LEARN. TEST. EARN TRUST.

Better decisions begin
with better evidence.

Follow the work behind each Bee. Training creates a candidate; independent evaluation decides whether it deserves a forward paper trial.

RESEARCH → VALIDATION → PAPER → LIVE ACCEPTANCE

Recorded daily training

04:00 · Africa/Lagos

Latest retained reports from the runtime. These are development screens and do not authorize execution. Reports generated before the current trainer update retain their original evaluation method.

Loading recorded evidence…

The strategy tournament

OFFLINE WORKER · AVAILABLE

Validate the data

One instrument, closed bars, consistent intervals, source and rights statement. Record the dataset hash.

Compare five candidates

Drift, Rebound, Breaker and two logistic thresholds. Train on the first 60%; select on the next 20% with purged boundaries.

Open the final holdout

Evaluate only the selected candidate on the last 20%. Compare 20, 40 and 80 basis-point round-trip costs and buy-and-hold.

Freeze and observe

Preserve the immutable report. A consumed holdout cannot become fresh evidence through another run. Forward paper validation remains required.

The first actual-market tournament is published in the Paper Workspace. Its selected candidate lost money in the final holdout at every tested cost stress and remains unapproved. Synthetic tests separately verify code behavior.

The research ecosystem

18 REFERENCES REVIEWED · 17 OCT 2026

Source and license review informs original OmieBot work. This catalogue identifies references, not installed integrations. CCXT packaged base modules also received a scoped REA static analysis; broker behavior remains separately verified.