The current protocol.
A public, human-readable summary of the frozen ENTRY Confirmation study. This page summarizes the scientific protocol; it does not authorize Fresh execution, Historical TEST or deployment.
Require repeated directional intent before entry.
The study asks whether a small number of consecutive identical directional requests can increase entry selectivity and reduce execution drag without changing directional symmetry, costs, leverage, sizing, risk limits, reward coefficients, market features, PPO hyperparameters or the 30-minute decision cadence.
Exactly 1, 2 and 3 confirmations.
Pending intent is Markov-visible.
While FLAT
FLAT cancels pending confirmation. A new LONG/SHORT starts count 1. The same direction increments the count. Opposite direction replaces pending direction and resets count to 1. Until threshold is reached, effective action remains FLAT.
While positioned
Same-position HOLD is preserved. Voluntary FLAT exit is immediate. In treatment arms, an opposite-direction request first closes to FLAT and begins pending confirmation for the opposite direction. Forced risk close always has priority.
472 parent features + 2 treatment-state features.
| Index | Feature | Encoding |
|---|---|---|
| 0…471 | Frozen parent observation | Unchanged definition/order/normalization/provenance |
| 472 | pending_direction | SHORT −1.0 · NONE 0.0 · LONG +1.0 |
| 473 | pending_confirmation_progress | count / 3 → 0, 1/3 or 2/3 |
One treatment variable; parent stack stays fixed.
Ten checkpoints per arm, none selected from TRAIN performance.
Candidate outer timesteps: 3,334 · 6,668 · 10,002 · 13,336 · 16,670 20,004 · 23,338 · 26,672 · 30,006 · 33,340
84 days · 120,960 rows · 360 episodes.
As of 18 Sep 2026 the full source does not yet exist, so candidate performance remains sealed.
120,000 controlled action minutes.
Positive economics are mandatory.
0 risk-terminated windows; strictly positive median and mean actual net economic return; no permanently-FLAT zero-return behavior.
Median return → mean return → mean maximum drawdown → exact residual tie = AMBIGUOUS. No least-negative fallback.
Not available for this decision stage.
The next Historical TEST is preregistered as the next contiguous 120,960 one-minute rows after Fresh Validation, expected 2026-11-16 through 2027-02-07. It remains locked unless Fresh selection yields exactly one champion and an independent future TEST authorization chain is frozen.