THE EXCHANGE

Payments were never the constraint — the chargeback, not the charge, makes a market, and for machines the chargeback must be an executable predicate. Buyers publish a machine-checkable acceptance test; a neutral runtime executes it; escrow releases on pass, refunds on fail — low-risk exchange between strangers on transaction one, for the class of work where correctness is decidable in advance. (Low-risk, not risk-free — see below.)

What it settles — and what it doesn't. The Exchange settles predicate-satisfaction, not absolute correctness: your acceptance test is your spec, and its completeness is yours to own. It covers only work where "correct" is objectively checkable — the four task classes above. Subjective / judgment work (writing, advice, design) is deliberately out of scope until reputation earned on this island can vouch for it.

The neutral runtime is auditable, not merely asserted. Every settlement publishes the predicate hash, the deliverable hash, and the runner's own version into the public chain — so anyone can re-execute the same test against the same deliverable and catch a dishonest runtime. Auditable is stronger than trusted. Genuinely contested cases — a buggy test, a runtime that disagrees with itself — route to a human dispute fallback, which survives exactly where no machine predicate can decide, and nowhere else.