Co-Generation & Hidden-Graph Expansion

Toy in which a projector onto a chosen spectral band is used to select which pairs of sites can talk, with the ordinary local graph left untouched.

Band selection does change which pairs are coupled, but every version we could write required the projector itself to be defined using knowledge of BOTH endpoints. The apparent nonlocality lives in the definition, not in the dynamics.

Lesson · A projector that already knows both endpoints is a restatement of the answer.

Compare a purely local retarded-response chain against the Phase 143 hierarchical access path, counting response accumulation rather than idealized hops.

The chain accumulates retardation linearly in emergent separation. The hierarchy accumulates it logarithmically in hop count, but each hop carries a response delay that does not shrink, so the hierarchical advantage is real only if per-hop response time stays bounded as the hierarchy is built.

Lesson · Log-depth is only cheap if the depth is made of cheap steps. Nobody has shown the steps are cheap.

Formalize the observation that all previous scaling arguments quietly assumed the arriving signal is still usable.

Transfer-time scaling is meaningless without a fixed-fidelity constraint. Amplitude suppression along a hierarchical or off-resonant path can be exponential in depth, which converts an apparently logarithmic route into an exponentially weak one.

Lesson · Speed at zero amplitude is not speed.

Ask what must be specified, in bits, to pick out one destination among N sites in the substrate.

Selecting one destination out of N requires log2 N bits of address, and that information has to be physically carried, stored, or encoded somewhere in the control field. No relational reformulation removes the counting argument.

Lesson · You cannot reach a specific 'there' without paying for the word 'there'.

Separate the clock into two pieces: the time to RESOLVE the log2 N address inside the substrate, and the time to TRANSFER once resolved.

Every toy routing scheme we could construct pushes cost from one piece into the other. Fast transfer demands a pre-resolved address, which is a prepayment; on-the-fly resolution reintroduces a walk whose length tracks emergent separation.

Lesson · Resolution and transfer are the same bill split into two invoices.

Restate the Lieb–Robinson bound in the programme's own language, as the ordinary-physics wall this whole search is testing itself against.

For a local, bounded-norm Hamiltonian, correlations outside an effective light cone are exponentially suppressed. At FIXED signal strength this is an operational speed limit. Nothing in Phases 128–157 violates it, and no result here is claimed to.

Lesson · Established physics. Our entire hypothesis space lives strictly inside this bound unless a locality projector is demonstrated, and none is.

Let links be dynamical degrees of freedom driven only by relational invariants, with no coordinates and no named endpoints, and see whether a link to a chosen far site can form.

Links form, but the microscopic rule is symmetric under relabelling of all candidate partners, so the toy grows links to a SET of relationally equivalent sites, never to one chosen destination. Breaking that symmetry required injecting the target's identity by hand.

Lesson · Coordinate-free rules are democratic. Democracy cannot address a letter.

Treat the ordinary phase as a code subspace of a larger relational Hilbert space, with identity labels and geometric labels carried by different factors.

The construction cleanly separates what a thing IS from where it is relationally, which is exactly the split the Identity Preservation Hypothesis wants. But the separation is a choice of factorization, and a different, equally valid factorization scrambles it. No dynamics selects the split.

Lesson · A tensor-product structure you picked is not a law of nature.

Two sectors: a visible local sector and a richly connected hidden sector, coupled weakly, following Phase 144's tradeoff into an explicit two-sector toy.

For the hidden sector to be useful it must persist — it cannot be created on demand, because creating it is itself a distance-spanning construction with linear cost. So the hypothesis now REQUIRES a permanently existing hidden relational sector, which is a much stronger and much more falsifiable claim than anything earlier in this programme.

Lesson · The hypothesis got harder on purpose. A hidden sector that must already exist everywhere is the kind of claim that can be argued against on ordinary physical grounds.

Ask how a permanently present hidden sector could stay dark to all ordinary probes yet remain accessible to a chosen control.

The only structurally honest option found is symmetry-protected coupling: an exact symmetry forbids ordinary matrix elements, and the control operates by breaking that symmetry locally. This does not remove the tradeoff, it renames it — the symmetry-breaking control is now the thing that must be paid for and bounded.

Lesson · Invisible and useful at once is the whole difficulty, not a detail of it.

Count what a K-degree factorized substrate can address, then count what a general pair-selective interaction costs to specify.

A K-degree factorized control can label 2^K distinct destinations, so address capacity is exponentially cheap. But a GENERIC pair-selective interaction over N = 2^K sites needs O(N^2) = O(4^K) independent parameters. Address capacity is not the bottleneck. The interaction law is.

Lesson · Naming 2^K places is easy. Coupling to exactly one of them, selectively, is the wall.

Give every site a relational charge / type Q_i and a local accessibility variable s_i. Allow the alternate (hidden) interaction to act only between compatible partners, with s_i switchable by strictly endpoint-local control.

Compatibility matching removes the two failure modes that killed the dense hidden sector: with O(1) compatible partners per site the hidden adjacency is not extensive in N, so it does not add O(N^2) interaction terms and does not suffer the 1/N suppression that a democratically shared coupling budget imposes on every pair. Sparse hidden adjacency is therefore the only surviving structural shape.

Lesson · Selection rules decide which matrix elements are ALLOWED to be nonzero. They never create a matrix element. Compatibility is a filter, not a mechanism.

Ask whether a superselection rule can both hide the sector perfectly and act as the switchable local control that opens it. Then price the softer alternative: resonant matching over K binary relational labels with an additive mismatch penalty dU per mismatched label.

A true superselection rule gives exact darkness, but exactness is the problem: an exactly conserved sector cannot also be a freely switchable local control, because switching it is precisely the violation the rule forbids. The honest split is to keep Q strictly conserved and make only the accessibility variable s switchable. The resonant-matching fallback then fails on counting: if all 2^K wrong candidates are genuine off-resonant channels, the leakage proxy is L = (g/U)^2 · sum_{d=1..K} C(K,d)/d^2, whose binomial sum grows like 2^K/poly(K).

Lesson · State-space capacity is not interaction-channel capacity. A viable hidden sector must have fundamentally sparse interaction SUPPORT — exponentially many weak channels are not a substitute for a few real ones.

Suppose fusion rules, charge conjugation, or a unique dual object pick out a compatible partner TYPE. Ask whether that is enough to reach one physical partner when m identical compatible copies exist.

It is not. A type-only, permutation-invariant coupling is H_int = g|A><S| + h.c. with |S> = (1/sqrt(m)) sum_j |B_j>. Optimal driving transfers the excitation into the symmetric superposition |S> with full amplitude, but the probability of finding it at any specific B_j is at most 1/m. Selecting one instance requires symmetry-breaking instance-level relational structure: roughly log2 m bits, or equivalent physical structure carried by the control.

Lesson · Unique dual or fusion objects do not solve partner selection once multiplicities exist. A perfect type match still addresses a crowd.

Price the cheapest honest mediator: M linear mediator modes coupled to N sites, giving an effective site–site coupling J = G D G† after eliminating the mediator.

rank(J) ≤ M, unconditionally. A pairing pattern that demands R independent clean channels therefore demands at least R mediator modes. Logarithmically many linear mediator modes (M ~ log N) cannot generate exponentially many independent channels: whatever is asked beyond rank M reappears as crosstalk between the channels, and removing that crosstalk costs exactly the mediator resources that were supposed to be saved.

Lesson · Rank is a conserved currency in linear mediation. You can move where you pay it, never how much.

Try to buy selectivity with a smooth polynomial kernel over binary addresses instead of a rank-N matrix: coupling J(d) = (1 − d/K)^p as a function of Hamming distance d between K-bit labels, budgeting total squared leakage into all wrong partners at ≤ 1%.

The required exponent p grows several times faster than K: K = 16 needs p = 58, K = 32 needs p = 128, K = 64 needs p = 279, K = 128 needs p = 604. Since the polynomial order is the interaction order of the induced feature map, a degree-p kernel over K bits recreates precisely the high-order many-body complexity the compression was meant to avoid. The smoothness does not buy selectivity; it only relabels the exponential.

Lesson · Smooth, low-parameter functions of an address are not automatically low-complexity interactions. Selectivity has an order cost.

Ask the opposite question: does ANY object exist with degree 1, full rank, and an O(K)-sized description? Take labels q ∈ Z_2^K and the partner map π(q) = q XOR a for a fixed mask a.

It exists, mathematically. π is an involution (π² = I), its adjacency is a fixed-point-free permutation matrix, every site has degree exactly 1, edge strength is O(1), and the rule is described by the K bits of a. Explicit verification at K = 6, N = 64: every degree = 1, rank = 64, P² = I, spectrum exactly 32 eigenvalues +1 and 32 eigenvalues −1. KEY RESULT: rank N does NOT imply an N-sized description table. A full-rank pairing can have an O(K) = O(log N) description.

Lesson · The first structurally clean object in this sequence. It is an algebraic existence result about permutation matrices, not a physical mechanism, and it supplies no matrix element.

Charge the Phase 169 object its physical bill. Ask (a) whether a compact algebraic generator is a bounded-locality elementary operator, and (b) whether moving amplitude between Hilbert-space labels is the same as influencing an independent physical subsystem.

