nucleation

CLAIM-407: A coarse-graining exactly informationally closed under a process's ordinary dynamics need not preserve its responses to an intervention on the mechanism: for a declared intervention distribution ρ, independent of the state, the deficiency Γ = I(S; Y^(I) | T(S), I) can be positive, as in the hidden governance bit, where it is ρ(h = 1) bits; and Γ = 0 exactly when, for every intervention ρ charges, the response kernels are constant on T's fibres over the states of positive probability

7 min read

Properties14
StatusActive
Version3
Rolethesis
Warrantdemonstrated
Tagscausality, information-theory, self-governance
Lineformal-apparatus, will-organization
Rests onCLAIM-350 — A coarse-graining preserves a deterministic judgment exactly when the judgment is constant on its fibres: the judgment then factors through the coarse description, uniquely on its image, and otherwise two states the judgment tells apart have been merged
GroundsTHEORY-820 — For a stationary discrete-valued process X and any coarse-graining Z of its past, informational closure at every horizon is the same condition as zero one-step transfer entropy from the micro past, Barnett and Seth's dynamical independence, and as causal closure, Z's ε-machine coinciding with its υ-machine; for a coarse-graining of the present state it also makes Z's causal states a coarse-graining of X's, which does not imply closure, and the closed coarse-grainings are not closed under common refinement, THEORY-424 — Mutual-information measures of individuality that read a system's next state off its present one measure association, not propagation: I(S_{t+1}; S_t) can be positive when the environment alone drives the system, and conditioning on the environment does not recover the interventional quantity, which can be zero where the conditional measure is one bit, CASE-125 — A hidden governance bit: an output that is fresh fair noise whatever a fixed hidden bit K is, until an intervention h makes the output read K, so the output alone is exactly informationally closed under ordinary operation and keeps nothing about the response to h, THEORY-848 — In an acyclic structural equation model, marginalising a variable that no declared intervention sets is an exact transformation, provided the coarse model's exogenous noises may be dependent; so exactness does not require the dropped variable to be causally idle, and it can drop a variable that decides how its children respond to an intervention that is declared
SupportsCLAIM-409 — For one state law, one intervention distribution and one outcome, a deficiency of sufficiency cannot increase under refinement: if T₁ = g∘T₂ then Γ(T₂) ≤ Γ(T₁); informational closure, by contrast, is monotone in neither direction, since in the three-bit chain of Rosas et al. (2024) A is closed, its refinement (A, B) is not and the full state refining that is closed again, while A⊕B, a coarsening of (A, B), is not closed either; closure fails to be monotone because refining a description changes the future it must predict as well as the evidence, so Blackwell dominance, which every deterministic refinement has over its coarsening, does not order closure, CLAIM-399 — Causal abstraction in the senses of Rubenstein et al., Beckers and Halpern, and Beckers, Eberhardt and Halpern does not secure governance-response sufficiency: exact transformation, τ-abstraction and zero approximate-abstraction distance all accept the hidden-bit coarse-graining, which loses ρ(h = 1) bits of the response to h, because each lets high-level noise carry what the coarse state drops; for one outcome and the identity correspondence, sufficiency is their compatibility condition with the high-level context fixed to the coarse state, and strong abstraction rejects the case only through interventions on the hidden parameter
ComplementsCLAIM-414 — Matched ordinary trajectories can conceal differences in an interpretation's efficacy, in the allocation of that efficacy, and in the maintenance of its enabling constraints: in two finite constructions, systems that agree under ordinary operation differ in each, and a specified intervention exposes the difference that passive observation of the matched trajectories does not, CLAIM-450 — Three standards for a communicative distinction come apart: statistical distinguishability of an interpreter's response profiles, measured locally by the Fisher metric and never raised by a parameter-independent garbling; usability by a bounded receiver, which a transformation can raise; and pragmatic efficacy in the act; and where response models are singular, as Watanabe's theory treats, a missing response can come from an absent attempt or an unresponsive receiver alike, and only intervention or later dynamics separate them
UsesTERM-067 — governance-response profile, of a state for a declared family of interventions on the mechanisms that govern efficacy: the tuple of its response kernels, J_I(s) = (K_i(· | s))_{i∈I} with K_i(y | s) = P(Y^(i) = y | S = s); a description is governance-response sufficient for (I, Y) when the profile factors through it, TERM-023 — will-organization
Illustrated byCASE-125 — A hidden governance bit: an output that is fresh fair noise whatever a fixed hidden bit K is, until an intervention h makes the output read K, so the output alone is exactly informationally closed under ordinary operation and keeps nothing about the response to h, CASE-152 — Absent attempt or unresponsive receiver: with Y ∼ N(ab, 1), a the strength of an attempted influence and b the receiver's responsiveness, the settings (a, b) = (0, 1) and (1, 0) give the same response distribution; K(a, b) = a²b²/2, the Fisher matrix has rank at most one and vanishes at the origin, and the real log canonical threshold is ½ with multiplicity 2, not the regular d/2 = 1, while changing b with a held fixed separates the two, CASE-164 — A 192-state organization on which one kernel gives several closure verdicts: the description by agents and phase is informationally closed on the intact support and fails global strong lumpability, its deficiency for the next pair of rule integrities under a declared intervention mixture is 0.55·H₂(1/3) ≈ 0.5051 bits and zero once the maintenance mode is kept, and finite-sample predictive gains misread both a positive and a zero information defect
Cited as “complements” byCLAIM-517 — A successor regulator keeps a viable set positively invariant if, at every boundary context, its response distribution is within γ/L in total variation of a regulator that certifies inward margin γ, the dynamics being L-Lipschitz in that distribution; an average distillation loss certifies this only through coverage of the boundary contexts, so fidelity measured in ordinary operation certifies nothing about viability, CLAIM-522 — Recovery under interventions on state cannot show where maintenance is realized: an outside controller that measures the constraint capacities and supplies the internal production law as state feedback yields the same trajectory under every perturbation of state, damage and recovery included, even with every capacity observed exactly; only interventions on the production mechanisms, relative to a declared boundary, separate the two
Cited as “corroborated by” byCLAIM-122 — Will-organization is not one more higher-order preference: changing the conditions of efficacy changes behaviour while the competing preferences stay fixed

