The claim
The reading names a tension it would “deliberately leave open”: “Realist: Communicative acts have relational identities that outrun every finite set of observations. Operationalist: Communicative identity is only specified relative to a family of observable consequences and decisions.” It says the record “sometimes seems to want the empirical rigor of the second while retaining the ontological ambitions of the first”.
Its middle position: “a proposed relational kind is a hypothesis about invariance across admissible changes of measurement and context, not simply an equivalence class under the currently available measurements. That doesn’t establish metaphysical realism. But it gives realism empirical content: hypothesized kinds should support successful extrapolation to new measurement families that weren’t used to construct them.”
“The difference between a merely convenient observational partition and a candidate structural kind would then be partly expressed through how robustly its invariants survive changes in probes, populations, and environments.” It closes with a restatement of the line’s question: “Ask which relations are invariant, under which transformations, relative to which observational capacities—and which of those invariances support predictions beyond the observations that established them.”
What it bears on
- CLAIM-142’s evidential half already requires probes “justified independently and evaluated on held-out cases”. This claim extends held-out cases to held-out probe families, populations and environments.
- CLAIM-209 (restricted recoverability) gets its place here: “which invariant signatures are recoverable under specified assumptions, which remain underdetermined, and which generalize to new interventions.”
- CLAIM-139’s ontological clause and CLAIM-152’s “no such realist commitment” are the two sides of the tension it leaves open.
What it does not say
It does not establish realism, and it does not say every invariant that extrapolates is a kind.
Note of 2026-10-30: what holds it back
The result its defeat condition names is missing: no kind identified by invariance under its constructing probes has been tested against an observational partition on held-out probe families, populations or environments. So whether the invariance adds predictive content, which is the empirical content the claim says realism gains, is untested.