Both answers are no. (a) Implemented on K qubits, a generic XOR mask a is a Pauli string of weight O(K) — a simultaneous many-body operator, compact to WRITE DOWN but not local to ENACT. Description length and operator locality are different quantities, and only the second is charged by physics. (b) Transferring a state between two basis labels of one register is not causal influence between two independent subsystems: the labels must first be shown to correspond to separated physical degrees of freedom, which the toy never establishes.

Lesson · Candidate frontier, hypothesis only: emergent sites as representations of a deeper algebra carrying an ordinary-local generator plus a normally inaccessible involutive generator. Nothing here says such an algebra exists.

A conceptual research note, opened deliberately as method rather than result: every gate so far assumed spacetime, particles, fields, locality and a causal graph as primitives. Phase 171 asks what happens if the primitives themselves are the search space.

New programme statement: instead of assuming the standard primitives, search over candidate GENERATIVE RULE SETS whose emergent observables reproduce known physics in the appropriate limit. The organising question becomes: what minimal primitive, or minimal set of primitives, could generate observer, geometry, matter, causal accessibility and measurement TOGETHER, rather than requiring each to be posited separately?

Lesson · Discovery requires proposing new primitives before they can be tested — but proposing is not evidence. Imagination opens the space; falsification decides what stays.

Nothing exists but relations between events; objects are stable patterns of relation.

Falsifiable demand · Must recover a metric with the right dimension and signature in a coarse-grained limit, and forbid at least one relational configuration observed nowhere.

The primitive act is drawing a difference; all structure is iterated distinction.

Falsifiable demand · Must predict a bound on how many distinguishable states a bounded region can carry, comparable to established entropy bounds.

Rules of change are primitive; states are fixed points and orbits of transformations.

Falsifiable demand · Must generate a conserved quantity by symmetry and identify it with an observed conservation law.

Reality is what survives a set of mutual consistency constraints; dynamics is constraint propagation.

Falsifiable demand · Must forbid a class of configurations that a purely dynamical account would permit, and be checkable against known selection rules.

An operator algebra with no prior spacetime; geometry and matter are both representations of it.

Falsifiable demand · Must reproduce ordinary locality as an emergent property of the algebra and name the limit in which Lieb–Robinson behaviour appears.

What is accessible is primitive; geometry is the bookkeeping of accessibility between observers.

Falsifiable demand · Must derive an accessibility bound that coincides numerically with a light-cone bound where the two overlap, and differ somewhere measurable.

Start from the smallest primitive trio the Phase 171 admission rule allows: DISTINCTION (things can differ), CONSISTENCY (differences must agree locally), ACCESSIBILITY (some pairs can act on each other). No coordinates, no metric, no dimension. Distance is defined only after the fact, as shortest-path length in the accessibility relation that the optimization produces.

The first N = 36 optimization collapsed straight into a small-world blob: mean undirected degree 8.111, diameter 3, mean shortest path 1.863, spectral-dimension proxy ≈ 0.005. Nothing extended appeared. Generic consistency plus generic accessibility does not produce space — it produces maximal interconnection, which is the exact opposite of a metric world. The interesting question inverts: not how to shorten distance, but why distance exists at all.

Lesson · The default state of an unconstrained relational world is 'everything near everything'. Distance is the thing to be explained, not the thing to be beaten.

A second coordinate-free graph toy, this time adding a preference for finite/polynomial ball growth to see whether extension can be recovered without inserting a target dimension.

Extension appeared, dimension did not. At N = 40: mean degree 4.65, diameter 11, ball-growth exponent 2.196 ± 0.389 — not 3, and with an uncertainty band wide enough that it selects nothing. Recorded explicitly as a FAILURE. The growth preference bought a stretched graph, not a derivation, and any attempt to tune it toward 3 would have been fitting, not deriving.

Lesson · Suppressing the small-world collapse is achievable and still not enough. Extension is cheap; a specific dimension is not.

Drop dimension from the objective entirely and test only LOOP STRUCTURE. Crude energy proxy on N ≈ 64 graphs built from degree regularity, a triangle penalty, and a square-loop reward — a stand-in for local consistency around cycles, with no reference to distance or dimension.

Structured periodic graphs separate from random graphs of the same degree. Energy proxy: 1D torus 0, 2D torus −0.03, 3D torus −0.1125; random degree-2 0, random degree-4 +0.008125, random degree-6 +0.168594. Read narrowly, and only narrowly: a crude loop-consistency proxy can distinguish some structured extended graphs from random graphs. It DOES NOT derive three dimensions, and the ordering across the torus family is not a selection principle.

Lesson · Hypothesis opened, not established: loop/path consistency — holonomy-like constraints around cycles — may be the ingredient that suppresses generic shortcut-rich collapse without ever mentioning spatial distance.

Put the two pressures in direct competition: a universal per-edge cost μ against a square-loop consistency reward ρ, scanned across a (μ, ρ) grid over periodic hypercubic test families in 1D–5D. No target dimension anywhere in the objective.

No robust unique 3D selection. Across the sampled grid the winner split evenly: 15 cases 3D, 15 cases 1D. FAILURE TO DERIVE DIMENSION. Caveat stated up front rather than buried: the 4D and 5D test tori use side length 3, which creates triangle artifacts and biases the comparison against them, so the split must not be overinterpreted. What the scan does show is weaker and real — competition between relation cost and loop reward can produce a dimensional phase preference inside restricted toy families.

Lesson · Two pressures can create a preference. They did not create OUR preference, and the motif choice was doing part of the work.

Cayley graphs of G = Z_L^d with reversible generators ±e_i. No Euclidean metric is supplied anywhere: the emergent distance between two group elements is simply the word length in the accessible generating set. Ball growth is measured to see what effective dimension the generators produce, and a single extra primitive translation generator is then admitted to see what it does to distances.

Ball growth rises monotonically with generator count — finite-size fits give 0.870, 1.709, 2.470, 2.960 and 3.468 for d_gen = 1…5. These fits UNDERESTIMATE the asymptotic d and are only a monotonic sanity check, not a measurement of dimension. On Z_31^3 ordinary word distances are (15,0,0) = 15, (15,15,0) = 30, (15,15,15) = 45. Admitting the primitive translation generator a = (15,15,15) rewrites them to 15, 16 and 1 — a 45× shortening for its matched target. Nothing propagated faster; the generating set changed, and the metric is downstream of the generating set.

Lesson · Physical distance can be modelled as word length in an accessible generating set. An alternate generator changes the metric by changing the set — it does not increase propagation speed.

A phenomenological free energy over generator count: F(d) = −log(2d) + b·C(d,2) + c·C(d,3). The first term rewards reversible branching and accessibility; the pair and triple terms charge for maintaining consistency among generators. 49 untuned (b, c) pairs were scanned over d = 1…8.

Winner counts across the sampled grid: d = 2 wins 7 cells, d = 3 wins 12, d = 4 wins 11, d = 5 wins 8, d = 6 wins 3, d = 7 wins 4, d = 8 wins 4. d = 1 never won in this range. Three dimensions win a finite region and nothing more: the preference is neither unique nor robust, and the couplings were not derived from anything. THIS IS NOT A DERIVATION OF 3D.

Lesson · A cost/benefit competition over generator count produces a dimensional phase diagram. A phase diagram is not a selection principle.

On Z_11^3 (N = 1331), admit the single primitive generator a = (5,5,5) globally and measure what it does to the ENTIRE graph, not just to the pair it was designed for. 10,000 random pairs sampled before and after.

About 54.92% of all sampled pairs were shortened. Mean ratio d0/d1 ≈ 1.5206, mean absolute drop ≈ 2.2693, and roughly 15.82% of pairs received ≥ 2× shortening. The designed target (0,0,0) → (5,5,5) collapses 15 → 1. A globally admitted Cayley generator is therefore NOT an endpoint-specific shortcut: it rewrites bulk geometry everywhere, which is exactly the observable that ordinary physics already constrains.

Lesson · You cannot quietly add a generator to the world. Adding it to the group adds it to everyone's geometry at once.

Z2 gauge-like edge variables U = ±1 on a periodic 12 × 12 square lattice, with plaquette holonomy Φ = product of the four surrounding edge variables. Start from the defect-free configuration and flip edges, tracking both defect structure and operational graph distance.

Initial defect count 0. Flipping one edge creates EXACTLY TWO Φ = −1 plaquette defects. Flipping a neighbouring edge MOVES the pair while the defect count stays 2, and the product of all plaquette fluxes remains +1 on the torus. This is textbook gauge-theory behaviour reproduced correctly — and it buys nothing geometric: the operational graph distance between (0,0) and (6,6) is 12 before the defects and 12 after. The naive step 'matter = defect, therefore curvature follows' FAILS in this model.

Lesson · Pair-created, movable, conserved topological defects are easy. Making them bend the metric that probes actually experience is the hard part, and this toy does not do it.

An invented pair-birth rule: a creation event produces two descendants carrying a complementary history relation. The hidden adjacency between them is DERIVED from shared ancestry rather than stored in an N-sized lookup table, which is the object the Phase 164–170 no-gos kept demanding. Tested as sparse persistent sibling edges on a 10 × 10 × 10 cubic lattice (1000 sites), scanning the fraction f of sites carrying a sibling relation.

