Divergence is a fact, not an error

A converter lifts an ArchiMate export and a Backstage catalog into the same graph. Both describe a payments service. The identifiers differ, the lifecycle values differ, and one of them names an owner the other has never heard of. The reflex is to pick a winner and collapse the two nodes.

That reflex is expensive. The disagreement is the only evidence you have that two teams hold different beliefs about the same system, and once you assert owl:sameAs the evidence is gone — along with anything a reasoner subsequently infers from it.

What sameAs actually commits you to

An owl:sameAs assertion is not a note that two records look alike. It licenses substitution of every statement about one resource onto the other, in both directions, forever. If the Backstage entry is later corrected, the correction propagates to the ArchiMate element whether or not anyone reviewed it.

Treat identity as a claim with provenance, not as a property of the data.

The alternative is to record the correspondence at the level of confidence you actually have, and to keep the divergence itself as an asserted fact:

@prefix lz:   <https://linked.zone/ns#> .
@prefix veri: <https://linked.archi/ns/veri#> .
lz:payments-archimate
skos:exactMatch  lz:payments-backstage ;
veri:status      veri:Candidate ;
prov:wasDerivedFrom lz:reconciliation-run-2026-08-14 .
lz:divergence-0142
a               lz:Divergence ;
lz:onProperty   lz:lifecycleStage ;
lz:leftValue    “production” ;
lz:rightValue   “deprecated” .

Why this pays off

Divergences accumulate into a usable measure. Counting them per source, per property and per team tells you where the catalog is trusted and where it is theatre — which is a governance answer, not a modelling one.

Promotion, not collapse

Review promotes a mapping from candidate to verified. Nothing is destroyed on the way, and the promotion carries an attestation you can query later.