Alpaca Expected Execution Cost Model
Generated 2026-07-19 17:20:43.477419+00:00 | version 20260719T172030Z
collecting_strictly_post_cutoff_paper_shadow
Promotion Decision
This immutable version is promoted to isolated paper shadow for expected-cost estimation only. It has no position-sizing or order authority, does not model tail slippage, and cannot enable live trading.
Temporal Validation
Training: 2026-04-30 through 2026-07-06; validation: 2026-07-07 through 2026-07-10; untouched test: 2026-07-13 through 2026-07-17.
Promotion Checks
| Check | Result |
|---|---|
| buy_adverse_selection_bps_bias | passed |
| buy_adverse_selection_bps_rmse_lift | passed |
| minimum_candidate_rows | passed |
| minimum_training_dates | passed |
| minimum_untouched_test_dates | passed |
| minimum_untouched_test_rows | passed |
| minimum_validation_dates | passed |
| sell_adverse_selection_bps_bias | passed |
| sell_adverse_selection_bps_rmse_lift | passed |
| strict_temporal_order | passed |
Known Boundary
The quoted half-spread remains better on median-oriented MAE because adverse selection is zero for many individual minutes. This model is promoted for expected-cost RMSE and bias calibration; tail-risk and per-order fill calibration remain separately blocked until prospective shadow depth and 100 actual paper-fill samples exist.