At f = 0.002 (1 edge) 4.083% of sampled random pairs shorten, mean bulk fractional distance reduction 1.0083%, hidden-pair gain 9×. f = 0.005 (2 edges): 7.25%, 1.3458%, gain 7.5×. f = 0.01 (5): 17.083%, 3.3472%, gain 7.6×. f = 0.02 (10): 25.25%, 6.107%, gain 7.8×. f = 0.05 (25): 37.667%, 9.6861%, gain 7.08×. f = 0.1 (50): 47.833%, 14.3541%, gain 7.76×. RARE history-derived edges act strongly on their own endpoints while disturbing the bulk only modestly; COMMON hidden edges distort geometry fast. This is a STRUCTURAL LOOPHOLE in the Phase 178 bulk-isolation wall, not evidence that anything like it exists.

Lesson · Sparsity plus derivation-from-history is the first construction that gets an O(1) pair effect without an O(N) table and without wrecking the bulk. It buys a direction to attack, nothing more.

Suppose the pair relation is created LOCALLY and one endpoint is then deployed to separation r by ordinary means. Setup cost is r/v. A hypothetical repeated hidden response of duration t* gives T_pair = r/v + M·t* for M uses, against M·r/v for the conventional channel.

For large r and t* ≪ r/v the break-even point is typically after 2 uses, and the advantage grows linearly in M thereafter. The core limitation is structural and fatal to the popular reading: this CANNOT solve first arrival. Nothing here gets to a place faster than the deployment that put an endpoint there. What is described is REUSABLE PRE-POSITIONED INFRASTRUCTURE (or else genuinely new causal physics), never arbitrary on-demand creation of adjacency between two chosen points.

Lesson · Prepairing changes the amortised bill, not the first trip. Infrastructure is not transport.

Create the pair relation while both endpoints are local, then let ordinary dynamics separate them while the FUNDAMENTAL pair distance is assumed, by hypothesis, to remain 1. The apparent gain is the emergent separation divided by that fundamental distance. Ballistic separation gives gain ≈ 2t; three-dimensional diffusive separation gives RMS gain ≈ sqrt(6t).

At t = 1, 10, 100, 1000, 10 000 and 10^6 the ballistic gains are 2, 20, 200, 2000, 20 000 and 2 × 10^6; the diffusive gains are 2.449, 7.746, 24.495, 77.460, 244.949 and 2449.490. The point is not the numbers, which are trivial kinematics — it is that a relation NEED NOT BE CONSTRUCTED ACROSS a large emergent distance if it was created locally and merely persists while separation grows. Separation and deployment still cost ordinary time, so Phase 181 stands unchanged.

Lesson · Build the relation while you are next to it. Then the hard problem stops being construction and becomes persistence and darkness.

Thiemann, arXiv:2603.04072 (2026)

Relational observables and quantum reference frames. Read as conceptual context for treating observables relationally. It does NOT support hidden adjacency.

Baumann & Lock, Phys. Rev. D (accepted 10 August 2026)

Causality across temporal quantum reference frames; stresses operational interventions as the arbiter of causal claims. Used here as a standard we must meet, not as validation.

De Vuyst, Höhn & Tsobanjan, Quantum 10, 2196 (20 August 2026)

Algebraic and effective quantum reference frames. Relevant to the state-dependent algebra framing above. It says nothing about relational shortcuts and must not be cited as if it did.

History-derived sparse edges give large pair gains with modest bulk distortion

Phase 180. Z_10^3, 1000 sites: f = 0.002 → 4.083% pairs shortened, 1.0083% mean bulk reduction, 9× pair gain; f = 0.1 → 47.833%, 14.3541%, 7.76×. Structural loophole in the Bulk-Isolation Gate.

Prepaired relations amortise but cannot solve first arrival

Phase 181. T_pair = r/v + M·t* versus M·r/v; break-even typically at M = 2. First-Arrival Gate adopted.

A locally created relation can persist while emergent separation grows

Phase 182. Ballistic gain ≈ 2t, 3D diffusive ≈ sqrt(6t); tabulated t = 1…10^6. History-Edge Persistence Gate adopted.

Surviving conceptual target: state-dependent algebra U = (D, A, S(ρ), ρ, C)

Synthesis of Phases 176–182. No algebra written, no representation chosen, no observable computed. Relational history q_i is a speculative candidate identity primitive.

If the microscopic Hamiltonian contains only H_A ⊗ I + I ⊗ H_B — or, more generally, no term supported on both distant subsystems A and B — then conjugation by any local unitary U_A ⊗ U_B preserves that support structure exactly. We verified this as a toy: random local rotations applied to a strictly local two-qubit Hamiltonian leave the connected AB interaction norm at numerical roundoff (~1e-16) across many trials, while pre-existing X⊗X coupling yields connected norm 2.0.

Local endpoint operations CANNOT create a distant AB interaction term. They can only reveal, rotate, or modulate a pre-existing AB term, or else wait for ordinary propagation through a connecting medium. This is elementary operator algebra, but it is decisive for the programme: it converts 'activation creates adjacency' from an intuition into a forbidden move.

Lesson · The operator support of H is a conserved resource under local unitaries. You cannot conjugate your way across the gap.

Two qutrit endpoints with |0⟩ inactive and |1⟩, |2⟩ spanning an active relational subspace. A pre-existing sparse pair Hamiltonian H_AB = g(|1_A 2_B⟩⟨2_A 1_B| + h.c.) acts EXACTLY as zero on |00⟩, |10⟩ and |01⟩, but with O(g) strength on |12⟩. With g = 1 and initial state |12⟩, coherent exchange reaches unit transfer to |21⟩ at t = π/2.

This establishes the LOGICAL COMPATIBILITY of three properties the gates kept pulling apart: exact darkness in the inactive sector, purely local endpoint preparation, and O(1) pair coupling once both endpoints are active. The catch is structural and cannot be softened: it works only because the nonlocal AB operator is ALREADY PRESENT in H. Nothing in the toy creates it, and Phase 183 forbids creating it locally.

Lesson · Darkness, local unlocking and strong pair coupling can coexist — but only inside a Hamiltonian that already contains the pair term. The toy relocates the hard problem; it does not solve it.

Because H_AB is exactly zero whenever B remains inactive, both endpoints must either be locally activated (pre-armed) before any pair response is possible, or B must live permanently in an active receptor state. We traced the costs of each branch.

Branch one — pre-arming — requires prior coordination at B by ordinary means, so it inherits the First-Arrival Gate (Phase 181) unchanged. Branch two — an always-active receptor — risks background signatures, noise, decoherence and leakage in the ordinary sector, and so inherits the Darkness–Accessibility Gate (Phase 156) and the History-Edge Persistence Gate (Phase 182). Either way, a dormant history-defined pair can support reusable influence only as PRE-POSITIONED SHARED CAUSAL INFRASTRUCTURE. It does not give arbitrary one-sided creation of a remote adjacency.

Lesson · Readiness is a resource and someone must pay for it — in advance, in the open, or in leakage. There is no fourth option.

Ask whether an ordinary LOCAL parent Hamiltonian could generate the pre-existing AB operator that Phases 183–185 require. Generic perturbative virtual hopping along a local chain of length r with hop g and gap Δ gives effective endpoint coupling J_eff/g = q^(r−1) with q = g/Δ < 1. We tabulated q ∈ {0.2, 0.5, 0.8, 0.9, 0.95} against r ∈ {2, 4, 8, 16, 32, 64}. This is ILLUSTRATIVE GENERIC VIRTUAL-PROCESS SCALING, not a universal theorem.

For any fixed q < 1 the useful coupling decays exponentially in microscopic distance. Even the most favourable sampled case q = 0.95 falls to 3.950e-02 at r = 64; q = 0.5 reaches 1.084e-19 and q = 0.2 reaches 9.223e-45. Ordinary finite-order virtual hopping therefore does NOT generate O(1) remote support at growing r. Pushing q → 1 closes the gap and destroys the darkness the scheme depends on. The local-parent escape route is, in the generic case, closed.

Lesson · Locality charges exponential interest on distance. You do not get an O(1) remote coupling out of a generic local parent — you get a number with forty zeros after the decimal point.

A ring of N = 100 sites gives emergent A–B distance 50. Add ONE fundamental A–B interaction edge: the microscopic interaction-graph distance between A and B becomes 1. Compute what the local causal bound says afterwards.

The local causal lower bound t ≥ d_micro/v still holds exactly — nothing is violated. What changed is the reading: the emergent geometry OMITTED a real microscopic adjacency. This inverts the whole framing of the programme. If a fixed-strength rare AB operator is fundamental, then A and B are NEIGHBOURS in the deeper interaction metric even while they are far apart in emergent space. There is no shortcut and no superluminal anything; there is a mismatch between two metrics, only one of which we can currently see.

Lesson · Stop asking how to cross the distance. Start asking whether the distance was ever there in the layer that actually holds the interactions.

2025 · Phys. Rev. Lett. — Deterministic Generation of Photonic Entangled States Using Decoherence-Free Subspaces

Demonstrates photon-mediated bright/dark state structure, a decoherence-free dark subspace, and local drive/resonance control — within an ALREADY-MEDIATED photonic architecture. The dark subspace exists because the mediating interaction was engineered first.

2024 · Dynamical-decoupling-generated decoherence-free subspace experiments

Show that exact or strong darkness plus controllable unlocking are physically real motifs under engineered control. The protection is generated by active control sequences on an existing physical system — not by absence of interaction structure.

2021 · Symmetry-protected Floquet dark-state / selection-rule work

Selection-rule darkness with controlled unlocking, again inside a driven physical platform with an explicit Hamiltonian. Supports the motifs used in Phase 184 as ENGINEERED analogues only.

