Escrow a bounty in USDC. Every dataset item clears a verification pipeline — dedupe, contamination screen, sandboxed tests, LLM review — before a single token moves.
Funded datasets accepting submissions right now.
Every accepted item earned its way through an automated pipeline — payment only follows acceptance.
Solutions run in the sandbox — tests, SQL results, and regex matches must pass mechanically.
Near-duplicates are caught by similarity scoring before they reach validation.
Every item is screened against public benchmarks so datasets don't leak eval sets.
Ranked validators spot-check accepted batches and earn bonuses for confirmed issues.
From spec to delivered dataset — every step is funded, verified, and auditable.
Sponsors fund, contributors build, validators keep everyone honest.
Completed datasets with public quality stats and payout records.
Whether you're buying training data or earning from building it, the pipeline treats everyone the same: accepted work gets paid, everything else does not.