[nmrg] New I-D: draft-janz-nmrg-adhoc-semantic-reconciliation-00

[Christopher Janz <[e-mail]>]

Dear NMRG,

Brad Peters and I have posted a new Internet-Draft, "Portable Semantic
Models and Ad Hoc Reconciliation by Cognitive Agents in Network
Management", which may be of interest to the group. It follows on from the
ontology reconciliation work we brought to IETF 126.

The draft is here:
https://datatracker.ietf.org/doc/draft-janz-nmrg-adhoc-semantic-reconciliation/


The draft sets out two operations a cognitive agent can carry out on
semantic models, and keeps them distinct. The first is the lift: producing
a single portable semantic model, complete enough that a cognitive system
can comprehend it on its own. The second is reconciliation: bridging two
such models so the systems that produced them can interoperate. We describe
each as a ReAct-style flow that the agent both orchestrates and, in part,
performs itself, interrogating the live system, constructing or recovering
references, and running virtual experiments that never alter the
operational system.

One point we try to make carefully is that a good deal of a signal's
operative meaning rides in its pragmatics: significance, authority, intent,
operative state. That is a layer that schemas are least likely to hold, and
the study suggests a sufficiently capable agent can carry or recover it
where a weaker one cannot.

We also look at thin lexical references: a small shared reference that each
system binds to once. What interests us here is the reduction they afford:
a reference the agents can bind to lowers the cognitive power and effort
that both the lift and reconciliation demand, since less must be inferred
or constructed afresh on each side. Such references are comparatively easy
to maintain and evolve, and where one is absent a capable agent can
construct it.

The claims are backed by a small reproducible harness: the lift and
reconciliation pipeline, the datasets, and a worked trace of the agent's
cognitive reasoning for both flows. It is archived here:

https://doi.org/10.5281/zenodo.23187404

Comments and criticism are welcome, on the list or to Brad and me directly.
We ask to present this at IETF 127 rather than the upcoming interim.

Thanks,
Chris