@servanda/node

L1 — the edge state machine — plus the §7 MCP surface that every client speaks to.

The transition table is the constitution's enforcement point

const result = verifyAssertionChain(edge, assertions);
// → { outcomes: [{ index, accepted, rejection_reason? }], finalState }

§4.3: "Any assertion violating this table is invalid and MUST be discarded." This function is where that becomes true. It reproduces all 26 conformance vectors — 7 chains a node must accept and 19 it must reject — with the exact rejection_reason strings the suite asserts on.

The 19 negative cases matter more than the positive ones. A verifier that accepts a confirmed assertion signed by the owner rather than the counterparty has silently discarded the entire confirm-first guarantee, and would still pass every positive test. Among the cases it must reject:

• the owner self-confirming their own proposal (M-2)
• an assertion attributed to the counterparty but signed with the owner's key
• release asserted by the owner instead of owed_to — release is forgiveness, and only the person owed can forgive
• an owner recording tacit acceptance before the acceptance window has elapsed
• an explicit open assertion (§4.3 marks that transition implicit)

Effective state folds confirmed into open and adds pending-acceptance to model the §4.4 window, which §4.3 has no row for. Both are interpretations, documented upstream (issues #12, #13), not invented silently.

The six tools (§7)

const node = new ServandaNode({ vault, activePersona, now });
node.commit({ intent, owed_to, due, propose });
node.expect({ expect, from, context });
node.confirm({ id, decision, edit });
node.openLoops({ view, persona, limit });
node.brief({ persona });

Exposed over MCP stdio via mcp/stdio.js. Clients — assistants, the web ledger, the TUI, the email brief — are interchangeable above this contract; that is the point of having one.

commit throws M1Violation if handed an owner that is not the caller's persona. You cannot record someone else's promise as theirs; the correct object is expect. assertNoForeignOwner is exported so the rule is testable in isolation.

What the node will not do

• M-11 — no reputation. It does not compute, store or serve cross-party fulfillment statistics. localPairwiseHistory is deliberately local and pairwise; M11Violation exists to make the boundary explicit rather than implicit.
• M-8 — no auto-escalation of unverifiable edges. An edge with no verification adapter, or an invalid collective decomposition, is marked unverifiable and stays quiet. Escalating something you cannot verify is how a notary becomes a nag.
• Undated commitments never time-escalate (§3.1). due: null is expected to be the majority case; those rank by age × blocking, never by an invented deadline.

Ranking

brief orders by the attention market — cost_of_delay × confidence × people_blocked — and never by manual sort. With persona: null the ordering is cross-persona, which is the sole exception §5.3 allows; each slot's content still originates from exactly one persona's pipeline (M-5).