Daily ModelOutput

What the model flagged in its latest run, published unfiltered as part of the research record. Quiet days are shown as quiet days; nothing here is a recommendation to trade.

How to read a signal

Arm 1: Novelty
How far the stock’s current fused embedding sits from its own historical baseline (0–1). High novelty means “something about this stock’s combined data looks unusual today” — it says nothing about direction on its own.
Arm 2: Prob
The supervised transformer’s estimated probability for the forecast direction. These probabilities are experimental and uncalibrated — 0.6 should not be read as “60% of such calls come true” until calibration is measured and published.
Concordance
Where the two arms agree. concordant = both point the same way; discordant = they disagree; novelty_only = only Arm 1 fired. Agreement is context, not proof.
Overall strength
The model score used for the listing threshold. Only signals at or above 0.5 appear here; weaker output is visible on each stock’s detail page. The threshold is a display choice, not a validated trading rule.

Full concept definitions are in the Guide; the pipeline that produces these numbers is diagrammed on the Technology page.

Academic research platform — not investment advice.