jgrusewski 5f2e8930d3 feat: trade lifecycle fix — 3-layer hold enforcement (action masking + kernel override + reward scaling) ...

4 bugs fixed:
1. Hold guard was dead code (!exiting_trade always false) — replaced with unified guard
2. Reversals bypassed hold — new guard covers exits AND reversals
3. Single-bar trades got full reward — reward scaled by hold_time/min_hold_bars
4. No action masking — experience_action_select now forces current exposure during hold

3 layers of defense:
- Layer 1: Action masking in experience_action_select (masks Q-values + blocks random exploration)
- Layer 2: Kernel override in experience_env_step (catches epsilon exploration slipthrough)
- Layer 3: Reward scaling (smooth gradient: 10% floor to 100% at min_hold_bars)

Config: min_hold_bars added to DQNHyperparameters (default 5), TOML, hyperopt [2,20].
Trailing stop respects min_hold_bars (was hardcoded > 2.0f).
Search space: 46D (was 45D).

Also fixes flaky test_batch_hierarchical_vs_flat_softmax_diversity assertion.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

2026-03-25 23:32:15 +01:00

..

asset_selection

…

backtesting

…

benchmark

refactor: remove 8 always-on use_ booleans — features are mandatory

2026-03-25 09:45:54 +01:00

checkpoint

…

cuda_common

…

cuda_pipeline

feat: action masking in experience_action_select — Layer 1 hold enforcement

2026-03-25 23:17:36 +01:00

data_loaders

…

data_pipeline

fix(ci): mark 8 heavy data-loading tests #[ignore] for CI

2026-03-19 13:38:49 +01:00

data_validation

…

deployment

…

diffusion

fix: wire real GPU backprop + SPSA gradients, fix checkpoint loading, eliminate candle from examples

2026-03-20 15:41:42 +01:00

dqn

…

ensemble

fix: update smoke tests for 9-action branching DQN (was 5-action)

2026-03-25 21:37:09 +01:00

evaluation

…

explainability

…

features

…

flash_attention

…

gpu_benchmarks

…

hyperopt

feat: min_hold_bars in hyperopt search space [2, 20]

2026-03-25 23:26:42 +01:00

integration

…

kan

fix: wire real GPU backprop + SPSA gradients, fix checkpoint loading, eliminate candle from examples

2026-03-20 15:41:42 +01:00

labeling

…

liquid

fix: wire real GPU backprop + SPSA gradients, fix checkpoint loading, eliminate candle from examples

2026-03-20 15:41:42 +01:00

mamba

fix: wire real GPU backprop + SPSA gradients, fix checkpoint loading, eliminate candle from examples

2026-03-20 15:41:42 +01:00

microstructure

…

model_registry

…

observability

…

paper_trading

…

ppo

…

regime

…

regime_detection

…

registry

…

risk

…

security

…

stress_testing

…

tests

…

tft

fix: resolve all CI test failures — TFT, DQN pipeline, benchmarks

2026-03-21 09:09:51 +01:00

tgnn

fix: wire real GPU backprop + SPSA gradients, fix checkpoint loading, eliminate candle from examples

2026-03-20 15:41:42 +01:00

tlob

fix: wire real GPU backprop + SPSA gradients, fix checkpoint loading, eliminate candle from examples

2026-03-20 15:41:42 +01:00

trainers

feat: trade lifecycle fix — 3-layer hold enforcement (action masking + kernel override + reward scaling)

2026-03-25 23:32:15 +01:00

training

…

transformers

…

universe

…

validation

refactor: remove 8 always-on use_ booleans — features are mandatory

2026-03-25 09:45:54 +01:00

xlstm

fix: wire real GPU backprop + SPSA gradients, fix checkpoint loading, eliminate candle from examples

2026-03-20 15:41:42 +01:00

.gitignore

…

batch_processing.rs

…

benchmarks.rs

…

bridge.rs

…

data_loader.rs

fix: data_loader finds futures-baseline + mamba2 benchmark #[ignore]

2026-03-20 11:38:52 +01:00

examples.rs

…

feature_cache.rs

fix: include MBP-10 + trades dirs in feature cache key

2026-03-22 23:31:05 +01:00

inference_validator.rs

fix: wire real GPU backprop + SPSA gradients, fix checkpoint loading, eliminate candle from examples

2026-03-20 15:41:42 +01:00

inference.rs

chore: remove last 4 candle references from ml crate (comments only)

2026-03-20 01:05:44 +01:00

lib.rs

fix: update evaluate_baseline for new backtest evaluator API

2026-03-22 15:16:11 +01:00

model_factory.rs

…

model_loader_integration.rs

…

model_registry.rs

…

portfolio_transformer.rs

…

preprocessing.rs

…

training_pipeline.rs

…

training_profile.rs

feat: add min_hold_bars config (default 5) — prevents per-bar churning

2026-03-25 23:06:28 +01:00

training.rs

…

walk_forward.rs

feat: add curriculum learning with ADX-based difficulty scoring to walk-forward windows

2026-03-23 12:54:01 +01:00