nucleation

CLAIM-142: Two communicative realizations can agree on every probe in a restricted family without being structurally identical, so observational equivalence is relative to the probes, and only independently justified probes evaluated on held-out cases are evidence that structure was preserved

6 min read

Properties15
StatusProposed
Version3
Rolethesis
Warrantinterpretive
Tagsepistemology, individuation, mathematics
Linepragmatic-transport
Rests onCLAIM-046 — Different hidden-state models can produce identical observations, so pragmatic identity and fidelity should be defined on observable responses before any equivalence of internal states is assumed, CLAIM-011 — Intertwining with an uninformative correspondence is vacuous or lossy, so the source–target correspondence must be anchored to independently measured distinctions and evaluated on tasks other than those it was estimated on, CLAIM-264 — Two objects can agree on every probe in a restricted family without being isomorphic, so equivalence under a probe family is coarser than identity unless the family is rich enough, CLAIM-322 — Agreement under a probe family is evidence that two objects share structure only to the degree that the family was chosen independently of the comparison and is evaluated on cases it was not chosen on, so a family selected to make two objects look alike shows only that it can be selected
GroundsTHEORY-032 — The Yoneda lemma determines an object only up to isomorphism, from its whole hom-functor, among objects the category already has; as a formal version of "an object is determined by its relations" it presupposes the relata, LIT-221 — Yoneda lemma, CASE-070 — Voita, Sennrich and Titov's BLEU-tied translation systems that differ on keeping ты or вы across sentences
SupportsCLAIM-129 — A factorization recovers only the structure visible in the relations it is fitted to, so a topic can survive a transformation that changes the communicative act, and a kind must be tested against probe families chosen for their bearing on the act, CLAIM-133 — Communicative identity is relational continuity under admissible transformation: translation transports relationally organized observational structure, correspondence fixes what can be compared, admissibility which transformations count, and fidelity which communicatively consequential distinctions remain recoverable, CLAIM-139 — A communicative kind's identity is grounded in its relational organization rather than in any expression or encoding; statistical methods recover aspects of that organization from restricted observations, translation transports it, and fidelity is which independently identifiable, decision-relevant relational distinctions persist, CLAIM-204 — A communicative kind's identity is grounded in its relational organization, for which competent participants' uptake is the standard; statistical methods recover aspects of it from restricted observations; translation transports it; and what persists is what the receivers can still recover, of which the analyst can certify the independently identifiable part
Objected byCLAIM-206 — If a communicative kind can outrun every probe, human same-kind judgements, themselves one probe family, can neither confirm nor defeat the October thesis; if it cannot, the kind does in the tests only what an observational class does, which the record's term for a kind says it is not
ComplementsCLAIM-050 — For report fidelity, fidelity is relative to the communicative decisions receivers make, and Blackwell's order on rendering procedures, over a designed distribution of situations, makes precise the ceiling a rendering sets on those decisions, CLAIM-149 — A probe family that supplies a transformation's training signal cannot also certify what the transformation preserves: preservation must be shown under probes that were not used to train it, CLAIM-217 — For a communicative kind, the standard of sufficiency is the uptake of competent participants who know the standings and norms in play: a defeasible standard, overridden only where the act's other constituents diverge from the judges' verdicts, so a probe family is tested against it, and a kind outruns the families coarser than that standard and the judges only through those constituents, CLAIM-209 — Probes and statistical models recover only the distinctions identifiable from the relational evidence they are given, and a signature sufficient for a stated family of communicative decisions outside the observed probes can be identified when the full latent factorization is not, CLAIM-213 — A proposed relational kind is a hypothesis about invariance across admissible changes of measurement and context, not an equivalence class under the measurements now available, so realism about kinds gains empirical content as extrapolation to probe families, populations and environments not used to construct them
UsesTERM-014 — communicative object, as an observational equivalence class
Illustrated byCASE-041 — FID and Projected GANs: one ImageNet-trained probe family used both to train a generator and to evaluate it
Cited as “complements” byCLAIM-193 — A linear probe family can tell apart only realizations whose representations differ in normalized kernel, and kernels fix a representation only up to the symmetry of what is observed: any distinction beyond that is supplied by the analyst
Cited as “rests on” byARG-029 — The operational account, kept as a rule of method, supports a realist kind that can outrun observation as the bearer of communicative identity

Graph View

The claim

A158 §5, after the owner’s U45 (“There’s definitely a connection here to Yoneda’s Lemma”), restricted an object’s profile to a selected family of probes: “This restricted profile may no longer distinguish all objects. And that is precisely our fidelity problem.”

A203 §8 made it a premise: “Two objects may be indistinguishable by all probes in 𝒬 without being isomorphic in the full category. This is our central epistemic distinction: Full structural identity and observationally recoverable identity are not the same thing.” A203 §14 drew the evidential rule: “Two realizations may have the same response distributions under every probe in 𝒬. That provides observational equivalence relative to the chosen probes. It does not guarantee complete structural identity … A probe family selected solely to make two representations appear equivalent provides weak evidence of structural preservation. A probe family justified independently and evaluated on held-out cases provides much stronger evidence.” A218 Ch7’s core claim: “A structural kind may be real while remaining indistinguishable under an insufficient family of empirical probes.”

The claim has two halves. The first is that observational equivalence (TERM-014) is relative to the probe family, and may be coarser than structural identity. The second is the rule of evidence, which is CLAIM-011’s remedy applied to probes: only probes justified independently of the comparison, and evaluated on held-out cases, are evidence that structure was preserved.

