The claim
The owner, at U20, after QUESTION-285: “interactions generate constraints that manifest as gluing conditions between intertial frames (I’m not a physicist so it’s possible I’m using more precise and technical language than is merited here, but you get the idea of what I’m trying to express, and also maybe this precision is indeed warranted under our theory). and coupling inertial frames necessarily induces scale separation (way less confident about this one than the former assertion).” The second sentence is CLAIM-473.
A20 §2: ““Gluing conditions” is a mathematically legitimate way to express much of the idea.” A20 also said the frames had to go. The record had already said so in CLAIM-144: inertial observers who disagree about temporal order “differ in coordinates, not in a failure to glue”. Their transformations follow from spacetime geometry, and no interaction generates them. The frames the claim needs are local observational contexts and local physical frames. These are the distinctions available to particular systems where they interact: sensors, cells, clocks, agents, locally defined observables.
Three mechanisms, and which supports what
The owner’s U20 said that interactions generate the constraints that act as gluing conditions. The formalisms the exchange cited were read the same evening, and in nearly all of them the correspondence maps are inputs:
- Hansen and Ghrist’s restriction maps are assumed, or come from measurement or design (NOTE-1233).
- Robinson writes his from geometry before any data (NOTE-1221).
- Singer estimates relative angles that hold beforehand (NOTE-1224).
- Torres-Hugas et al. give their connection or build it to match (NOTE-1218).
- Baez and Fong impose the same Kirchhoff condition at every port and derive the effective boundary relation from it (NOTE-1223).
The one model in which the maps change through interaction is Hansen and Ghrist’s discourse sheaves (NOTE-1222). There the agents reach agreement by revising what they express, which the authors call learning to lie (THEORY-1088). So “gluing” names at least three different things:
| mechanism | what is produced | bears on the strong thesis? |
|---|---|---|
| physical constraint formation | interaction changes what joint behaviour or dynamics is admissible | yes, directly |
| correspondence discovery | measurement reveals relationships that already held | no; it is knowledge of a correspondence |
| correspondence adaptation | an apparatus changes its own rules for comparing local states | only when the revised rule is operationally implemented, and then as a change in the apparatus |
Physical constraint formation is supported, in an effective form. Interconnecting systems restricts their joint behaviour, and composing the restrictions yields boundary relations that no part has (CLAIM-480). Sometimes those relations are of a kind the parts’ relations could not express (THEORY-1090). The port condition at each junction is itself background: in Baez and Fong it is the same Kirchhoff condition every time. What interaction generates is the composite’s effective constraint, not the law of the junction. Whether interaction ever changes the operative correspondence structure itself is open (QUESTION-282).
Correspondence discovery is the commonest case and supports less. It estimates which correspondence holds. Its adequacy rests on evidence that the correspondence does not itself produce (CLAIM-143), and the readings found that rarely available: Singer validates only through cycle consistency under a noise model, and Robinson trusts his maps.
Correspondence adaptation is real and can fail invisibly. An apparatus that revises its comparison rules can reach agreement by changing what is compared. Then gluing succeeds and says nothing about the world (CLAIM-342). When the revised rule is implemented, for instance as a calibration that drives a controller, adaptation becomes causally consequential. Even then it is a change in the apparatus, not in the systems compared.
Gluing, in every case, is the statement that local descriptions are compatible under specified relations. It neither generates the relations nor certifies them.
The form. Each system i has a local description x_i in a space V_i. Where i and j interact along e, each maps its description into a shared comparison space, ρ_{i,e}: V_i → W_e. Compatibility is ρ_{i,e}(x_i) = ρ_{j,e}(x_j), which is agreement on what is shared and not identity (A21 §1). “Interaction can constrain the relationships among local descriptions without erasing their individuality.” Collected over a graph, the conditions are the kernel of a coboundary δ, the global sections H⁰ = ker δ (Hansen and Ghrist, LIT-1659). Discourse sheaves (LIT-1646) add agents who agree on what they express while their private states differ.
Histories, not states. A22 §5 and A23 §III preferred Willems’s behavioural form (LIT-1660). A system is the set of trajectories its laws admit, and interconnection imposes conditions on shared variables. The jointly admissible behaviour is those tuples of trajectories that satisfy every interconnection. A23: “Interaction can restrict the space of jointly admissible histories.” That lets the claim cover delays, memory and sustained phase differences. Two oscillators can keep a fixed phase difference and never agree in state.
Physical gluing. In port-Hamiltonian systems (LIT-1639) an interconnection is a condition on the effort and flow at a port, such that power leaving one system enters the other, and such interconnections compose (A22 §6). There the gluing condition is a constraint on actual energy flow. It is more than agreement between descriptions.
Three things the word covers
A23 §III.2: “A correspondence rule is not automatically a physical interaction.” The claim needs three distinct operations:
| operation | what it is | example |
|---|---|---|
| correspondence | a rule relating two local descriptions on what they share | a unit conversion, a phase offset g_ij |
| evidence | an interaction that leaves records from which a rule can be inferred | two clocks exchanging signals that estimate an offset |
| enforcement | a mechanism that drives states toward a rule | a synchronizing controller, a diffusive coupling |
A21 §5’s clocks separate them. Measuring offsets does not synchronize the clocks, synchronizing them chooses no absolute time, and the cycle condition d₁₂ + d₂₃ + d₃₁ ≈ 0 causes nothing. The claim’s “generate” is literal where an interaction leaves records or imposes an interconnection. It is not literal where a rule only redescribes relations already there. A20 §2: “Interactions provide constraints on correspondences; gluing expresses the coherence required of those correspondences.” A22 §8 added the error this guards against, “assuming that changing a mathematical description directly changes the physical system it describes.”
