Where the matter is, and when it usually moves.
Eight stages, inferred from event types and never entered by hand. Days in stage, expected exit, expected review volume and discovery intensity, each with the number of comparable matters behind it. Timing and size only; nothing here estimates who prevails.
A stage is a conclusion from event types, with a confidence.
The classifier reads entries, not merits. A scheduling order opens discovery; a dispositive motion filed moves the matter to dispositive; a judgment resolves it. The stage_confidence is shown everywhere the stage is.
A range is only as honest as the matters behind it.
Comparables are matters in the same court and category with similar custodian counts and posture; the count is printed next to every range; below 100 we show nothing and say so.
An estimate you can trust is one that declines when it should. Every range you do see is built the same way: same court, same category, similar custodian counts and posture, and the number of matters behind it printed next to it.
Medians over a judge's own matters say how long things take, not how they will be decided; the product has no outcome field and the API returns unsupported_query if asked for one.
Timing, size and process. Never merits.
That line is what lets a general counsel, a bar association and a regulator trust the number.