Graph View

The claim

A16 §3: “A representation can be perfectly informationally closed under the ordinary dynamics while failing to preserve a governance-relevant counterfactual distinction.” A17 §4.2, its Proposition 1: “There exists a stationary finite-state stochastic system and a spatial coarse-graining that is exactly informationally closed under its ordinary dynamics, yet fails to preserve the response profile for a specified intervention on a governance mechanism.”

The terms. S is the state, with a law P. I is a family of interventions, drawn by a declared ρ independently of S. Y^(i) is the outcome under intervention i, with response kernel K_i(y | s) = P(Y^(i) = y | S = s). The profile of s is J_I(s) = (K_i(· | s))_{i∈I} (TERM-067). A16 §4’s deficiency of a coarse-graining T = f(S) is

Γ_ρ(T) = I(S; Y^(I) | T(S), I) = H(Y^(I) | T(S), I) − H(Y^(I) | S, I).

The witness (CASE-125). S_t = (K, Y_t), K a fair bit fixed in time, Y_{t+1} fresh fair noise under h = 0 and Y_{t+1} = K under h = 1; T = Y_t.

  • Closed. Under h = 0, Y_{t+1} is independent of every past, so I(S_{≤t}; Y_{t+1} | Y_{≤t}) = 0. By THEORY-820 (a) that is closure at every horizon.
  • Γ = ρ(h = 1). Under h = 0 the response is independent of S, so its term is 0. Under h = 1 it is K, with H(K | Y_t) = 1 bit and H(K | S) = 0. Averaging, Γ(Y_t) = ρ(h = 1), which is 0.5 bit at A16’s ρ = ½.

When Γ = 0. Since I is independent of S, Γ = Σ_i ρ(i) I(S; Y^(i) | T(S)). Each term is non-negative, and zero exactly when Y^(i) is independent of S given T(S), that is, when K_i(· | s) = K_i(· | s′) for all s, s′ of positive probability with T(s) = T(s′). So Γ = 0 is CLAIM-350’s factorization of J_I through T (its probabilistic version), for the interventions ρ charges and on the states of positive probability, and no more. A17’s Appendix A asks for exactly these “support assumptions under which (\Gamma=0) is equivalent to exact response-profile preservation”. With ρ(h = 1) = 0, Γ = 0 for a constant T while the profile does not factor through it.

Checked on 2026-10-21 by exact enumeration of the joint law of (K, Y_t, h, Y_{t+1}): closure 0 at one to three steps of history; Γ(Y_t) = Γ(constant) = 0, 0.25, 0.5 and 1 bit for ρ(h = 1) = 0, ¼, ½ and 1; Γ(K) = Γ((K, Y_t)) = 0 throughout; and the ρ = 0 case above.

