Model Registry
Manage alpha model configurations and experiment results
Total Configs
0
Active Configs
0
Total Experiment Runs
0
Last Experiment
Never
Shadow Model Comparison
Candidate models running in "canary mode" alongside production
No shadow predictions yet. Run the shadow_score agent to generate comparisons.
Run shadow_score agent to populate shadow predictions
Alpha Models(0 configs)
No alpha model configs yet. Run the model_experiment agent to generate candidates.
API Reference
GET/api/admin/models?secret=ADMIN_SECRET— List all configs
GET/api/admin/models/compare?horizon=20— Shadow vs production comparison
POST/api/admin/models— Create new config
PATCH/api/admin/models/[id]— Promote/rollback (action: "promote" | "rollback")
# Promote a config
curl -X PATCH "/api/admin/models/2?secret=..." -d '{"action":"promote","notes":"Better IC20d"}'
curl -X PATCH "/api/admin/models/2?secret=..." -d '{"action":"promote","notes":"Better IC20d"}'
Interpretation Guide
- IC ≥ 0.1: Strong predictive signal (Spearman correlation)
- IC 0-0.1: Weak but positive correlation
- IC < 0: Inverse correlation (model predicting wrong direction)
- Spread: Top decile avg return - bottom decile avg return
- Combined Score: 70% IC + 30% Spread (normalized 0-100)
- Only promote configs with consistent positive IC across multiple runs
Model experiments run via model_experiment agent. Promotions require human review.