Research API

This endpoint returns the research dataset: the de-identified, opted-in contributor stream (trials from accounts that opted into anonymous research contribution). It is dataset-level counts and statistics only — never any per-account data, choices, outcomes, or timestamps. The broader community activity figure (every trial that passes integrity checks, which may include automated actors) is shown on the transparency page, not here.

Endpoint

GET /api/research/aggregate

No API key required. Responses are cached for 5 minutes.

Parameters

paradigm	Optional. Restrict to one paradigm (e.g. `forced_choice`). Omit for all.
format	Optional. `json` (default) or `csv`.

Examples

# All paradigms, JSON
curl https://fathomresearch.org/api/research/aggregate

# Forced-choice only, as CSV
curl "https://fathomresearch.org/api/research/aggregate?paradigm=forced_choice&format=csv"

Response

JSON returns an overall row plus a byParadigm array. Each row reports trials (valid n), hits (matched k), hitRate, and — once the sample is large enough — the two-sided z / p against the chance baseline, a Wilson 95% interval, and the log Bayes factor (logBf10, uniform prior). Below the minimum sample size the inferential fields are null— we don't report z/p/BF on a handful of trials.

{
  "filters": { "paradigm": null, "from": null, "to": null },
  "overall": {
    "paradigm": "all",
    "trials": 1280, "hits": 657, "hitRate": 0.513,
    "z": 0.95, "p": 0.34,
    "wilson95": { "lower": 0.485, "upper": 0.540 },
    "logBf10": -1.82
  },
  "byParadigm": [ { "paradigm": "forced_choice", ... } ]
}

Honest framing

This dataset is gated to accounts that opted into contribution so the population is self-selected, not merely integrity-valid (passing the HMAC + timestamp checks proves a trial was committed before its outcome — it does not prove a human made the choice). A hit rate is always paired with its interval and test statistics — never reported bare — and Fathom takes no position on what the aggregate shows. See our methodology for data provenance and how to read these numbers.