q_mean|). Small drift = tiny penalty, large
drift = hard correction. At q_mean=3.5: 12.25× stronger than linear.
Cost curriculum: linear ramp (epoch/20) → sigmoid centered at epoch 10.
Gradual start (find raw edges), steep middle (force cost adaptation),
gradual finish (fine-tune at real costs). Prevents strategy breakage
from sudden cost increases that caused training Sharpe oscillation.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
mamba2_scan_backward kernel reads d_h_enriched [B, SH2] but was passed
self.grad_buf [TOTAL_PARAMS] — wrong buffer, wrong size. At batch_size=4096
the kernel read 1M floats from a 582K buffer → 2749 OOB reads.
Fixed: use self.bw_d_h_s2 [B, SH2] which is the actual trunk activation
gradient from the cuBLAS backward pass.
Also increased smoke test batch_size to 4096 to catch scale-dependent OOB.
compute-sanitizer: 0 errors at batch_size=4096.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
CRITICAL: epsilon_buf memcpy_htod used full max_batch_size buffer but
eps_host was batch_size. Fixed: slice_mut(..batch_size) to match.
HIGH: update_q_mean_ema read pinned memory before GPU finished writing.
Moved after cuStreamSynchronize to ensure kernel completion.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The compute_expected_q and quantile_q_select kernels read per_sample_support[i*3+0/1/2]
(3 floats per sample), but the experience collector was passing eval_v_range_ptr which
is only 2 floats (v_min, v_max). Every sample after sample 0 read out of bounds.
Replace the u64 pointer field with a proper CudaSlice<f32> buffer [alloc_episodes, 3]
that is tiled with [v_min, v_max, delta_z] once per epoch via update_per_sample_support().
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Two tests:
- test_generalization_kernels_load: no data, verifies all CUDA cubins load
- test_generalization_components_smoke: 3 epochs on fxcache, verifies
AdamW, cost_anneal, gamma_anneal, DSR, walk-forward state, Q-gap,
atom utilization, all kernel launches succeed, finite metrics.
Passes in 2.7s on RTX 3050 with 5000 bars.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
HIGH: c51_loss_kernel now uses atom_positions per-branch in shmem_support
(was ignoring adaptive positions → forward/loss atom mismatch).
MEDIUM: adaptive_gamma wired into C51 Bellman projection via
set_adaptive_gamma(). Config gamma replaced with adaptive_gamma in
both launch_c51_loss sites.
MEDIUM: c51_grad z_norm uses adaptive atom positions when available
(was assuming linear grid for spread gradient).
LOW: adaptive_gamma field added to GpuDqnTrainer, initialized from
config, updated via passthrough from training loop.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
G5: Ensemble variance gates magnitude Q-values via sigmoid scale.
High disagreement forces conservative (Small) position. Pinned
var_ema threshold — no cuMemcpy in hot path.
G6: 6-way cosine similarity penalty on branch hidden activations
(lambda=0.01). Informational — gradient integration deferred.
G10: Lipschitz penalty on Q-diffs between consecutive similar states
(lambda=0.005, threshold=0.95). Uses atomicAdd accumulation.
G14: Confidence-weighted PER flagged for follow-up (needs on-GPU
priority modification to avoid memcpy).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
G2: Transaction costs anneal 0%→100% over 20 epochs via pinned
device-mapped cost_anneal_ptr. No memcpy — CPU writes, GPU reads.
G15: Commitment penalty lambda=0.01, tau=5.0 bars. Also scales
with cost_anneal to ramp together.
G11: Per-branch Q-anchoring to neutral actions (Flat/Small/Market/Normal).
Focuses model capacity on alpha signal over doing nothing.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
q_mean_scratch was CudaSlice with memcpy_dtoh readback every step.
Now pinned device-mapped: GPU writes via dev_ptr, CPU reads via
host_ptr directly. Dead inline EMA code removed, wired through
update_q_mean_ema(&self) method (DRY). No hot-path memcpy remains.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
AdamW weight_decay was in kernel but hardcoded to 0.0 in launch.
Now uses config value (1e-4) gated by per-param mask (1.0 for trunk+value
indices 0-7, 0.0 for branches). L1 proximal step on w_s1 (first layer)
induces automatic feature selection from 42-dim input.
Also updates decision_transformer.rs Adam launch to match the new
3-arg kernel signature (uniform mask, L1 disabled).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Reward rank normalization now operates on Sharpe contributions
(return - mean) / std instead of raw returns. Aligns reward signal
with Sharpe ratio goal. High-return trades during volatile periods
get lower rank weight than same return during calm periods.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
9d: Q-mean EMA (alpha=0.01) tracks epoch-level drift. Drift penalty
(lambda=0.01) in c51_grad pushes Q-distribution back toward zero.
