The model's estimate of the next short-interest settlement — the one FINRA has already measured but has not published yet. This is a PREDICTION, not a reported figure: never cite it as FINRA data. It carries a likely range (P10-P90) and a confidence band, and it is withdrawn as soon as the official figure for that settlement exists. Returns 404 when there is nothing to forecast, which is the normal state in the days after a release.

The settlement date being estimated. FINRA has measured it; the official figure is not out yet.

The date FINRA is expected to publish the official figure for that settlement, after which this forecast is withdrawn.

The last settlement with an official figure — the anchor this estimate grew from.

The official position at previousSettlementDate (shares), restated onto today's split basis so it is directly comparable with the estimate.

The estimated short position at settlementDate (shares), on the same split basis. A model output, not a reported figure.

The estimated change against previousReportedPosition, in percent.

Low end of the model's own likely range (shares, same split basis) — its 10th percentile.

High end of the model's own likely range (shares, same split basis) — its 90th percentile.

How firm the direction call is: high, medium or low, from how far riseProbability sits from the model's decision cut. Bands are cut at measured quantiles, not chosen. Measured on held-out models of the same configuration over 2025 to July 2026, the high band called the direction correctly about 72% of the time and the medium-and-above band about 69%, against about 62% over all estimates.

The model's own probability, 0 to 1, that the short position RISES at settlementDate against previousReportedPosition.

This is a separate model output, not the sign of estimatedChangePercent, and the two can disagree. Read direction from expectsIncrease, which applies the model's own decision cut; that cut is not 0.5, so comparing this value against 0.5 will give the wrong answer near the boundary.

Whether the model expects the position to rise, applying its own decision cut.

The last FINRA daily short-volume date the estimate reads. Its inputs stop here.

Says in words what every field above is, so an answer built from this response cannot present it as a reported position.