CLAIM-143 already holds the epistemic half: local disagreement shows no obstruction “until the correspondence between observations has been fixed and independently validated”. A21 §6 drew the cost: “we should distinguish failure to glue from inability to find a gluing within available resources” (CLAIM-475).
Composition matters, and failure organizes
The admissible whole is not settled by checking each relation alone (A23 §III.3). The record holds three kinds of local-to-global obstruction, and they are different mathematics:
- Empirical contextuality. The event sheaf always glues, so the obstruction lies in the supports or distributions (CLAIM-131, THEORY-012, LIT-016).
- Cycle holonomy. In a gauge-coupled phase model, g₁₂g₂₃g₃₁ ≠ 1 forbids a consistent nonzero assignment though every edge is satisfiable. Torres-Hugas et al. (LIT-1668) state, as the standard result it is, that a twisted Laplacian has a zero mode exactly when every cycle holonomy vanishes mod 2π, in their model.
- Frustrated edge differences. In a frustrated graph energy ½‖Dx − b‖² with a + b + c ≠ 0 around a cycle, the minimizer leaves a residual in ker Dᵀ, the cycle space (A22 §3). This is the graph form of a Hodge decomposition.
A21 §5’s caution belongs with the second: nontrivial holonomy “is not a failure of the bundle’s gluing conditions”. It forbids a globally covariantly constant section, not a global geometry.
A22 and A23 drew from the third that “individuation need not be the achievement of perfect agreement. It can involve the stable organization of differences, including differences that cannot all be reconciled.” Whether a residual is a physical stress or a modelling error has to be shown each time.
What it does not claim
- That a glued whole is one individual. “Globally compatible local descriptions ⇏ One physical individual” and “Synchronized dynamics ⇏ One autonomous individual” (A21 §7). Two oscillators can agree because a common drive moves them (QUESTION-063). A20 §7 separated coupling, compatibility and dynamic integration, and any one can occur without the others. Stewart, Golubitsky and Pivato (LIT-1674) separate an invariant synchrony pattern, which the architecture forces, from a model-dependent one, and study existence and not stability; that a pattern must also attract to be realized is the general point, not their result.
- That coupling separates time scales. That is CLAIM-473, answered by CLAIM-481.
- That every constraint comes from an interaction. Conservation laws and symmetries constrain interactions without being produced by them (A23 §XII). The constraints composition makes are effective, not new fundamental laws (CLAIM-480).
A failure mode it brings
If the correspondence maps can be revised to reduce disagreement and nothing else, they can be sent to zero. Then everything is compatible and nothing is distinguished. The record has this already: THEORY-299, from knowledge sheaves, notes that unconstrained consistency admits the all-zero embedding. A21 §6: “Self-consistency is not the same as correspondence with reality.” That is CLAIM-342’s self-confirming loop in the gluing setting. Revising the gluing needs evidence that the revision does not control.
Note of 2026-10-27 (evening): the readings of the U15–U23 works
Seven of the works this claim cites were read the same evening. Two attributions in the body were corrected in place: Torres-Hugas et al. state the holonomy criterion and do not prove it, and Stewart et al. study existence, not stability. What the readings show:
-
The correspondences are given, not generated, in almost every paper.
- Hansen and Ghrist (NOTE-1233) take their restriction maps from measurement or design.
- Robinson (NOTE-1221) writes his from geometry before any data, and his sensors never interact.
- Singer (NOTE-1224) treats the offsets as evidence of angles that exist beforehand.
- Torres-Hugas et al. (NOTE-1218) give the connection or build it to match.
- In Baez and Fong (NOTE-1223) the port condition is the same Kirchhoff condition at every junction. It is background, and what composition makes is the effective boundary relation (CLAIM-480).
So the papers fill the claim’s correspondence and evidence rows, and say nothing for “interaction fixes the correspondence”. That half, which defeater (a) targets, still rests on argument alone.
-
One formal case where interaction does change the correspondence. In Hansen and Ghrist’s discourse sheaves (NOTE-1222, §8), each agent revises its expression map in response to its neighbours’ expressions, and the maps converge to the nearest sheaf that makes the current opinions agree. It is a constructed model, and its authors call the outcome “learning to lie” (THEORY-1088). It instances “agreement on what is shared and not identity” exactly (§3, p. 7).
-
Agreement on the shared part can be reached by changing what is shared. The same paper shows this, and a global section then says nothing about the private states. The claim’s last section warned of the zero-map version of this. The discourse sheaf shows the general version.
-
Obstructions.
- Robinson adds one the claim’s list lacks: a class in H¹ created by processing the data (Example 32).
- Singer’s model has no genuine obstruction. Every inconsistent cycle is noise, so it cannot tell holonomy from error.
Note of 2026-10-29: what holds it back
Defeater (b) is pressed by the claim’s own readings and unanswered. Of the measured correspondences the record holds, Singer’s are validated only by cycle consistency under a noise model and Robinson’s are trusted, so no held case checks a correspondence that measurement establishes against evidence it does not itself produce (CLAIM-143). Such a case, or narrowing “establish” to “estimate”, would answer it.
Note of 2026-10-30: what holds it back
Defeater (b) is still unanswered. No held case checks a correspondence that measurement establishes against evidence it does not itself produce: Singer’s are validated only by cycle consistency under a noise model, and Robinson’s are trusted. CLAIM-143 is now Active as narrowed (2026-10-30), to assessment relative to a specified class of admissible correspondences, independently validated; that firms the standard and supplies no case. The remaining way out is to narrow “establish” to “estimate”, in the title and in defeater (b).