Non-standard Cross-domain Provisioning — cognitive agent-to-agent live demo

demo byChristopher Janz
speed Ad-hoc reference ON
Two systems with no standard in common must exchange provisioning requirements unambiguously. Agent X — Meridian is a bespoke transport OSS (circuits, bearers, wavelengths, grades); Agent Y — Cascade is a bespoke IP/VPN controller (services, endpoints, VLANs, throughput). Cascade's service rides Meridian's circuit as its underlay — the one seam their worlds must meet at. They construct a minimal shared reference between themselves and pin every requirement to a canonical example, then validate by worked provisioning & read-back. Toggle the ad-hoc reference ON/OFF to compare the work.

About this demo — what you're looking at

The first two demos let each agent recognise a standard — TAPI, IETF, TM Forum — and lean on a reference that already mapped to it. Here that crutch is gone. Meridian and Cascade are wholly home-grown, and they are in different domains: one provisions optical transport, the other IP/VPN services. Their models overlap only where a Cascade service must ride a Meridian circuit. Neither can name a standard to disambiguate; both are competent, interrogable agents that must find common ground from first principles.

Two things have to happen, and both are shown. First, the agents construct an ad-hoc reference between just the two of them — the minimal set of categories the exchange needs (endpoint, committed rate, protection, latency, and the cross-domain transport-underlay seam), each defined by a canonical example they can both instantiate. It is deliberately small and ad hoc, not universal. Second, they pin every provisioning requirement so it cannot be misread: is a rate a line rate or a committed payload? protection against which failure? a latency tier or a number? These are exactly the ambiguities that silently mis-provision across domains.

Verification is a worked provisioning, read back. With no standard and two different domains, round-trip identity is meaningless — nothing looks alike. So the test is concrete: run one real order through the exchange, have the receiving side provision it and read the instance back, and confirm every pinned requirement is honoured. Watch the cognition meter (heaviest at planning, the standard-free lift, constructing the reference, and pinning the requirements), and the pragmatics — realm (transport vs IP ownership) and authority (who is source of truth for each field).

Stage Press ▶ Play to begin. The agents will plan, lift with no standard to lean on, construct a shared reference by example, bind, align, pin every requirement unambiguously, co-refer endpoints, build the exchange mapping, and validate by worked provisioning and read-back.
Cognition load
—
Gap closed by
—
Verified by · measure
—

🟢 Agent X — Meridian (transport OSS)bespoke · circuits / bearers

🟡 A2A message busTMF Agentic / JSON-RPC

🔵 Agent Y — Cascade (IP/VPN controller)bespoke · services / endpoints

Ad-hoc reference
Alignment
Knowledge graphs & instances
Pin requirements
Exchange mappings
Provisioning & read-back
Measures & cognition
Work ledger · with vs without
Semantic model
This reference is not adopted from a standard — it is built here, between just these two agents, for just this exchange. Each category is proposed, then defined by a canonical example both can instantiate, then agreed. It is minimal and ad hoc, not universal. The amber entry is the cross-domain seam.
Two passes. Pass 1 — label: cheap candidates on names alone. Pass 2 — gloss + example: the definition and canonical example confirm, reject, or refer to pinning. With no standard, the trap is worse: the same word (grade) means unrelated things in the two systems.
No alignment yet — press Play.
The lift makes each implicit ontology explicit — and here there is no standard to recognise, so it is recovered from instances and explanations alone (heavier, lower initial confidence). Endpoint co-reference — a separate step — settles which hand-off and which attachment are the same demarcation.

Agent X — lifted Meridian ontology

— not yet lifted —

Agent Y — lifted Cascade ontology

— not yet lifted —

Endpoint co-reference — which physical hand-off and which logical attachment are the same demarcation (the IP-over-transport seam)

Endpoints co-refer in stage 8.
The crux of unambiguous exchange. Each provisioning requirement is probed and pinned to a canonical example both agents instantiate and agree — so it cannot be silently misread. These are the errors that would otherwise slip through: a line rate taken for a committed payload, protection against the wrong failure, a tier read as a number.
RequirementMeridian saysCascade saysAmbiguity → pinned meaning (by canonical example)✓
No requirements pinned yet.
The negotiated artefact: two directional exchange mappings. Y→X — a Cascade service request yields a Meridian work order for the underlay circuit. X→Y — a Meridian circuit event updates the Cascade service. Cross-domain, so not inverses; each carries the pinned meanings and the co-referred endpoints. Cognition spent once; runtime is deterministic.

Y→X — Cascade service ▶ Meridian work order

— generated after mediation —

X→Y — Meridian circuit event ▶ Cascade service

— generated after mediation —
The right test with no standard and two domains: run a real order through the exchange, provision it, and read the instance back to confirm every pinned requirement is honoured. Ambiguity would surface here as a read-back mismatch — not as a schema error.
No provisioning exchanges yet.
The per-stage ledger. Note the verification story with no standard: reference categories agreed by canonical example, requirements pinned by canonical instance, and the exchange validated by worked provisioning + read-back.
StageCognitionClosed byVerified byMeasure
Fills in as stages complete.
Same reconciliation, both ways. With no standard, the honest cost of the reference is that it must be built, not adopted — a couple of turns up front. That cost is paid once and reused; without it, the two bespoke models are aligned pairwise, the grade cognate must be caught by conversation, and every requirement's ambiguity is re-litigated per pair.
The form-independent semantic model — meaning (semantics and pragmatics) lifted clear of both bespoke forms, with provenance. It is ad hoc, not universal: built for this pair and this exchange, anchored to the reference they constructed. An agent consumes this directly; the Meridian and Cascade forms are projections of it.
Assembled at the final stage — press Play, or jump with Step.

Meridian view — transport circuits (optical)

site / ROADM fiber span circuit (bearer)

Cascade view — IP/VPN services (logical)

honoured margin / advisory service (pipe) underlay → circuit
Staged demonstration — the conversation is scripted for legibility; a code-backed run is scaffolded separately. Both data models are deliberately non-standard (bespoke to each system). No published standard is used or recognised; the shared reference is constructed ad hoc between the two agents. A2A envelopes follow the TM Forum Agentic (A2A) JSON-RPC message/send shape.