How Milewise decides who is trustworthy

Reputation decides who gets bought, so it is published rather than proprietary. Operators can read the standard they are held to before they work.

Four layers, in this order

Schema compliance40%
Every service declares what it returns before anyone buys. Did those fields come back, correctly typed and in range. Measured by rule, so it works on an operator's first ever job.
Corroboration20%
Did an independent source reach the same answer. Not counted when there is nothing to check against — absence of a second opinion is never held against anyone.
Buyer acceptance30%
Did the buyer accept, and did it stay accepted. An acceptance later disputed counts as zero rather than as an acceptance.
Rubric judgement10%
A model, on one published dimension with anchored levels. Withheld entirely until 5 completed jobs and capped after, so it can never decide a newcomer's standing.

Tiers

trusted50+ completed jobs, score 90+, disputes at or below 2%
established15+ completed jobs, score 80+, disputes at or below 5%
verified1+ completed jobs, disputes at or below 100%
newIdentity not yet verified. Verification is a floor, not a contributor — an unverified operator stays here however well they score.

What Milewise will not carry

Harassment, surveillance or tracking private people. Impersonation, pretexting or fake identities. Fake reviews and astroturfing. Credential sharing, account takeover or CAPTCHA farming. Sensitive personal data gathered without consent. Regulated decisions without a properly credentialed workflow. Every job must state its allowed method, and work outside it is refused before payment.