What the work actually is

You write a 60–120 word question in your own words — no copied phrasing, no search hints — whose answer is a single definitive fact: a name, a date, a catalogue number, a documented title. Behind that prompt sits your real deliverable: four to six independently verifiable clues drawn from different reputable open English sources, each captured as a verbatim quote with a URL and the date you checked it. Wikipedia and paywalled pages are out. You also document the plausible wrong answers, name the specific clue that eliminates each one, and write up the search path you took to get there.

The task is then run against frontier models. If a model solves it instantly, it was too easy and you refine it. If a reviewer finds a clue ambiguous, an undefined term, a logical leak that gives the answer away early, or a quote that doesn't match the page when opened logged out, it comes back. Only approved tasks count toward pay.

What the platform screens for

Turing's assessment is looking for people who have done sourced research under accountability — journalism and fact-checking, library or archival work, genealogy, due diligence, graduate primary-source research, citation-heavy editing. Expect probing on the mechanics: search operators, reverse image and reverse citation lookups, going straight into institutional catalogues rather than through a search engine, the Wayback Machine, preprint servers, court and legislative registers, specialized sports and film databases. Expect equal probing on judgment: primary versus secondary, original versus republished, what a source states versus what it implies. LLM tools are allowed; unverified LLM output is the fastest way to fail review.

Logistics and economics

  • Fully remote, contract, any time zone, paid in USD; NDA required because task materials are client IP.
  • Roughly four hours per completed task, at an observed rate of about $30 per approved task — do that arithmetic before you apply, and note that rejected or reworked tasks are unpaid effort.
  • Stated as a six-week project, 20+ hours per week minimum and ideally up to 40.
  • Work is asynchronous but review is iterative; you should expect revision requests rather than fire-and-forget submissions.