The exact half

When a restricted family identifies. The restricted Yoneda functor A ↦ Hom(−,A)|_𝒬 is fully faithful exactly when 𝒬 is dense. Examples: the one-point set in Set; the vertex and the edge in graphs. When 𝒬 is only a strong generator (in a category with finite limits), the functor reflects isomorphisms without capturing all morphisms. Example: ℤ in groups, where Hom(ℤ,−) is the underlying set. Any other family identifies objects only up to an equivalence coarser than isomorphism, and that coarser equivalence is what TERM-014 calls an “equivalence class … relative to a chosen family of measurements”. A158 did not name the condition.

What the record’s reading of the lemma adds. THEORY-032, from the nLab page (LIT-221, read in NOTE-195), sets three conditions on it:

  • The profile that identifies an object is the hom-functor with its action by composition, not a list of hom-sets. Two non-isomorphic objects can have hom-sets in bijection for every probe (the ℤ≥0 / ℤ≥1 example). An empirical probing profile, a table of outcomes, is of the bare kind. So the categorical ideal it is said to approximate is a stronger structure than probing records.
  • The lemma needs identities. It fails for semicategories.
  • The lemma presupposes the objects it identifies. y(c) is indexed by every object of the category, and the conclusion is an isomorphism between two of them.

What probe distributions can and cannot certify. A184 §2: (φ_i)#P = (φ_i)#Q for every i “does not generally imply” P = Q, and matching each probe’s marginal leaves the joint across probes unconstrained. FID keeps only two moments of one probe. A181 §5: “a standard Gaussian and a symmetric distribution supported on {−1,+1} have equal mean and variance”, so the population FID between them is zero. But a rich enough family does separate: all one-dimensional linear projections determine a distribution (Cramér–Wold).

What it does not say

  • It does not say a full Yoneda profile exists for communicative acts, or that one is observable. Writing 𝒬 ⊆ 𝒞 presupposes a category 𝒞 of communicative realizations that nobody has specified. A158’s notation Hom_C(Q,A) also puts contexts and communicative objects in one category. A communicative object is a presheaf on contexts, not an object among them, so its restricted profile is its value at each probe.
  • It does not say probabilistic fidelity follows from Yoneda. A173: “Avoid claiming that probabilistic fidelity follows directly from Yoneda’s lemma.”
  • It does not withdraw CLAIM-046’s methodological rule. Evidence still comes from observable responses. What changes is what that evidence is of.
  • A data matrix of probe outcomes is a table of hom-set-like entries without composition. Even a complete one is not a Yoneda profile. The exact theorem for counts from probes is Lovász’s (1967), noted in THEORY-032 and NOTE-195: finite relational structures are isomorphic exactly when they have the same homomorphism counts from every finite structure.
  • It says nothing about contextuality. A158 §4: “Yoneda does not say that every empirical presheaf is representable. Indeed, a contextual empirical model may be precisely the sort of structured collection for which no globally admissible realization exists.” Being representable and having a global section are independent. On the poset of subsets of a set X, the representable y(U) for U ≠ X has no global section. The event presheaf U ↦ O^U is not representable in general, yet it always has global sections. Contextuality is a failure of the empirically constrained local data, supports or distributions, to extend from the cover to a global realization (CLAIM-038), not a failure of representability. Applied to the event presheaf, the lemma restates the definition of a communicative object (TERM-002, now superseded by TERM-042) and adds no constraint of its own.
  • It says nothing about which probes to select. That is CLAIM-050’s question, and QUESTION-031 asks when a family is rich enough.

Note of 2026-10-13: the general content is filed separately

Under ADR-040, probe-relative equivalence and the evidential rule is filed on its own as CLAIM-264, in the formal-apparatus line, resting on no claim. This claim now rests on it for the general content, and keeps CLAIM-046 and CLAIM-011, for the realist reading of communicative identity and the anchored-correspondence remedy. Claims in other lines that need only the general content cite CLAIM-264, so setting pragmatic-transport aside does not take it from them.

Note of 2026-10-19: the strong-generator sentence is corrected in CLAIM-264

The exact half above says a strong generator reflects isomorphisms and that any other family identifies objects only up to a coarser equivalence. The first is right, as conservativity of the restricted functor, but read with the second it implies that a strong generator identifies objects, and it does not: (ℤ/p)³ and the Heisenberg group mod p, for p odd, have isomorphic profiles through ℤ and are not isomorphic. Identifying objects does not need density either. CLAIM-264, version 2, states the three cases apart, and is now Active. This claim’s application is unchanged.

Note of 2026-10-29: what holds it back

Its premises and grounds are held, and the part of CLAIM-206 aimed at it, the defeat condition a one-probe family met, was written into version 2. What the objection still reaches is the realist half: “structurally identical” and “every pair of realizations that judges reliably distinguish” need a standard beyond any analyst’s family, and the record’s argument for one, competent uptake (CLAIM-217), is Proposed and rests on CLAIM-151, which is held back too. CLAIM-217 held is what would promote this claim.

Note of 2026-10-30: what holds it back

Rechecked after CLAIM-151’s version 5 (2026-10-30), which only drops CLAIM-004 from its rests_on; CLAIM-151 is still Proposed. So CLAIM-217, the competent-uptake standard that the realist half of this claim needs against CLAIM-206, still rests on a premise not held, and CLAIM-217 held is still what would promote this claim.

Backlinks