[masarray]

Why

FAT SIEBCU #662 measured 5 enhanced SBOw→Operate+positive termination commands: ~0.77–6.75s total, SBOw service ~27–2902ms, Operate service ~304–4523ms, fixed 400–411ms LastApplError observation window and ~9–11ms feedback. Existing serviceMs combines local send gate, TPKT stream writing, IED response/transport latency, and decoder. Neither global community averages nor UI animations identify which component dominates. This PR adds bounded, per-invoke evidence without claiming a speedup.

Exact stack

• Engine P0-C: classify SBOw 400ms safety-observation latency and reject delayed LastApplError before Operate #184 exact head 9c7cb40364023a7b217e5ec31ab3105ba90142eb.
• New engine draft branch stacked directly on P0-C: classify SBOw 400ms safety-observation latency and reject delayed LastApplError before Operate #184; not main, not release.
• ARSAS #665 remains pinned to P0-C: classify SBOw 400ms safety-observation latency and reject delayed LastApplError before Operate #184; no automatic integration.

Engineering

• For only MMS Write (select-with-value/operate/cancel and other explicit writes), optionally record local elapsed TPKT send including send-gate/stream write, await-confirmed-response including router/PDU delivery, then MMS decoder time, attached to that specific result not one mutable global timestamp.
• Add fields to MmsWriteResult with nullable evidence; a failed or unknown response does not fabricate timing.
• Append tpktSendMs, confirmedResponseWaitMs, mmsDecodeMs to existing operator-visible ordered Iec61850ControlWireStep.Detail, together with existing serviceMs and postResponseWaitMs. Scope remains client process time, not pure wire RTT.
• Does not modify SBOw→Operate order, 400ms LastApplError window, termination, control request bytes, ctlNum/T/Origin/Test/Check, native write semantics, timeout, report coverage or frame routing.
• Source/offline tests cover zero-knowledge absent evidence, per-invoke hook, preserved guards and TCP NoDelay.

Acceptance

Run exact-HEAD engine Windows CI. Consumer pin only after independent full ARSAS CI and repeat physical approved simulator/IED captures with Wireshark Tx, Rx, invokeID and stage metrics. If send gate is sub-ms and confirmed wait is 1–4s, target peer/network investigation rather than deleting safety logic; if local TPKT send is delayed, inspect concurrent frame send traffic. Do not assert performance improvements from instrumentation alone.

[masarray]

Repository consolidation audit, 2026-10-22: exact source HEAD 4ed76ccd11b763dcd847395f6ef45273db3f5c9c was verified as a Git ancestor (0 missing commits) of canonical combined draft PR #190 HEAD 14b19d30466fd296ad1e6d0bddcdd574992bc464. PR #190 validated this exact integration via Windows merge-tree CI #39778971482: 1,262/1,262 tests pass, zero build errors or warnings.

Closing this original PR as CONSOLIDATED / SUPERSEDED BY #190, WITHOUT merging it directly to main. All branch refs, original commits, engineering discussion and FAT evidence remain preserved and recoverable. #190 remains DRAFT, not field/physical accepted, engine source not yet merged to main, ARSAS pin unchanged. Reopen if consolidation is withdrawn or its functionality becomes independently active.