9e: 500-step LR warmup for Mamba2 and other new components. Prevents
gradient shock from new modules destabilizing trained trunk.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Allocate rewards_5bar and rewards_20bar buffers [B] for future n-step
return computation. Wire multi_horizon_value_forward into training step
(currently degenerate d2d copy of 1-bar logits, ready for full GEMMs).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
atom_position_gradient kernel computes entropy-based gradient for
spacing_raw parameters. SGD decay toward uniform (lr=1e-3, every 50
steps) prevents atom positions from drifting. Concentrates atoms
where return distribution has mass.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2 extra value heads (W_v1/W_v2 pairs) for 5-bar and 20-bar horizons.
Regime-weighted blend: trend_weight=sigmoid((ADX-25)/5) mixes 1-bar
and 20-bar Q-values. Urgency branch learns temporal opportunity
structure. ~144K extra params. NUM_WEIGHT_TENSORS: 56 -> 64.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ensemble_size was a dead config field — ensemble_count drives the
actual ensemble head count in fused_training.rs. Changed from 1 to 3
to actually enable 3 independent Q-network heads.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ensemble_size: 5 -> 3. Each head shares the trunk but has independent
branch weights. Ensemble variance replaces C51 aleatoric variance
for exploration: Q_select = Q_mean + sqrt(ensemble_var).
~174K extra params from triplicated branch heads.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
W_regime[4,4] + b_regime[4] produce softmax importance weights from
[ADX, CUSUM, Q_gap, atom_utilization]. Scales per-branch Q-values
so direction matters more in trends, magnitude matters more in ranges.
NUM_WEIGHT_TENSORS: 50 -> 52.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Three new GPU kernels for OOS performance enhancement:
- regime_branch_gate: learned per-branch importance weights from regime features
- adaptive_atom_positions: softmax-normalized learnable C51 atom spacing
- mamba2_temporal_scan + mamba2_update_history: selective SSM for temporal context
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
train-w2z55 rewound to epoch 0 (improvement_rate=17.3, val_Sharpe=17.3)
instead of epoch 20 (improvement_rate=0.4, val_Sharpe=45.4). The model
lost ALL learned behavior and couldn't re-learn at low cosine LR.
Fix: sort checkpoints by val_Sharpe (highest first). Rewind targets the
BEST state (val_Sharpe=45), not the fastest-climbing moment (epoch 0).
Added val_sharpe field to TrajectoryCheckpoint for proper ranking.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Observed in train-w2z55: rewind to epoch 0 weights at epoch 30, but
cosine LR at t_local=10 (50% decay). Model couldn't re-learn at low LR.
val_Sharpe stuck at 17.39 for 4+ epochs after rewind.
Fix: cosine_epoch_offset set to current epoch on rewind.
t_local = (epoch - offset) % T_restart = 0 after rewind → full LR.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ROOT CAUSE: run_backtracking_epoch_end received epoch_sharpe (training
Sharpe from experience collection, ~0.65-0.79 oscillating) instead of
val_Sharpe (deterministic backtest, 33.49 frozen for 56 epochs).
Training Sharpe oscillates even when the model is frozen → sharpe_frozen
was always false → AND condition never met → backtracking never triggered
despite 56 consecutive frozen epochs in train-2tgs7.
Fixes:
1. Pass val_sharpe (-val_loss) to run_backtracking_epoch_end
2. prev_val_sharpe field on BacktrackingState (not sharpe_history)
3. improvement_rate uses val_Sharpe delta (not training Sharpe)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
1. Checkpoint save: improvement_rate used sharpe_history.last() which is
the CURRENT epoch (just pushed). Always compared current to itself → 0.
Fix: use sharpe_history[len-2] (previous epoch).
2. Branch Adam reset: required !checkpoints.is_empty() but branch reset
doesn't USE checkpoints — it just zeros momentum in-place. Removed
the checkpoint guard so the lightweight fix fires even without saved
checkpoints.
Both bugs prevented backtracking from ever acting in train-l7g9t despite
detecting 6 consecutive frozen epochs.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Stable Q-gap + rising val_Sharpe is healthy convergence, not a plateau.
OR logic would fire branch Adam reset at epoch ~12 when Q-gap stabilizes
at 0.318 while val_Sharpe climbs 17→40. AND requires BOTH Q-gap frozen
AND val_Sharpe stagnating — true plateau, nothing improving.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
states_buf has stride pad128(state_dim) for CUTLASS alignment. The
concat_ofi kernel was reading with stride state_dim → OOB for sample 1+.
With state_dim=50, pad=128: sample 1 OFI at offset 128+42=170, but
kernel read at 50+42=92 (sample 0 padding area). Silent data corruption.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
