A team built to improve.
One director. Accountable specialists. A continuous cycle of prediction, evaluation, and learning.
Connecting to your private workspace
Verifying owner access before loading private predictions.
End-of-day review has not loaded.
Results, lessons & follow-up →AI runtime
Inspecting readinessChecking actual execution state.
Automatic race coverage
Not enabledNo source run is recorded.
Leadership & mission
DIRECTORRacing Director
Improve the accuracy and calibration of AU and Japan horse-racing predictions.
Independent evaluation governs change. A completed task is not proof of a better prediction.
Operational attention
View monitoring →Current work
Manage queue →Actual AI work coverage
22 AI roles share one cloud AI worker. Queued roles take turns; an active role is not proof of completed work. Open a role or task for its latest accepted result and current wait. Open work queue. Task completion does not establish prediction accuracy.
Inspect recent redacted blocked assignments
Specialist responsibilities with explicit boundaries. Routine checks use code; AI handles reasoning and investigation.
Commission a task
DURABLE QUEUEAll tasks
Detect → assign → repair → independently verify → close. Duplicate symptoms share one incident; unresolved work stays visible.
Today's predictions
Win probabilities: not estimated by a trained model. AI selections below are qualitative.
“No model prediction” means a 1/N benchmark, with no estimate of real differences between horses. “Reviewed · no pick” means AI finished the review without a supported selection; it does not mean the race is cancelled. Queued and reviewing are unfinished work.
Coverage and unavailable forecasts
Performance and pre-race source reviews
Pre-race source reviews
T−30 and final checks re-read published fields and conditions. A planned check is not a completed review; gaps stay visible.
WIN5 research preparation
Saved formal preparation records
Published outcomes and saved forecast history
Published outcomes
A published result without explicit official clearance stays provisional. These observations are separate from verified prediction performance.
Full-field probabilities are saved before the cutoff. Uniform benchmarks and unfitted form priors have no demonstrated prediction advantage; model changes require evaluation.