The forecast model
Forecast Scorecard
Weeks before each conference, this site writes down a forecast for every paid petition on the list and never changes it. This page holds those forecasts to the orders that follow.
Loading the scorecard…
Forecast and outcome, petition by petition
Each dot is a paid petition at its logged forecast, on a log scale. Rows are what the Court did. Hover or tap a dot for the case.
Where the grants ranked
Each grant's place among the petitions logged for its conference. A tick at the left edge means the model ranked it first.
Grants found, reading down the list
Sort every logged petition by its forecast and read down. The line is the share of grants found by each point; the diagonal is a list in random order.
Calibration
Petitions grouped by forecast. The open diamond is the average forecast in the group; the dot is the share actually granted, with a 95% interval. A well-calibrated model puts the dot on the diamond. Groups with few petitions have wide intervals.
Conference by conference
Expected grants is the sum of the logged forecasts. Scores use paid petitions only; IFP petitions carry no forecast. Granted ever counts grants at this or any later conference.
On deck
Conferences not yet held, with the forecasts already in the ledger. These are the numbers they will be scored on.
How this page is scored
The forecasts
Each paid petition gets two forecasts at each conference: the chance it is granted at that conference, and the chance it is ever granted. The scores here use the first. The second can be shown on the dot plot and is totalled in the table. Only paid petitions are scored.
Written down in advance
The forecasts come from the site's forecast ledger. The nightly run adds a petition the first time it appears on an upcoming conference's list and never edits the entry afterwards, and nothing is logged for a conference that has already happened. These are the numbers the site committed to, often weeks ahead. Each entry also carries a final forecast: re-scored every night while its conference is ahead and frozen at noon UTC on conference day, before the Justices meet, so it is the last forecast the site made with everything it knew. The switch at the top of the page scores either one. Both can differ from the conference pages, which re-score past conferences with the current model and the case as it stands now. An entry written by code later found to be broken is re-scored if its conference is still ahead, and left out if it is not; the ledger keeps the original numbers either way.
The outcome
Read from the orders released between the conference and the next one. Granted and GVR'd are as ordered. Carried over means the order lists named no disposition: relisted, held, rescheduled, a call for the Solicitor General's views, or only the IFP motion decided. Carried over and GVR'd both count as not granted at this conference. The granted-ever view uses everything ordered since the conference. Habeas, mandamus, prohibition and rehearing entries are left out.
The yardstick
The backtest figures are the conference model's rolling-origin results: trained on earlier Terms and scored on OT2019–OT2024, 11,716 petition-conferences with 346 grants. See the methods note. A Term's live record will be noisy for months: the Court grants a few dozen petitions a Term, and AUC and calibration from a handful of grants carry wide intervals.
Forecasts from the site's forecast ledger (cases/forecasts.json); outcomes from the Court's order lists as parsed for the order list pages. The data behind this page: scorecard.json.