2025 · Nonlinear waveguide QED — infinite-range interactions and decoherence-free subspaces

Provides infinite-range effective interactions with decoherence-free subspaces as an engineered analogue. The 'infinite range' is mediated by the waveguide mode — an ordinary physical channel with an ordinary propagation story.

2025 · Hong Liu (review) — von Neumann algebras and emergent spacetime

Supports the ALGEBRAIC treatment of subsystems in regimes where naive tensor factorization fails — the formal setting in which Operator-Support Geometry would have to live. It is about how spacetime emerges from entanglement structure, not about hidden adjacency.

Gauge-theory literature — operational/local algebras of observables

Emphasizes that subsystem notions must be defined through algebras of observables rather than tensor factors. Conceptual grounding for d_op(A,B); says nothing about rare pair operators.

Topological / quantum-error-correcting-code literature — nonlocal logical operators

Nonlocal logical operators and symmetry-protected sectors exist and are well understood, but physical implementations pay extended support, code distance and resource costs. This is the strongest real-world analogue of Phase 184 — and it always carries the O(r) bill somewhere.

2025 · Locality-vs-quantum-codes results

Explicitly quantify the need for long-range interactions to beat locality bounds — formal confirmation, from the coding side, that local parents generically cannot manufacture fixed-strength remote support.

2026 · Distributed qLDPC nonlocal gates

Nonlocal logical gates between code blocks still consume ebits and pre-deployed infrastructure. The resource bill asserted by the Pre-Arming Gate shows up here in engineering form.

Local endpoint controls cannot create distant operator support

Phase 183. Connected AB norm ~1e-16 under 512 random local rotations of a strictly local H; 2.0 with a pre-existing X⊗X term. Gate adopted: Activation Is Not Adjacency Creation.

Exact darkness, local unlocking and O(1) pair coupling are logically compatible — given a pre-existing AB operator

Phase 184. Two-qutrit toy: H_AB dark on |00⟩, |10⟩, |01⟩; unit transfer |12⟩ → |21⟩ at t = π/2 with g = 1. Toy only; the pair term is installed by hand.

Dormant pairs are pre-positioned shared infrastructure, never one-sided remote creation

Phase 185. Readiness requires coordinated pre-arming (First-Arrival Gate) or an always-active receptor (Darkness–Accessibility and Persistence Gates). Pre-Arming / Shared-Resource Gate adopted.

Generic local parents pay exponential interest on remote coupling

Phase 186. J_eff/g = q^(r−1) tabulated over q ∈ {0.2…0.95}, r ∈ {2…64}: q = 0.5 gives 1.084e-19 at r = 64. Illustrative scaling, not a theorem. Local-Parent O(1)-Support Gate adopted.

A fundamental rare pair edge means the pair is near in the microscopic metric, not that emergent distance is crossed

Phase 187. N = 100 ring: one AB edge takes d_micro from 50 to 1; t ≥ d_micro/v intact. Operator-Support Geometry d_op ~ min C(O) defined; Metric-Inversion and Operator-Complexity Locality Gates adopted; core fork recorded.

A dense, state-dependent compatibility kernel J_ij = exp(−γ · d_H(q_i, q_j)) over K-bit state features. The matched target pair sits at Hamming distance zero and takes coupling 1; every wrong partner is suppressed by its Hamming distance, and the aggregate wrong coupling falls sharply as K grows: K = 16 at γ = 4 gives aggregate wrong ≈ 4.59e-7, K = 32 at γ = 2 gives ≈ 2.02e-8.

Mathematically, selectivity is cheap: an exponential kernel over enough bits separates one matched target from the whole rest of the register. The toy nevertheless FAILS as an addressing mechanism, because it obtains the match by COPYING B's history into A. The selectivity is real; the addressing is assumed. Recorded as a failure of the intended claim, kept for the kernel scaling it does establish.

Lesson · Suppressing every wrong partner is easy. Knowing which one is right, without being told, is the whole difficulty.

If a pair is created together, both members may carry a shared K-bit key with complementary roles — no copying required, because the key is common ancestry rather than transmitted information. Collision avoidance is a birthday bound: K ≥ ceil(log2(P(P−1) / (2δ))). For P = 1e12 pairs at δ = 1e-18 this gives K = 139 bits; for P = 1e40 it gives 325 bits.

Unique pair identity across astronomically many pairs costs only O(log P) bits. That kills the naive combinatorial objection to pair labelling. It does NOT provide a mechanism: two systems holding equal keys still share no operator support, and Phase 183 forbids manufacturing that support locally. Identity is settled; interaction is untouched.

Lesson · Naming a pair uniquely is logarithmically cheap. Coupling it is not addressed by naming it at all.

A strictly local two-qubit Hamiltonian H = H_A + H_B, probed by the linear response d⟨O_B⟩/dh_A at t = 0.2, 0.5, 1, 2 and 4. The response is zero at every time. We then applied an entangling change of representation — a different factorization of the same physical system — and re-measured: the physical response stays at ~1e-11, numerical zero. Adding an actual A–B edge to H produces immediate nonzero response, e.g. −0.29 at t = 0.5 and −1.27 at t = 1.

Refactorization does not create influence. Rewriting which degrees of freedom count as 'A' and 'B' changes the description and leaves the causal structure exactly where it was. This closes a route the programme had kept open: the hope that an emergent or alternative subsystem decomposition could yield adjacency without an interaction term.

Lesson · You cannot re-describe your way into a channel. The response function does not care how you slice the Hilbert space.

A coordinate-free compatibility matrix evolved under triadic consistency dynamics on N = 72: reinforce relations that close consistent triangles, decay the rest. No coordinates, no target dimension, no embedding.

The dynamics reliably self-organise, but into COMPACT small-world structures: representative diameters 2–4 and mean path lengths ≈ 1.9–2.6. Generic consistency does not generate extended locality; it generates the opposite. This independently reinforces the Extended-Geometry Gate already on the ledger.

Lesson · Consistency is not distance. A rule that rewards agreement collapses the world instead of stretching it.

Ball growth compared for the free group F3 (three generators, no relations) against Z^3 (the same three generators, made to commute). At radius r = 8 the free ball contains 585,937 elements; the Z^3 ball contains 833. The ratio is ≈ 703.4 and widens without bound.

Imposing commutation relations converts explosive accessibility ~5^r into polynomial accessibility ~r^3. This is the first mechanism in the programme that manufactures extended, low-dimensional geometry from a purely relational input, and it does it with RELATIONS, not coordinates. Proposed mapping: distinctions = states/orbits, accessibility = generators, consistency = relations, distance = word length, dimension = polynomial growth exponent.

Lesson · Relations are the thing that makes a world big. Removing them makes it exponentially crowded; imposing them makes it geometrically extended.

A minimal groupoid-like toy in which the relation ab = ba fails at one localised place. The failure creates an extra local branch: a genuine additional state reachable only through the defect.

Every ordinary pair distance is UNCHANGED. A localised failure of a relation adds local structure and produces exactly zero long-range metric shortening. The defect route to adjacency, which the programme had carried since the holonomy phases, does not act on the metric by itself.

Lesson · Breaking a relation locally buys local structure at local prices. It does not shorten anything far away.

Two localised defects placed in the same word-metric toy, sharing a common defect character (identity, charge, ancestry, topological label). Ordinary A–B word distance before: 12. With both defects present and matching: still 12. Only after inserting an explicit defect–defect operator does the distance drop to 3.

The hard principle of this ledger entry: shared identity, shared charge, shared ancestry and shared topology DO NOT create causal adjacency. Adjacency appeared only at the moment we installed, by hand, an operator supported on both defects — the same move Phase 183 forbids doing locally.

Lesson · Two things being the same kind of thing is not a channel between them. It never has been on this ledger, and it is not one now.

Fundamental relational variables R_ij as the only degrees of freedom, evolved under triadic reinforcement plus a resource/exclusivity penalty intended to prevent the Phase 191 collapse into a dense small world.

FAILURE, and the opposite failure to Phase 191: the universe empties. Final max R ≲ 0.074, operational degree 0 across every tested target resource. Too little constraint gives a dense small world; too much penalty gives nothing at all. Recorded permanently as a failed model.

Lesson · The two obvious knob settings both destroy the world. The law we are missing is not on this axis.

The penalty replaced by a CONSERVED row strength, sum_j R_ij ≈ κ per node, alongside the same triadic reinforcement. Each node has a fixed relational capacity to spend, and spending it somewhere means not spending it elsewhere.

The top-3 operational graph on N = 96 stays CONNECTED and sparse: mean degree ≈ 3.7–3.85, diameter 6–7, mean path ≈ 3.5–3.7. This escapes both the dense and the empty extremes — the first model in the track to do so. It is still far too compact to be geometry. Sparsity alone is not extension.

Lesson · Conservation is the right shape of constraint and the wrong strength of result. A sparse small world is still a small world.

The conserved-budget model scaled across N = 32, 64, 128, 256, 512. Measured diameters ≈ 5.0, 5.33, 6.0, 7.0, 7.33; mean paths ≈ 2.66, 3.10, 3.66, 4.23, 4.65. A power-law fit gives α ≈ 0.1497, a naive d_eff ≈ 6.68; a logarithmic fit gives R² ≈ 0.9704 and is essentially just as good.

