Rig-gated — the OSC bindings move to /rlx/v1

Not iGPU-gated and not runnable at the desk either. It needs the lighting rig — Arena, the TouchDesigner patch and whatever show file the set runs from. Extracted from Plan 0152 at that plan’s close (2026-09-05) so the plan could close on its four completed dev phases without waiting on a rig session, exactly as every item above was.

The break is silent, and this check is its only detector. ADR-0164 moved the OSC address root from /lmv/v1 to /rlx/v1 in one clean break — no transition period, no dual-emit. OSC has no negotiation and no error channel, so a binding left on the old root simply stops firing: the fixture holds its last value, nothing in the app, the console or the protocol reports it, and from front of house that is indistinguishable from a fixture that happens not to be moving. Nothing in CI, and no synthetic run, can see the binding half.

• Re-point every OSC binding from /lmv/v1/… to /rlx/v1/…. Fourteen addresses; the table in Configuration is the roster. Only the root changes — no payload, argument type tag, address suffix, vocabulary or send cadence moved, so /v1 stays and a binding that re-points its root and changes nothing else is correct. Report each of the fourteen driving what it drove before, confirmed against a playing track with ritmolux --osc <host:port>. A “looks fine” reading does not discharge this — a missed binding is invisible until the moment the show needs it. Keep the old show file until every address is confirmed: a build from before Plan 0152 and a build after it cannot drive the same bindings, and /v1 deliberately does not distinguish them on the wire. (Plan 0152 Phase 5, extracted at that plan’s close 2026-09-05.)

What is already known, and what it does not cover. A synthetic acceptance run on 2026-09-05 took the app side of the break: the shipped sink aimed at a local UDP receiver, a throwaway decoder self-tested against three known-good encoded messages first, 1450 sets over ~24 s against a live loopback track. It establishes that all fourteen addresses publish under /rlx/v1, none on the old root and none outside it; that every argument type tag matches ADR-0144; that every datagram is a multiple of 4 bytes; and that n was equal across all fourteen, so no partial set was sent. The reading is in the plan’s ## Implementation log. It reaches everything on the app side of the socket and nothing on the binding side, which is the half this item exists for.