Finality & Settlement

State machine

PENDING → VERIFIED → FINALIZED → SETTLED.

PENDING

Runner has submitted a candidate checkpoint/result.

VERIFIED

The verification mechanism required by this World's verification class has accepted it.

FINALIZED

Challenge/finality conditions are complete and the result is the canonical economic result.

SETTLED

The settlement layer has converted the terminal result into claims exactly once.

v1 verification class

For large Worlds, the current practical target is independent deterministic replay + signed attestations.

This provides reproducibility and a measurable trust model but is not a mathematical proof. Colluding verifiers can lie. v1 documentation must say so.

Exposure limits

Because verification classes differ, protocol risk limits should differ too. A REPLAYED World can have conservative maximum economic exposure until stronger adjudication is proven.

Conflict behavior

If independently recomputed roots disagree:

• do not advance finality;
• freeze at last FINALIZED checkpoint;
• preserve both claims/evidence;
• resolve through the applicable challenge process;
• do not move player principal based on either unfinalized branch.

Longer-term directions

• optimistic application-level fraud proofs;
• zk/validity proofs via a selected zkVM;
• hybrid proof strategies.

These are later research tracks, not v1 guarantees.