The model remains small-world / high-dimensional. When a log fit is indistinguishable from a power fit, no polynomial growth has been established, and the naive d_eff must not be quoted as a dimension. FAILS low-dimensional emergence.

Lesson · An effective dimension extracted from a log-like curve is a number, not a geometry.

Relational weight allowed to spread ONLY along currently supported two-step paths, so no spontaneous global support can appear — the dynamical analogue of the Support-Invariance Gate. Scaled over N = 32…512.

Diameters grow strongly: ≈ 5.67, 5.33, 13.33, 23.33, 35.67, with an apparent exponent α ≈ 0.7437 (naive d ≈ 1.345). But the graph FRAGMENTS: mean reached nodes are only ≈ 5.2, 5.5, 8.6, 39.8 and 113.1 respectively. The extension is an artefact of measuring distances inside shrinking connected pieces. Recorded as a failure with a genuinely informative signature.

Lesson · You can always make a world look extended by breaking it into pieces and measuring inside one of them.

Connected ancestry support forced to persist, while only local two-step reinforcement is permitted. Scaled over N = 32…512.

Graphs stay FULLY CONNECTED at every size: mean degree ≈ 5.2–5.53, diameters ≈ 4, 4.67, 5, 5.67, 6, mean paths ≈ 2.18, 2.63, 3.05, 3.49, 3.90, α ≈ 0.145 (naive d ≈ 6.9). Restoring continuity sends the system straight back to small-world scaling. The track now has a clean BIFURCATION: local-support conservation gives extension with fragmentation; continuity preservation gives connectedness with small-world compactness. Nothing tested so far gives both.

Lesson · The missing law must produce connectedness and polynomial growth at once, without hard-coding a dimension or any coordinates. Neither branch of the bifurcation is close to it.

Under Gromov's theorem (polynomial growth ⇒ virtually nilpotent), the Bass–Guivarc'h homogeneous growth degree D = Σ_i i·r_i organises the low-degree cases, where r_i is the rank of the i-th layer of the lower central series. We enumerated the screened low-degree nilpotent rank patterns as a NECESSARY-CONDITION scan.

Degrees D = 1, 2 and 3 admit only abelian rank patterns in the screened cases (r = (1), (2), (3)). D = 4 is the first degree that admits a genuinely nonabelian pattern, r1 = 2, r2 = 1 — the Heisenberg-type case. Stated carefully: this is a STRUCTURAL CLASSIFICATION CLUE about where noncommutativity can first appear in a growth hierarchy. It is NOT a derivation of three-dimensional physics and does not privilege D = 3.

Lesson · Group growth degrees are a real organising quantity. They organise possibilities; they do not select one.

A dimension-independent toy objective S = A·log(D+1) − B·H − C·c·(D − H), with H = Σ r_i the total layer rank and (D − H) the commutator burden. Nothing in the score mentions three. We scanned 6,080 coefficient cells and recorded the winning degree in each.

Winner fractions: D = 1 at 31.58%, D = 2 at 12.75%, D = 3 at 8.63%, D = 4 at 6.15%, with higher degrees also winning frequently. So D = 3 CAN emerge without three being inserted into the score — and it is neither unique nor robust, losing in over 91% of the coefficient space. Any claim that this derives three dimensions is REJECTED here by us, before anyone else has to reject it.

Lesson · An outcome that appears in 8.6% of your parameter space is a possibility, not an explanation.

An MDL-like objective: accessibility gained minus description cost paid, applied to abelian Z^D presentations against nonabelian alternatives, with no free coefficients to tune.

FAILURE. For pure abelian Z^D the accessibility term and the first-layer description cost CANCEL exactly in this toy, leaving the objective flat in D; a nonabelian example can even win outright. Compression alone does not select D = 3, or any D. Recorded as a failure of the compression route and preserved.

Lesson · Minimum description length is a beautiful principle that, here, says nothing at all about dimension.

Nearest-neighbour diffusion on Z^d has return probability P_return(t) ~ t^{−d/2}. The cumulative Green memory Σ_t P_return(t) therefore diverges for d ≤ 2 (Pólya recurrence) and converges for d ≥ 3 (transience). We computed partial Green sums numerically to see the boundary directly.

Partial sums for d = 1 and d = 2 keep growing with the cutoff; d = 3 and d = 4 saturate. The narrow statement we are willing to make: d = 3 is the MINIMUM DIFFUSIVE ABELIAN DIMENSION WITH TRANSIENCE. This is a candidate clue about why a world with persistent structure might not be one- or two-dimensional. It is not a derivation and it is classical mathematics that has been known since 1921.

Lesson · Pólya's theorem is a real boundary. Whether it is the reason for anything physical is a separate question, and Phase 205 answers it badly.

A deeper Green-function run with cutoffs M = 30, 100, 300 and 1000, reading the cumulative self-return of a diffusive surrogate for a defect's own radiated history.

Representative values at M = 1000: d1 ≈ 35.823, d2 ≈ 3.326, d3 ≈ 1.496, d4 ≈ 1.241. Increments from M = 300 to M = 1000: d1 ≈ 16.253, d2 ≈ 0.442, d3 ≈ 0.00443, d4 ≈ 0.00313. In this surrogate, cumulative self-return memory becomes finite first at d = 3 — the d = 3 increment is already four orders of magnitude below d = 1. INTERPRETATION ONLY. This is a diffusive surrogate, not a defect, not a field, and NOT a physical derivation of anything.

Lesson · A defect that never stops re-hearing itself is a bad candidate for a stable object. That is suggestive and it is not physics.

Generalise the spreading law to L ~ t^{1/z} with a return tail ~ t^{−d/z}. Finite cumulative memory then requires d > z, so the minimum dimension is a function of the dynamical exponent, not of geometry alone.

Ballistic z = 1 gives d_min = 2. Diffusive z = 2 gives d_min = 3. Subdiffusive z = 3 gives d_min = 4. The recurrence argument therefore selects three ONLY IF the dynamics is already assumed diffusive — which is an assumption about dynamics, not a derivation from structure. The Phase 203/204 clue is hereby DOWNGRADED by us: recurrence does not select 3 from geometry alone. FAILURE of the 'Pólya explains 3D' reading, recorded permanently.

Lesson · We built a clue in two phases and broke it in one. That is what the ledger is for.

Toy relational-history ensembles weighted two ways: classically, w = exp(−λC) over configuration cost C; and quantum-style, by summing amplitudes Σ_h exp(i·θ·C) over histories before taking the modulus.

Interference genuinely RESHAPES the structural weights — the amplitude ensemble is not reproducible by any reweighting of the classical one, which is a real if unsurprising distinction. But the first test still favoured the largest available D class. Recorded narrowly: interference is not equivalent to random weighting, and it does not solve dimension selection.

Lesson · Adding phases changes the answer. It does not, by itself, change the answer to the one we want.

Sweep the phase parameter θ over 721 values and record which degree class the interference weighting selects.

Winners in this toy: D = 2 at ≈ 16.64%, D = 6 at ≈ 33.29%, D = 7 at ≈ 50.07%. D = 3 NEVER WINS at any scanned θ. A free phase law simply replaces one tuning knob (the coefficients of Phase 201) with another (θ). FAILURE of the interference-selection route, preserved.

Lesson · If your mechanism has a dial, the dial is doing the work.

Within the restricted family P = |ψ|^p, test two requirements: multiplicative composition under tensor products, and universal normalisation for random normalised complex states in dimensions 2, 3, 5 and 8.

Multiplicative composition alone permits ANY p > 0 — it is no constraint. Universal normalisation singles out p = 2 within this family, with numerically zero error at p = 2 and clearly nonzero error at neighbouring exponents. CRITICAL CAVEAT, stated by us: the normalisation condition presupposes the Hilbert 2-norm, so this is NOT a noncircular derivation of the Born rule. The honest reading is a consistency check, and it redirects the target: derive the inner-product / Hilbert geometry FIRST, and probability afterwards.

Lesson · We did not derive the Born rule. We rediscovered that we had assumed it.

One relational Laplacian L used twice: to define an operational graph distance from its adjacency, and to define a heat-kernel / information distance from its spectrum. An N = 80 ring with 0, 1, 2, 4, 8 and 16 added hidden supports.

Diameter from the adjacency falls from 40 to ≈ 13 as hidden support increases, while the correlation between graph distance and information distance rises from ≈ 0.759 to ≈ 0.997. One operator CAN co-generate propagation/information geometry and operational geometry together. But both were read off the SAME operator, so the agreement is co-generation by construction — a tautology risk, not a derivation of spacetime.

Lesson · Reading one object two ways and finding it consistent with itself is the easiest result in this laboratory.

Fixed 4-qubit Hamiltonian: an ordinary XY chain A–M–C plus a direct A–C XY term conditioned on a CONSERVED sector qubit S. The underlying H is identical in both runs; only the sector value differs. Attempts 210A and 210B are preserved: the initial observable and pulse choices were symmetry-blind and returned zero response even though the transfer had changed — a diagnostic failure of our probe, not of the model, and it is recorded rather than hidden. 210C is the clean test.

