A complete earnout lifecycle on Canton — four parties, rules locked at signing, a forged verdict that fails on the ledger — and then the harder test: earnout clauses from the public record that I did not write, replayed against their own history.
3 min 47 s · English captions. The ledger walkthrough was recorded on LocalNet; the same model now runs on the shared DevNet node, with the update ids in docs/devnet.md.
What you are watching
Signing. Buyer proposes metric definitions, thresholds, evaluation logic, attestor and arbiter. Seller accepts. Both signatures are required for the contract to exist.
A clean quarter. The bank attests the period's account snapshot; the buyer can only submit bank-signed data; the contract evaluates it against the signed threshold and records a verdict carrying both parties' signatures.
A contested quarter. Q3 lands below target, but the ledger flags it because a one-off outflow would have covered the gap. The seller disputes; the arbiter named at signing resolves it and the reasoning is recorded on-ledger — not a forum chosen after the argument started.
The attack. After a missed quarter the buyer tries to write a favorable verdict alone. A verdict is a dual-signatory contract, so the participant node rejects the submission — the error shown on screen is the ledger’s, not the app’s. (Moving the thresholds is not even attemptable: the model has no choice that amends signed terms.)
Settlement. Three of four quarters met, so finalization creates the obligation and Settle moves the tranche and closes the contract in one transaction.
A clause I did not write. Proving the mechanism on my own clause proves little, so five earnout clauses from the Delaware record are replayed against their own history. In Prosser v. PharmaLogic — $6.6M if EBITDA reached $7M, buyer reported $6.8M, nothing paid — the ledger reaches the same zero without ever seeing the case.
What it refuses. Four of the five cannot be expressed at all — tiered payouts, a rising target, FDA milestones that are events rather than figures — and each is published with its reason. Of nine earnout disputes in the public record, two are the shape this model holds.
The self-check. Eight questions about the shape of a clause, no figures, nothing leaving the browser: would yours have compiled, and what happened to the public clauses that look like it.