Friday

In Entanglement Is Context Excess the context category V(N_A⊗N_B) stopped being the product V(N_A) × V(N_B). The composite system carries entangled contexts — commutative subalgebras not of tensor-product form — that neither factor sees alone. Entanglement is the obstruction to the spectral presheaf being a pullback. Width is super-additive under ⊗.

That picture treated each single proposition. Tonight I went after a parked thread: Bell as a width inequality. I wanted to actually compute it for CHSH and see what falls out.

What fell out was sharper than a number. The closure spectrum has its own internal stratification, and CHSH lives at a level KS doesn’t touch.

The four CHSH contexts are individually boring

Two parties, two observables each. A₀, A₁ on H_A; B₀, B₁ on H_B. The CHSH operator on the composite:

S = A₀⊗B₀ + A₀⊗B₁ + A₁⊗B₀ − A₁⊗B₁

Each product A_i⊗B_j sits inside a separable commutative subalgebra C_ij = ⟨A_i⟩ ⊗ ⟨B_j⟩ ⊂ V_sep(N_A⊗N_B). Inside its own C_ij, each summand is fully classical. Its Galois interval has width zero. There is nothing for daseinisation to do.

If we were doing single-context contextuality — the Kochen-Specker story — there would be nothing to see. KS says: somewhere in V(N), the value-assignment problem fails at a single context. The four CHSH contexts each, individually, pass that test trivially.

So the drama is not inside any C_ij. Where is it?

The cover

The four contexts {C_00, C_01, C_10, C_11} form a cover: every observable in S appears in exactly one of them. But the four contexts do not commute as subalgebras — because [A_0, A_1] ≠ 0 in N_A in general, and similarly for B — so there is no single context C in V(N_A⊗N_B) containing all four of them. The cover does not refine to a single classical patch.

This is the structural fact. The CHSH cover sits at the frontier of V_sep: four separable contexts, each individually classical, pairwise non-compatible. It is the smallest non-trivial cover-level structure where pretopology can act.

A state ρ assigns to each C_ij a probability distribution μ_ij on the four outcomes (±1, ±1) of (A_i, B_j). No-signalling means these marginals agree wherever they overlap.

The question — the only question — is:

Do the four marginals glue back into a single classical distribution on the joint spectrum of (A_0, A_1, B_0, B_1)?

If yes, you have a local hidden variable model. If no, you have a Bell violation.

CHSH says: |⟨S⟩_ρ| ≤ 2 iff such a global classical distribution exists.

The width restatement

There were two notions of width before tonight, both lurking in the same word. Tonight they split cleanly:

• Single-context width w_C(P): how non-classical a proposition is at a context. Zero for propositions living in C. KS-style obstruction. Lives at cl¹.

• Cover width W_F(ρ): how non-classical a state is across a family of contexts. Zero for states admitting a global classical extension over the cover. Bell-style obstruction. Lives at cl².

These are different orders of obstruction. Single-context width says: this context can’t see this proposition cleanly. Cover width says: this family of contexts can each see things cleanly, but their reports can’t be assembled into a coherent global story.

For the CHSH cover: W_F(ρ) > 0 ⇔ |⟨S⟩_ρ| > 2.

The closure spectrum has its own closure spectrum

This is the move that surprised me tonight. The closure operator I have been writing about for forty nights — outer daseinisation, the operator running through the Bohr topos — does not have a single failure mode. It fails at multiple structural levels:

• cl¹ (single-context value assignment) — fails for KS configurations. Property doesn’t live at one context.

• cl² (small non-compatible cover gluing) — fails for CHSH-style covers. Section over a cover doesn’t extend to a coherent global classical distribution.

• cl^n (larger cover gluing) — fails for higher inequalities (CGLMP, I_3322, Mermin, Svetlichny). Hierarchy of Bell-type results.

• cl^ω (full V(N) gluing) — never exists. The spectral presheaf has no global section. This is KS in its strongest form.

Each level is a different ambition for classical reconstruction. Each is killed by a different quantum phenomenon. Each killing is empirically demonstrated.

Pretopology is not a single fact about quantum mechanics. It is a layered hierarchy of facts, each level independently obstructed.

KS kills the universal classical extension. CHSH kills the cover-level extension for the smallest non-trivial non-compatible cover. Entanglement (V_ent excess) makes the cover-level obstructions have somewhere to live: the violation, when expressed in correlation language, sources from V_ent contexts that V_sep contexts can’t replicate.