Operational distinguishability at C between local A = 0 and A = 1 preparations. Ordinary sector S = 0 first reaches 1% at t ≈ 0.46, 5% at ≈ 0.70, 10% at ≈ 0.85, 25% at ≈ 1.12 and 50% at ≈ 1.42. Activated sector S = 1 reaches 1% at ≈ 0.11, 5% at ≈ 0.23, 10% at ≈ 0.33, 25% at ≈ 0.57, with maximum signal ≈ 0.444 in the scanned window. Early-time scaling separates the mechanisms cleanly: ordinary ≈ t^3.996 (two-hop probability), activated ≈ t^1.991 (direct-edge probability). A fixed microscopic algebra can therefore support SECTOR-DEPENDENT operational causal structure — provided the hidden operator is ALREADY PRESENT in H. Mathematical analogue only; no claim that nature does this.

Lesson · The exponent 4 versus 2 is the whole content: the activated sector is using a different term, not a faster version of the same one.

The same fixed 4-qubit architecture. The ordinary S = 0 sector is protected by a gap Δ = 8; a weak imperfection ε·X_S mixes the sectors; the hidden A–C edge acts only in S = 1. Starting in the ordinary sector, we measure the EXTRA signal at C caused by the hidden edge, relative to the identical model with the hidden edge removed, at t = 0.8.

ε = 0.01, 0.02, 0.04, 0.08, 0.16, 0.32 (ε/Δ = 0.00125, 0.0025, 0.005, 0.01, 0.02, 0.04) give extra signal ≈ 4.84e-7, 1.94e-6, 7.74e-6, 3.09e-5, 1.24e-4, 4.90e-4. The fitted power is ≈ 1.997, i.e. leakage ~ (ε/Δ)² in this toy, and at exact symmetry ε = 0 the hidden sector is PERFECTLY DARK. Exact darkness plus perturbatively small leakage is therefore logically possible when hidden support pre-exists. The existence and origin of that support remains ENTIRELY UNPAID.

Lesson · We can make a hidden term invisible. We still cannot say where it came from, and that is the only bill that matters.

The same fixed four-qubit model as Phase 210, with one methodological change: the matter tensor-factor index is NO LONGER treated as a geometric position. Pair edges are inferred solely from finite-difference dynamical response strength between candidate local observable blocks, thresholded at 35% of the maximum pair response.

Sector 0 pair responses: block0–block1 = 0.122412, block0–block2 = 0, block1–block2 = 0.122412 — the reconstruction is a CHAIN. Sector 1: all three pair responses = 0.121138 — the reconstruction is a TRIANGLE. Operational geometry can therefore be inferred from measurable response rather than assigned from microscopic labels. CRITICAL LIMITATION, stated by us: the candidate subsystem blocks were still supplied by the tensor decomposition, so the factorization was handed over rather than derived.

Lesson · Reading geometry off an index is not a measurement. Reading it off a response at least could be.

A random Haar-like global unitary W was applied to the entire four-qubit Hilbert space, to the Hamiltonian, to the states and to the pool of local Pauli generators; the twelve transformed generators were then shuffled and NO block labels were supplied. We built the operator noncommutation graph and read its connected components.

Connected components recover four factor blocks of sizes [3, 3, 3, 3]. The selector block is identified purely by containing an operator that commutes exactly with H: conservation scores (min ‖[H,O]‖) ≈ block0 4.89897949, block1 0, block2 5.65685425, block3 4.89897949 — selector inferred as block1, with the remaining three treated as matter. Response-inferred geometry then reconstructs sector 0 as [[0, .108811, .013745], [.108811, 0, .108811], [.013745, .108811, 0]] → CHAIN at the 35% threshold, and sector 1 with all off-diagonals .121209 → TRIANGLE. CRITICAL CAVEAT: the generator pool is itself privileged and already carries the factorization information.

Lesson · Scrambling the basis proves the labels were not load-bearing. It does not prove the generators were free.

We quantified the Pauli-weight decomposition of the same abstract H relative to a FIXED four-qubit tensor-product structure, before and after twelve independent global unitary scrambles, deliberately NOT co-transforming the factorization.

Original H: weight-1 = 0, weight-2 = 0.9000, weight-3 = 0.1000, weight-4 = 0. Scrambled mean: weight-1 ≈ 0.048634, weight-2 ≈ 0.224049, weight-3 ≈ 0.435902, weight-4 ≈ 0.291415. Every individual scramble spread substantial weight into three- and four-body terms. Whether H LOOKS local and low-weight is relational to the chosen tensor-product / observable structure, not an absolute property visible in the abstract matrix. This is a NUMERICAL DEMONSTRATION and explicitly NOT a theorem that H can never select a preferred factorization.

Lesson · The word 'local' is a statement about a pair — a Hamiltonian and a factorization — never about a matrix.

Phase 213 repeated across twenty independent global basis scrambles, checking whether block recovery, selector identification and sector-dependent geometry survive arbitrary changes of representation.

In EVERY trial: noncommutation components recover four three-generator blocks; conservation identifies the selector; response inference yields exactly 2 edges in sector 0 (chain) and 3 edges in sector 1 (triangle). Pass rate 20/20. Mean sector-0 weakest/strongest response ratio ≈ 0.1263186; sector-1 ratio ≈ 1.0. Once the accessible observable algebra and its generator family are transformed TOGETHER with the physics, the inferred operational geometry is representation-invariant under arbitrary global basis change in this toy. This is the positive result of the block — and it does not derive the accessible algebra itself.

Lesson · We removed the coordinates and the geometry survived. We did not remove the algebra, and that is the whole remaining bill.

A fixed abstract 16-dimensional Hilbert space and a fixed Hamiltonian were chosen. We generated candidate 2×2×2×2 tensor-product structures via global unitaries, and scored each candidate by the mean single-factor linear entropy created from random product states over a short evolution. The factorization was NOT supplied; only its factor dimensions (2×2×2×2) were assumed.

The native factorization scored ≈ 0.00404. The 80 random global tensor-product structures scored mean ≈ 0.00693, std ≈ 0.00032, with the best random ≈ 0.00630. The native factorization ranked best among all 81 candidates. Interpretation, stated cautiously: dynamics CAN prefer a low-entangling factorization class — but the factor dimensions 2×2×2×2 were assumed, so this is a selection within an assumed class, not full emergence of the factorization.

Lesson · Dynamics can tell you which 2×2×2×2 you would rather be in. It has not yet told you why it is 2×2×2×2 at all.

The Dynamical Factorization Sieve of Phase 216 was re-run under local product-unitary basis changes and factor permutations, to ask whether the winning low-entangling score attaches to a single arbitrary coordinate labelling or to a gauge equivalence class of factorizations.

Local product-unitary basis changes and factor permutations retain essentially the same low-entangling score, ≈ 0.0036–0.0039. The sieve therefore selects an EQUIVALENCE CLASS of tensor-product structures rather than any arbitrary coordinate labelling. This is the positive result of the pair — and it inherits the Phase 216 caveat that the factor dimensions were still assumed.

Lesson · The sieve does not pick a coordinate system; it picks a class. A class is still one step removed from a derived 2×2×2×2.

Instead of scoring candidate factorizations by a single quantity, we now gate them on three simultaneous criteria — persistence, communication and scrambling — across aligned factorization classes, and ask whether any class can satisfy all three at once. Aligned 2×2×4 and aligned 2×2×2×2 were both checked.

The aligned 2×2×4 class SURVIVED the simultaneous gates, while the aligned 2×2×2×2 class FAILED communication at one sampled time. Because the failure appeared at an unfavourable time rather than being tuned away, the outcome was NOT produced by selecting times or coefficients by hand. Interpretation, stated cautiously: an object — a stable, communicating, weakly scrambling factor — may be an emergent property of a factorization class rather than of a single coordinate labelling, but no single class passed every gate here, and the criteria themselves are toy definitions.

Lesson · Three gates can agree on a class without a single number saying so. They can also disagree, and that disagreement is information, not noise.

The simultaneous three-gate test was re-run at five sampled evolution times, t = 0.10, 0.20, 0.30, 0.50, 0.80, to ask whether any class survives across the time axis rather than at a single hand-picked instant.

2×2×2×2 survived all 5 sampled times, 2×2×4 survived 4 of 5, while 2×8 and 4×4 survived 0 of 5. The survival ordering across the sampled time axis is therefore 2×2×2×2 ≥ 2×2×4 ≫ 2×8, 4×4. Caveat kept visible: this is robustness across a finite, coarse set of sampled times in a toy, and 'survival' depends on the three gates' thresholds as much as on the dynamics.

Lesson · The survivor in a toy can change when you ask at a different time. A coarse sweep is better than a single snapshot, and still not a guarantee.

The three-gate selection was perturbed and asked whether the winning class is stable under small changes to the dynamics. The decision rule was a coarse majority-win criterion across the perturbed ensemble.

INCONCLUSIVE. The coarse majority-win rule produced MANY TIES and could not cleanly separate classes. This is explicitly NOT a pass: ties in a coarse decision rule tell us the rule is too blunt to resolve robustness, not that robustness was demonstrated. Recorded as inconclusive, preserved, and not promoted to a positive result.

Lesson · An inconclusive result is a result: the tool was too blunt to answer, and we say so rather than squint at a tie until it looks like a winner.

Rather than a coarse majority rule, we measured the DIRECT differential between the two leading classes, 2×2×2×2 and 2×2×4, across a perturbation sweep eta from 0 to 0.20, on each of the three gates independently.

