Reference-Lexicon Ontology Reconciliation — revised live demo

demo byChristopher Janz
speed Reference lexicon ON
Reconciling two independently-authored models of one OTN network. Agent T speaks ONF TAPI; Agent I speaks IETF TEAS/OTN. They plan the attack, lift to ontologies, bind to a thin shared reference lexicon, align in two passes, reconcile the residual, co-refer instances, and validate by round-trip. Toggle the reference lexicon ON/OFF to compare the work — especially the verification work.

About this demo — what you're looking at

This is the fully-cognitive, agent-to-agent case: both sides are competent, interrogable agents. Rather than merge schemas, each agent plans how to attack the reconciliation, lifts its data model to an explicit ontology, and — when the reference is on — binds its concepts to a thin, generic reference lexicon that pins identity only (id, names, a shallow class, a disambiguating definition, one canonical example). Correspondences then fall out by shared reference; only the residual (granularity, scope) is negotiated, and instance co-reference is settled separately, from evidence.

Three things are made explicit. (1) At every stage — how the gap is closed and how it is verified, with a measure. (2) Where the hardest cognition acts: planning, the lift, and the residual reconciliation — watch the cognition meter. (3) The value of the reference lexicon: flip it OFF and the same task costs more conversation and, above all, more verification — see the Work ledger.

Planning is part of the cognitive workflow. The agents do not run a fixed pipeline; they choose the order in which gaps are closed and re-plan when they meet a surprise — the false cognate, a scope gap. The planning beat and its re-plans are shown in the agent windows.

Stage Press ▶ Play to begin. The agents will plan the attack, lift to ontologies, bind to the reference lexicon, align, reconcile the residual, co-refer instances, and validate.
Cognition load
—
Gap closed by
—
Verified by · measure
—

🟢 Agent T — TAPI cognitive proxyONF TR-547 / tapi-topology

🟡 A2A-T message busTMF Agentic / JSON-RPC

🔵 Agent I — IETF TEAS cognitive proxyRFC 8795 + otn-topo

Reference lexicon
Alignment
Knowledge graphs & instances
Residual decisions
Translators
Traces
Measures & cognition
Work ledger · with vs without
Semantic model
A thin, generic reference — neither TAPI nor IETF/ACTN, only the essential categories. Each side binds its concepts here; entries light up as they are bound. Two entries have a scope gap on one side (amber).
Two passes. Pass 1 — label: cheap candidates on names alone (deliberately over/under-proposes). Pass 2 — gloss + example: the disambiguating definition and canonical example confirm, reject, or mark a gap.
No alignment yet — press Play.
The lift makes each implicit ontology explicit (heavy cognition). Instance co-reference — a separate step — settles which individuals are the same, from evidence.

Agent T — lifted TAPI ontology

— not yet lifted —

Agent I — lifted IETF ontology

— not yet lifted —

Instance co-reference — which individuals co-refer, from evidence (shared key → attribute → topology)

Instances co-refer in stage 8.
The residual after alignment — granularity and scope gaps — closed by derived transforms and by self-extension, each verified and logged.
GapTAPIIETFClosure & verificationConf.
No residual decisions yet.
The negotiated artefact: two directional translators (dictionary + structural transforms + identity map), composed from the committed decisions and the reference bindings. Cognition is spent here, once; runtime executes with no inference per message.

T→I translator

— generated after mediation —

I→T translator

— generated after mediation —
Deterministic application of the translators for each live change — inbound native instance, rule firings, emitted native instance. No model inference per message.
No translations yet.
The per-stage ledger: how each gap is closed, how it is verified, the measure, and the cognition load. This is requirements (1) and (2) made durable.
StageCognitionClosed byVerified byMeasure
Fills in as stages complete.
Same reconciliation, both ways. The reference lexicon does not just save conversation — it saves verification: correspondences arrive as near-identity bindings (checked once, reused), and the NEP≈TTP false cognate never becomes a candidate to be rejected.
The form-independent semantic model — meaning (semantics and pragmatics) lifted clear of any form, with provenance. This is the artefact a reasoning agent consumes directly, and the hub the translators merely project from. Ad hoc for this network and anchored to the thin reference lexicon; expressed as an RDF/OWL graph with SKOS lexicon bindings and PROV-O provenance. Not a universal model.
Assembled at the final stage — press Play, or jump with Step.

TAPI view — single multi-layer topology

ROADM Access node OMS link spur

IETF view — stacked per-layer networks

te-node te-link supporting-tp
Staged demonstration — the conversation is scripted for legibility; a code-backed run is scaffolded separately. Schema fragments are faithful but abbreviated. Reference-lexicon entries are generic (neither ONF nor IETF), pinning identity only. A2A envelopes follow the TM Forum Agentic (A2A-T) JSON-RPC message/send shape.