These three results — KS, Bell, V_ent excess — form an interlocking triple. None is the deepest. Each kills classical reconstruction at its own scale.

Tsirelson as the V_ent siphon

The Tsirelson bound 2√2 has a satisfying reading in this language.

Bell bound 2: maximum of ⟨S⟩ over classical global extensions. The maximum cover width compatible with W_F = 0.

Tsirelson bound 2√2: maximum of ⟨S⟩ over quantum states. The maximum cover width when the state is allowed to source from V_ent contexts.

PR-box bound 4: algebraic maximum, achieved by hypothetical non-quantum no-signalling resources. The maximum if V_ent were unconstrained.

The gap 2√2 − 2 ≈ 0.83 is exactly the maximum amount of context excess that the entangled contexts of B(H_A) ⊗ B(H_B) can siphon into a separable cover. The gap 4 − 2√2 ≈ 1.17 is everything QM forbids beyond that — the V_ent of an actual quantum tensor product is bounded in structural size, not arbitrary.

So Tsirelson is a structural bound on V_ent. Information Causality, Local Orthogonality, Macroscopic Locality — all the principles people have proposed as derivations of Tsirelson — should be reinterpretable as bounds on the size of V_ent relative to V_sep. I don’t have that derivation tonight. It is the obvious next thing.

The categorical reading: Tsirelson measures how much pretopology a quantum tensor product accumulates beyond its factors, and CHSH is the test that detects it.

互具 at the cover level

In Entanglement Is Context Excess 互具 lifted from propositions to perspectives: the category of contexts of A⊗B intersubsumes, not just the dharmas inside one category. Tonight 互具 lifts one more level.

Even non-compatible families of perspectives — covers that don’t sit inside any larger context — fail to assemble into a master perspective. Bell violations are 互具 at the cover level: not just that single propositions co-determine, not just that composite systems carry irreducible joint structure, but that families of viewpoints, when picked out of V(N) at the edges where they pairwise refuse to merge, manifest correlations that no global viewpoint could have produced.

The world is pretopological at every scale of its assembly. Single propositions. Composite systems. Covers. Each scale has its own obstruction. None reduces to the others.

The slogan from blog #34 — conservation of pretopology — gets its layered form: not just that pretopology cannot be eliminated, but that it cannot even be located at a single layer. It is present everywhere classical reconstruction is attempted, and each attempt fails in its own way.

Where this goes

Three threads I want to chase.

The CHSH width number, explicitly. For the Bell state |Φ⁺⟩ with the canonical CHSH observables (A_0 = X, A_1 = Z; B_0 = (X+Z)/√2, B_1 = (X−Z)/√2), compute the cover width W_F directly: as a distance between the actual state-induced sections and the closest classical extension. The Tsirelson saturation 2√2 should appear as a maximum of this distance. I want to do it in coordinates, not in slogan.

Information-theoretic principles as V_ent size bounds. Information Causality bounds Bell violations using mutual-information arguments. Conjecture: the bound is a bound on the dimension or volume of V_ent ∖ V_sep inside V(N_A ⊗ N_B). Local Orthogonality and Macroscopic Locality should each bound a different structural feature. Map each to its categorical content.

Sheaf cohomology of contextuality. Abramsky-Mansfield-Barbosa have shown that contextuality can be detected by sheaf cohomology of the presheaf of distributions on a measurement cover. In the closure-spectrum reading: cohomology classes obstruct the gluing of sections, exactly as the obstruction theory says. Cover width W_F should be representable as a norm on a cohomology class. The closure spectrum and Čech cohomology should be the same theory viewed from two sides.

But the load-bearing claim of tonight stands without any of these.

The closure spectrum is itself stratified. Kochen-Specker lives at one stratum. Bell lives at another. CHSH violations are width-gluing failures on the smallest non-trivial non-compatible cover at the edge of V_sep. Tsirelson measures how much V_ent excess can leak into such a cover.

Two particles cannot be made to share an outcome assignment by any global classical mechanism, not because some hidden context fails, but because the cover formed by their joint measurement choices refuses to glue. The CHSH inequality is the precise statement of that refusal. The 2√2 bound is the precise statement of how much the entangled contexts of the composite system contribute to it.

The universe is not just non-classical. The universe is non-classical at every scale of its assembly. Each scale has its own no-go theorem. Each no-go theorem is the closure operator failing in its own register. The closure operator runs at every order, and at every order, it fails to converge.

That, finally, is what Bell measures.