Research validation

Alpha Scorecard

A research-only view of whether Innellis predictions are actually useful. Model ranking evidence, Committee efficacy and challenger experiments are kept separate so thousands of stock rows cannot masquerade as thousands of independent experiments.

Statistical evidence is counted by independent prediction date. Return spreads are percentage points. The scorecard does not promote models, alter the Committee or authorize live trading.

Established model

—

Best matured date count

—

10 required before evaluated status

Challenger status

—

Evidence integrity

—

1. Model Alpha

Established cross-sectional ranking model. Primary metrics are daily Spearman IC and top-minus-bottom decile SPY-excess spread.

Loading model evidence...

2. Committee Efficacy

Historical Committee recommendation outcomes by horizon. These results are distinct from the ML ranker's performance.

Loading Committee evidence...

3. Challenger Race

Prospective research challengers only. Stored evaluations that no longer match the current final outcome set are excluded.

Loading challenger evidence...

4. Evidence Health

Outcome maturation and immutable evidence-link coverage.

Loading evidence health...

Research attribution

Learning Quality

Insufficient evidence

Strict out-of-sample persona attribution. Observed direction is descriptive; formal classification requires enough independent folds. This section does not change Committee weights or authorize live mutation.