How the score works
Nothing here is proprietary. If you disagree with a weighting, you can see exactly which one to argue with.
The composite
Every scored name gets a composite between -100 and +100. It is a weighted blend of two readings of the same options chain — where today's contracts actually traded, and where standing positioning already sits.
| Volume bias | 62% | (call volume − put volume) ÷ total volume, × 100. What traded today. |
| Open-interest bias | 38% | (call OI − put OI) ÷ total OI, × 100. What was already on the books. |
When a name has volume but no open interest (or the reverse), the composite is simply the reading that exists — it is not padded with a zero for the missing half, because a missing reading is not a neutral one.
Why some names carry a momentum reading
Separately from the composite, a name gets a momentum score when it has both volume and open interest. It asks whether today's flow is pushing in the same direction as the standing position, scaled by turnover — today's contracts against the open interest they are trading into. Flow that agrees with a large book at high turnover scores highest; flow fighting the book scores negative.
Confidence is not conviction
Composite says how one-sided the flow is. Confidence says how much there is to stand on. A name can print a near-perfect composite off a handful of contracts — that lands as high composite, low confidence, and it is the most common way to get hurt by this tool. The verdict shown on every name folds confidence in, which is why a +99 composite can still read "verify before sizing".
Quality
Quality grades the data underneath the score rather than the score itself: how much chain there was to read, and how much of the reading rests on a single print. A thin score is not a wrong score — it is a score built on very little.
The one verdict
Direction and trust are combined once, in VerdictLib, and every surface reads that same function — the desk, a name's own page, the sector listing and search results. It exists because three different definitions of "bullish" were once visible at the same time. Trust is gated on confidence, quality, choppiness, stability and total contract volume; failing any one of them keeps the direction but changes the read ("Bullish mildly, high chop" rather than "Bullish").
What this does not do
It does not forecast earnings, treat a headline as truth, or know your position size. Flow into a print is positioning, not prediction. The research models in the AI Lab are labelled as research and are never allowed to reach the auto-trade path. Track record exists because we would rather you judged this on resolved calls than on a description of the method.
Check it yourself
The composite and momentum formulas are in api/lib/score_lib.php; the verdict thresholds are in api/lib/verdict_lib.php. System report shows what has actually run and when.
Judge it on the record
The method is public. The track record is public. Read both and decide for yourself.