Tick: fresh audit of system-laws.md traced into flow-substrate.md and the substrate consumers. The engine and its faithful consumers verify (FlowGraph + Schedule pins, reach device graph, WorkGrid, message schedule, account graph; no rival engine anywhere), but the per-domain riding claims were written before messages and economy landed: no social FlowGraph with per-channel edges exists (channels are typed message fields with domain read conditions), and AccountFlow recurrence is per-flow cadence/next_tick reschedule-on-fire, not Schedule events. Both pages now state the as-built shape and when the social/filing graph becomes necessary (the routed-evidence migration). Defense: wiki/overview.md — law and spec pages are the binding surface, and these described an architecture the tree does not have; the next flow-system implementer would hunt for or silently build the missing half. The code is right: flow-substrate.md's own non-goal declares recurrence a domain concern, and B1 delivery needs no topology until detection.md's routed-evidence migration. The law's invariant — one engine, thin domain layers — is preserved and now precise.