Prospective benchmark

5,809 ongoing trials, open for forecasts

The benchmark covers the eligible ongoing trials of companies listed in the United States, including trials without an announced readout date. Forecasts for every trial were recorded before outcome disclosure and stay fixed as outcomes accumulate. The trial list is public so that any method can be scored on the same trials as outcomes are adjudicated.

Download the trial list (CSV) How to submit forecasts

How the cohort was built

ClinicalTrials.gov, queried in October 2026

  1. 121,961records with an ongoing status: recruiting, not yet recruiting, active and not recruiting, or enrolling by invitation
  2. 10,607with a lead sponsor or collaborator listed on the NYSE, Nasdaq, NYSE American or Cboe, including American depositary receipts and subsidiaries mapped to their listed parent
  3. 8,686interventional studies
  4. 7,418with a specified phase, including early phase 1
  5. 6,855not enrolling healthy volunteers
  6. 5,983after removing 872 trials with a disclosed final result or an interim stop, found by web search of the 1,736 trials whose primary completion date had passed or that had posted results
  7. 5,809after excluding 86 trials whose dossiers listed no intervention, 20 in the development cohort, 25 in the retrospective test cohort and 43 of the initial prospective cohort whose readouts had been disclosed before the forecasts were recorded

What is recorded

  • Forecasts. Four forecasts per trial, recorded before outcome disclosure: the GPT-5 reasoning expert of TrialOutlook and Claude Opus 5.5 in Claude Code as an additional reasoning expert, given the same prompts and TrialDossier evidence dossiers, and zero shot forecasts of Claude Opus 5.5 in Claude Code and GPT 6.1 Sol in Codex, given only the NCT ID. Forecast values and evidence dossiers are not published on this site.
  • Outcomes. Adjudicated under the same criteria as the initial prospective cohort, each with its supporting source, and updated monthly; disclosures that support neither label stay unresolved.
  • Scoring. Submitted methods and TrialOutlook are compared on shared eligible trials by ROC AUC, average precision and classification errors, and by Brier score and expected calibration error for probability forecasts. Updates include uncertainty estimates, forecast coverage and the numbers of adjudicated, unresolved and pending trials. Original forecasts remain fixed as outcomes accumulate.
  • Your forecasts. Submit for any subset of trials while they are pending; see Contribute.

Trial list

Select a phase, then type in the box under any column heading to search that column. Each identifier links to its ClinicalTrials.gov record.

Company: US listed lead sponsor or collaborator at cohort assembly (American depositary receipts and subsidiaries mapped to the listed parent). Phase: ClinicalTrials.gov phase at cohort assembly. Drug, drug targets (target genes) and disease: from the TrialDossier evidence dossier.