Where it was before

  • CLAIM-350’s “What it does not say”: “That observational adequacy is causal adequacy. A state space with a distribution can satisfy the probabilistic version while interventions tell its merged states apart (A22’s Construction B; CLAIM-127).” Its causal bullet defines intervention-consistent coarse-graining “for a declared family of interventions” as a definition, not a theorem. That is this claim’s qualitative point, on file since the exchange of 2026-10-17.
  • CLAIM-122’s note of 2026-10-18. A7 §5’s supervisory test and A12’s Proposition 2 of 2026-10-17 (“Construct two systems whose default action trajectories are identical”), of which the note says “No parameters were given.” A15 §1 of 2026-10-20 cited the test, and A15 §2 built it as a counterexample. A17 then listed it as “Established by the paper’s self-contained finite counterexample” without the note.
  • The causal-abstraction literature, read on 2026-10-21 (NOTE-1037, NOTE-1032, NOTE-1045). It has relativity to a declared intervention set (Rubenstein et al. §5; Beckers and Halpern’s I_L), which is this claim’s I; the support caveat (Beckers and Halpern’s Examples 3.4 and 3.11 leave behaviour off the support unconstrained), which is this claim’s restriction to the interventions ρ charges and the states of positive probability; and a macro variable merging micro-states with different interventional consequences (Rubenstein et al.’s cholesterol case). It does not have this claim: its criteria compare interventional laws, and exact transformation, τ-abstraction and zero approximate distance all accept the hidden-bit coarse-graining (THEORY-848, THEORY-849, THEORY-843). How sufficiency sits beside them is CLAIM-399.
  • An instance of the shape outside the record. Szeidl and Szucs’s model of media capture (THEORY-860, NOTE-1039): the observed slant gap can be the same whether it comes from owners’ ideology or from favor exchange, and a ban on targeted government advertising, an intervention on the mechanism, tells them apart (THEORY-852). The join is the record’s; the paper says nothing of closure or of Γ.

What it does not say

  • Nothing new as information theory. A17: “This is a deliberately restricted application of ordinary causal sufficiency: the interventions have been selected because they concern governance. We should not advertise it as a wholly new mathematical concept.”
  • Not that Γ can be had from passive data. A16: “Ordinary passive data may not identify (\Gamma).”
  • Not a contradiction of THEORY-820, which says closure is “Not interventional causation unless X is fully observed”.
  • Not a measure of will, or a showing of agency (CLAIM-116). A16: “Nor does the example establish agency. (K) is simply a hidden mechanism parameter.”
  • Not that the system governs itself. The profile records how the system would respond if its governance were set from outside, P(Y | do(G = g)). A27 §3 of the same exchange: “That is not the same as establishing that the system can alter the mechanism itself. An ordinary machine can respond dramatically to a technician rewiring it.” Whether the system produces its own transitions of governance is CLAIM-174’s third condition (its policy would adapt if its actions’ consequences changed), not something Γ measures. A15 §6 had the same limit: identifying the causal mechanism of governance “does not settle whether its authority is endogenous, historically uncoerced, contestable, or normatively justified.” Neither establishes ownership (CLAIM-022, QUESTION-071).
  • Not an instance of causal abstraction, nor a new theory of it. A25: “We should not present governance-preserving coarse-graining as a new general theory of causal abstraction.” That stands, and the papers’ notions do not imply the condition for Γ = 0 (CLAIM-399).
  • Not a test of CLAIM-336. The construction has no meaning dynamics.
  • Not one scale across regimes. A16 §5: “one must not compare deficiencies numerically across different intervention distributions or outcome targets as though they were measurements on a common scale.”

Note of 2026-10-26: a second exact witness

CASE-164 is a second exact witness, built separately from CASE-125 (dmarx/PerspectiveLab@7e2edc2 experiments/finite_organization/). In a 192-state organization the maintenance mode plays the hidden bit: the intact dynamics are the same in all three modes. The agents with phase are informationally closed on the declared intact law. Their Γ for the next pair of rule integrities, under a declared mixture that gives ordinary operation 0.45, is 0.55·H₂(1/3) = 0.5051 bits, since each other intervention splits the three modes one against two. With the mode kept Γ is 0, the claim’s zero condition.

Two of the claim’s cautions show in the construction. The closure is on a declared support of a reducible chain and is not global strong lumpability. And Γ moves with the source law: for the ordinary intervention alone it is 0 on the intact support and 0.918 bits on the damaged one. v1.9 (dmarx/PerspectiveLab@9f82bfd) estimates Γ from sampled intervention outcomes, as held-out log-loss: 0.7718 bits at 300 windows and 0.5103 at 19,200, against 0.5051. Small samples overstate it. The claim is unchanged.

Note of 2026-11-01: a reduced circadian model as a possible instance

Bechtel and Abrahamsen (LIT-1936, NOTE-1492) report two models of the Drosophila clock.

  • Leloup and Goldbeter model light as a doubling of TIM’s degradation rate, which reproduces phase shifts (p. 325).
  • Smolen et al.’s reduced model merges PER and TIM into one variable and still reproduces the 24-hour oscillation (p. 326).

A coarse-graining that merges the two is adequate to the ordinary dynamics. It may not preserve the response to an intervention that acts on TIM alone, since it has no variable for light to act on separately. That is this claim’s structure: ordinary adequacy without preservation of the response to an intervention on the mechanism. Same structure, not shown same mechanism. Neither the paper nor the models compute anything like Γ, and whether the reduced model mispredicts light responses is not reported.

No status, title or warrant changes.

Backlinks