Autoresearch Pipeline Accuracy — ShurAI Advisory Board

69.9

69.9%

Optimized Accuracy

Baseline

23.5%

Optimized

69.9%

MOTPE/Optuna TPE — 30 optimization trials

"The optimizer proved the graph HAS signal — it just needed the right interface configuration."

First Out-of-Sample Validation

Predictions Evaluated

Methods Compared

Weeks of Data

Companies Tracked

Method Comparison — Out-of-Sample Accuracy

Method	Description	Score
Persistence	Assumes next week = this week. Near-zero accuracy confirms scores are dynamic, not static.	4.8%
Naive Momentum	Extrapolates recent trend direction. Captures some signal but lacks structural context.	28.6%
Mean Reversion Best Performer	Predicts regression toward historical mean. Highest single-week accuracy of all methods tested.	85.7%
KG Augmented Not Yet Deployed	Default config: 4.8%. Optimized config reaches 69.9% — exists but has not yet been deployed to production pipeline.	4.8%

Cumulative Accuracy Over 2 Weeks

38 predictions evaluated across the full validation window

68.4%

Mean Reversion — 68.4% cumulative

Mean Reversion sustains strong performance across multiple weeks. When the optimized KG Augmented config is deployed, the combined approach is expected to exceed this baseline.