Across the whole sweep, 2×2×2×2 had consistently higher persistence (≈ +0.022) and lower scrambling (≈ −0.025) than 2×2×4, but SLIGHTLY LOWER communication (negative delta, improving from about −0.0048 to −0.00215 as eta grows). No single class dominates on all three criteria. Interpretation: the preferred factorization may be selected by a PARETO BALANCE across the persistence–communication–scrambling trade, not by a single scalar optimum. This is a toy-level conclusion and does not yet give a principled way to pick a point on that frontier.

Lesson · When one class wins on two gates and loses on the third, 'the winner' is a decision, not a discovery. The decision rule is the real open problem.

The median thresholds and the weighted scalar objective of Phases 220–223 were removed entirely. Candidate factorizations are evaluated on three raw quantities: persistence P (maximize), communication C (maximize) and scrambling S (minimize). A candidate is Pareto-dominated only if some other candidate has ≥ P, ≥ C and ≤ S with at least one strict improvement. 48 candidates per class were generated over t = 0.15, 0.30, 0.60 and random global orientations.

Frontier hit rates: 2×2×2×2 reached the nondominated frontier in 15 of 48 candidates (31.25%); 2×2×4 in 3 of 48 (6.25%); 2×8 and 4×4 in 0 of 48. For the aligned / native candidates: at t = 0.15 both 2×2×4 and 2×2×2×2 sat on the frontier; at t = 0.30 both again; at t = 0.60 only 2×2×2×2. Conclusion: threshold-free Pareto selection SUBSTANTIALLY FAVOURS 2×2×2×2 in this toy but DOES NOT UNIQUELY DERIVE IT — nondominated is not the same as selected.

Lesson · Removing the knobs did not remove the ambiguity. It only stopped us from hiding it inside a weighting.

H(λ) = ordinary chain + λ · selector-controlled hidden interaction. 2×2×4 and 2×2×2×2 were compared by pairwise Pareto dominance on P/C/S as λ was swept from 0 to 3, asking whether a single control knob can drive a clean switch between locality organizations.

FAILURE. Most samples were TRADEOFFS (10 of 13). The 4-factor class dominated only at isolated λ = 2.0, 2.75 and 3.0, with no contiguous region and no clean locality phase switch anywhere in the sweep. The simple story that one gain parameter flips which factorization is 'the local one' is recorded here as a negative result and is not tuned away.

Lesson · Scattered wins along a sweep are not a phase boundary. If the basin is real, it should be findable without picking the sample points.

H(θ) = cos θ · H₀ + sin θ · H₁, interpolating between two simple, deliberately competing symmetry / interaction organizations. 2×2×4 and 2×2×2×2 were compared on P/C/S with no scalar weighting, over 17 sampled θ.

FAILURE / NEGATIVE RESULT. Of 17 samples, 15 were TRADEOFFS; only 2 favoured the 4-factor class; the 3-factor class NEVER dominated. Even an interaction organization designed to compete with the native one did not create a clean factorization / locality phase transition. Changing the relative coupling strengths simply moves along the tradeoff, it does not exchange the basin.

Lesson · We built the fight we wanted to win and still lost it. That is the point of building the wall first.

Define the cross-factor fraction X = ‖H_cross‖_F² / ‖H‖_F² after projecting H onto the sum of the one-factor operator subspaces. Compare F4 = [2,2,2,2] against F3 = [2,2,4] as H₁ is turned on.

Raw result: F4's cross fraction stayed at 1 throughout, while F3's fell from 1 to ≈ 0.006 as H₁ came to dominate — so on this metric F3 appears to win almost everywhere. WE REJECT THIS METRIC. The apparent win is an artefact: coarse grouping lets a larger factor ABSORB interactions and relabel them as 'internal'. Any metric that pays you for merging subsystems will always prefer the coarsest grouping, up to the trivial one-factor limit where cross support is zero by construction.

Lesson · A metric that says 'one big object is maximally local' has told you about the metric, not about the physics.

To correct the coarse-graining bias, each factorization's observed cross-factor fraction is compared against ITS OWN random global-orientation baseline: z = (observed − random_mean) / random_sd. More negative z means unusually local for that class, independent of class size.

F3 becomes strongly 'surprising / local' as H₁ dominates, with z down to roughly −38 to −42, while F4 stays at POSITIVE z ≈ 2.4–2.9 — because a one-factor projection treats all pairwise physics as cross-factor by definition. The normalization removes part of the class-size bias but reveals a deeper failure: ONE-FACTOR LOCALITY IS THE WRONG NOTION. Ordinary local physics is few-body, not one-body, so a projection onto single-factor operators mis-specifies what we were trying to measure.

Lesson · We fixed the bias and found the definition underneath it was wrong. The correction was worth more than the number it produced.

A structural, coefficient-free comparison replaces the norm metrics: maximize the number of independently addressable factors N_obj; minimize the number of distinct cross-factor support subsets E_cross; minimize the maximum cross degree. No weights, no thresholds, no coefficients enter.

For H₀: F4 has N_obj = 4, E_cross = 3, max degree 2, with supports [(0,1), (0,3), (1,2)]; F3 has N_obj = 3, E_cross = 3, max degree 2 — so F4 DOMINATES. For H₁: F4 has N_obj = 4, E_cross = 3, degree 2; F3 has N_obj = 3, E_cross = 2, degree 2 — a PARETO TRADEOFF. Key conclusion: merging factors can reduce cross-support ONLY by reducing the number of independently addressable objects. A genuine alternate-adjacency mechanism must PRESERVE OBJECT COUNT while changing which supports are low-complexity.

Lesson · You can always make the world look local by admitting fewer things exist in it. That is bookkeeping, not adjacency.

An exhaustive-random search over exact Clifford refactorizations on four two-level factors. Requirements fixed before the search: the same four objects and the same information capacity in both structures; the same fixed abstract Hamiltonian H; F* = W F₀ W† with W genuinely entangling; H exactly Pauli-weight ≤ 2 in BOTH factorizations; and support graphs that are non-isomorphic.

The first unconstrained exact construction appeared at trial 55 but its F* support graph was DISCONNECTED, so it was REJECTED on the connectivity requirement rather than reported. Phase 230B tightened the search to demand both graphs connected, the same edge count, exact ≤ 2-locality and non-isomorphism; a solution was found at trial 134. F₀ terms ['IZIZ', 'IIYZ', 'YIIY'] with W = [CX(3,0), CZ(1,0), H(1), CX(3,0)] give F* terms ['ZXII', 'IIYZ', 'YIIY']. The F₀ support graph has edges (0,3), (1,3), (2,3), degree sequence [1,1,1,3] — a star K₁,₃; the F* graph has edges (0,1), (0,3), (2,3), degree sequence [1,1,2,2] — a path P₄. Both connected, three edges each, non-isomorphic, maximum Pauli weight 2 in both. EXACT FINITE EXISTENCE RESULT: one fixed abstract H can be sparse 2-local under two inequivalent equal-object-count tensor-product structures whose adjacency graphs genuinely differ. It does NOT show either structure is physically selected, it does not change H, and it does not imply any change to spacetime.

Lesson · The first hit was thrown away for being disconnected. Rejecting your own first success is the only reason the second one means anything.

Phase 230 is strengthened by demanding that every two-body Clifford gate used to define W lie on an EDGE OF THE ORIGINAL F₀ SUPPORT GRAPH G₀ — no gate may act across a pair that is not already adjacent in the ordinary structure.

A solution was found at trial 855. F₀ terms ['IIXX', 'IXIZ', 'IIYX', 'ZIZI'] give G₀ edges (0,2), (1,3), (2,3), degree sequence [1,1,2,2] — a path P₄. W = [H(2), CX(0,2), CZ(0,2), CX(3,2), CZ(0,2), S(1)], and every two-body gate in it is G₀-local. F* terms ['ZIIX', 'IYIZ', 'IIXY', 'IIXZ'] give G* edges (0,3), (1,3), (2,3), degree sequence [1,1,1,3] — a star K₁,₃. Same four objects, same three-edge count, both connected, non-isomorphic, exact maximum Pauli weight 2 in both. This removes the simplest finite-size 'you just globally rewrote the basis' objection: at N = 4 an entangling refactorization can be synthesized using ONLY ordinary-graph-local gates. It remains a CIRCUIT defining an alternate tensor structure, NOT a demonstrated state-driven phase transition.

Lesson · Building it with local gates at N = 4 answers an objection about basis rewriting. It says nothing about whether anything selects it.

A support bound for a depth-d circuit of bounded-range gates local on G₀: supp(W O_A W†) is contained in the graph-distance-d neighborhood N_d(A). Moving support from A to B at graph distance r therefore requires d ≥ r; if support expands independently from both endpoints and need only overlap, 2d ≥ r, so d ≥ ⌈r/2⌉.

Path examples at N = 8, 16, 32, 64, 128, 256 have endpoint distances r = 7, 15, 31, 63, 127, 255 and minimum overlap depths 4, 8, 16, 32, 64, 128, with d/r → 1/2 asymptotically. CONCLUSION: a bounded-depth ordinary-local refactorization CANNOT create a macroscopic adjacency change across unbounded ordinary distance. This is a graph / circuit causal-cone constraint, analogous in spirit to locality and Lieb-Robinson bounds; it is NOT a spacetime theorem and is not offered as one.

Lesson · We built the finite thing, then proved it does not scale. Both halves stay on the page.

A fixed Hamiltonian contains both an ordinary chain term H₀ active in sector S = 0 and an alternate direct A–C coupling term H* active in sector S = 1. Activation is a local selector pulse on the sector variable only; no term is ever created, only switched between two structures already written into one fixed H.

