Capture
Collect the agent intent, transaction evidence, market identity, balances, order events and relevant indexed state.
OPEN METHODOLOGY · NO BLACK BOX
From fragmented evidence to a bounded decision.
Last Known Book does not ask a model to invent a story. It reconstructs a DreamDEX incident in stages, labels what is known versus inferred, and lets evidence determine whether the safest response is action, retry, escalation — or doing nothing.
Reconstruct
Align the request with the actual DreamDEX venue path so expected behavior and observed behavior can be compared directly.
Classify
Tag each material claim as OBSERVED, INFERRED or UNKNOWN. Unknowns remain visible instead of being smoothed over.
Explain
Apply DreamDEX-native causal semantics such as MINT_A_PAIR, resting escrow, lifecycle state and event-chain reconciliation.
Decide
Produce a safe next action and Authority Receipt. Inference alone cannot authorize spend.
Directly supported by captured or live-read evidence: receipts, balances, lifecycle state, market IDs and native events.
A bounded interpretation supported by observed evidence. It can explain, prioritize or recommend — but it does not become authority.
An unresolved gap. Unknown evidence can stop a write, force a retry read, or require escalation.
VENUE-NATIVE CAUSALITY
Why the method cannot be generic.
DreamDEX Event Contracts have semantics that change what an execution means. Last Known Book treats marketId, current lifecycle, MINT_A_PAIR, resting SELL escrow and native order events as part of the reasoning substrate rather than branding around a generic analytics engine.
marketIdlifecycleMINT_A_PAIRescrowOrderPlacedOrderRestedOrderCancelled