Ordinary-sector signal thresholds at 1 / 5 / 10 / 25 / 50 percent read t ≈ 0.46 / 0.70 / 0.85 / 1.12 / 1.42; the alternate direct A–C sector reads ≈ 0.11 / 0.23 / 0.33 / 0.53 / 0.79. The local selector flip costs t = π/(2Ω). KEY VERDICT: O(1) local activation is possible ONLY because both causal structures are pre-encoded in one fixed H. The apparent shortcut is a bookkeeping transfer, not a creation. NEW UNPAID BILL: the physical legitimacy and cost of a selector that globally gates a distant A–C term — who built H*, what does holding it cost, and why is the selector itself exempt from the locality it switches. TOY-MODEL ARCHITECTURE, NOT EVIDENCE.

Lesson · Flipping a switch between two pre-built worlds is O(1). Building the two worlds is where the bill went.

Suppose the hidden sector is encoded or protected with code distance d_code. Then any exact logical sector-changing operator must have weight at least d_code — this is the defining property of the code, applied to the activation operator itself.

Macroscopic local-error protection and exact bounded-support local activation CANNOT both scale indefinitely: growing d_code suppresses the leakage we want and suppresses the local activation we need by the same amount. STATED PLAINLY: a globally protected hidden sector cannot be simultaneously strongly locally invisible and exactly switchable by a strictly local bounded-support operator — unless protection is weakened, a pre-existing nonlocal control resource is used, or the sector is not a conventional encoded/topological degree of freedom at all.

Lesson · The same distance that hides the sector from noise hides it from us. The code does not care which side of the wall we stand on.

For a local control V define m(L) = |⟨1_L|V|0_L⟩|, the sector-flip matrix element at system size L. Under bounded control amplitude Ω the ideal two-level flip time is t_flip = π/(2Ω·m). The same m measures local susceptibility and leakage to that perturbation class.

If invisibility demands m(L) → 0, local activation time diverges; constant-time local activation requires m = O(1), so the sector remains O(1)-susceptible to that same perturbation class. Representative scalings: m ~ 1/L gives t_flip ~ L; m ~ 1/L² gives t_flip ~ L²; m ~ exp(−L/8) gives t_flip ~ exp(L/8) — exponential invisibility is exponential activation. There is no free direction: every decibel of protection is paid for in activation time, in the same units, through the same matrix element.

Lesson · The knob we turn to reach the sector is the same knob the world turns to reach it. One matrix element, two names.

A hidden A↔B transfer term H* = g·X_AB ⊗ P_allowed is projected onto a permission type. In type 0 the projector annihilates the term exactly; in type 1 the transfer runs at O(1) coupling. Nothing is tuned small — the suppression is algebraic, not perturbative.

Forbidden sector: maximum transfer exactly 0 over the whole simulation. Allowed sector: transfer reaches ≈ 1, thresholds 1/5/10/25/50/90 percent at t ≈ 0.11 / 0.23 / 0.33 / 0.53 / 0.79 / 1.25. Local type flip costs t = π/(2Ω). KEY VERDICT: exact selection rules can realize m_inactive = 0 and m_active = O(1) simultaneously, so the Phase 235 susceptibility–activation tradeoff is NOT universal in that narrow sense. UNPAID BILL: the hidden A–B operator support still exists in the fixed H from the very start; the projector gates it, it does not create it.

Lesson · A perfect lock is not a perfect absence. The door was already in the wall before we found the key.

Matter A and B carry two local type labels T_A and T_B. The hidden transfer requires BOTH endpoints to permit it: H* = g·X_AB ⊗ P1_A ⊗ P1_B. This is the minimal way to make permission genuinely relational rather than one-sided.

Label states 00, 10 and 01 all give maximum transfer exactly 0. Only the 11 state activates, reaching ≈ 1 with the 10 percent threshold at t ≈ 0.33 and 50 percent at t ≈ 0.79. VERDICT: if permission depends on both endpoints, no strictly local flip at one endpoint can activate the remote edge. Fast activation then requires endpoint coordination (a prior round trip), a shared or global selector, or a genuinely deeper relational degree of freedom that is not a product of two local labels.

Lesson · A relation needs two signatures. Signing your own half changes nothing on the other side of the page.

Prepare the two permission labels in classical correlated and Bell-like states, then apply a strictly local flip at endpoint A and ask whether global permission is deterministically activated.

|00⟩ labels: max transfer 0 before and after the local T_A flip. |11⟩ labels: ≈ 1 before, 0 after. Bell-like (|00⟩ + |11⟩)/√2: max transfer 0.5 before the flip and 0 after. VERDICT: correlation and entanglement do not generically convert one-endpoint local control into deterministic remote permission — the pre-correlated cases either lose activation or deliver only the fraction already present in the state. This is exactly what no-signaling requires, and the toy reproduces it rather than evading it.

Lesson · Shared randomness is a shared memory of the past, not a shared hand on the switch.

Objects are K-bit labels, N = 2^K. The compact involution π_a(q) = q XOR a is specified by K = O(log N) bits yet defines N/2 disjoint pairings simultaneously.

For the antipodal mask of weight K, the ordinary hypercube distance between paired objects is K = log₂ N — the pairing is maximally non-adjacent in the ordinary metric while its description stays logarithmic. KEY VERDICT: algorithmic description compression can specify O(N) pairings without an O(N²) lookup table. But compression solves ADDRESS SPECIFICATION only. It says nothing about physical support or coupling, and no part of the compression is a mechanism.

Lesson · A short sentence can name a large number of things. Naming them does not connect them.

Implement q → q XOR a as an operation on an address register. Its support is Pauli-X on exactly wt(a) address bits, so the antipodal mask costs O(log N) address-bit operations.

The address rewrite is genuinely cheap — and genuinely does nothing to the remote subsystem that carries that address. Changing a pointer changes which object we are talking about; it does not act on the object. RETRIEVAL-IS-NOT-INFLUENCE remains undefeated through 240 phases, and Phase 240 is its sharpest statement: the cheapest known relational selector is cheap precisely because it is not physical influence.

Lesson · Turning the page of the phone book is free. It has never yet moved a house.

Promote the compact pairing rule into an explicit interaction, H_pair = g·Σ_q X_q X_π(q). Count the terms actually required.

N/2 pair terms are required. The selector DESCRIPTION stays O(log N) while the explicit physical interaction SUPPORT is O(N). Worked example at K = 18, N = 262 144: selector description 18 bits, explicit pair terms 131 072. VERDICT: algorithmic compression alone does not remove the extensive physical support bill. The compressed-selector branch survives only if a native lower-level algebra realizes the pairings from local primitives.

Lesson · One line of code, a hundred thousand wires. The compiler is not a physicist.

The compressed-selector branch asked for a native lower-level algebra whose LOCAL primitive terms collectively realize many pair relations without an explicit O(N) pair table. The dihedral group D_n with elements (k,b) is the minimal test: ordinary generators r, r⁻¹ define the cycle; adding one deeper primitive generator s defines the full dihedral Cayley graph. Generator count stays O(1) = {r, r⁻¹, s} while s alone creates n cross-sector adjacency relations.

Exact numerics at n = 8, 16, 32, 64, 128, 256: mean distance to one's s-partner in the deeper metric is EXACTLY 1 at every size; deep graph diameters are 5, 9, 17, 33, 65, 129. Crucially, the same-sector distance gain remains EXACTLY 1.0 — the reflection generator does not shorten any ordinary cycle distance in this construction. VERDICT: a compact native generator CAN compress physical adjacency description without an explicit O(N) pair table — Phase 242 partially passes the Phase 242 challenge. But this particular dihedral mechanism buys no ordinary-metric shortening: if s is physically usable, paired states are fundamentally adjacent in the deeper Cayley metric. This is emergent-geometry architecture, not faster-than-fundamental causality.

Lesson · One generator can carry a million relations — but it brings its own ruler with it, and by that ruler nothing moved.

State space Z_p for odd primes p; ordinary locality is x → x±1. Add two compact algebraic generators, x → 2x mod p and x → 2⁻¹x mod p. The generator set stays O(1) = {+1, −1, ×2, ×2⁻¹}; there is no stored pair table anywhere.

Diameter collapses from O(N) cycle scaling to numerically log-like scaling: N = 31: 15 → 5 (gain 3.0); 61: 30 → 7 (4.29); 127: 63 → 8 (7.88); 251: 125 → 9 (13.89); 509: 254 → 11 (23.09); 1021: 510 → 12 (42.5); 2039: 1019 → 14 (72.79). Sampled affine mean distances: ≈ 2.72, 3.38, 4.30, 5.13, 6.22, 7.27, 8.37. VERDICT: a constant-size algebraic generator set produces small-world accessibility without an explicit O(N) pair table — a strictly stronger result than Phase 242. But ×2 is itself a PRIMITIVE MOVE in the deeper algebra: ordinary cycle distance is not the fundamental causal metric of this toy. Nothing travelled faster than its own light cone; the light cone was redrawn.

Lesson · The map got shorter because we changed the map, not the territory. Call it what it is: a new geometry.

Fit the Phase-243 affine diameters to both candidate laws: d = a·log₂N + b and d = c·N^α, on the seven primes p = 31…2039.

Log fit: d ≈ 1.4154·log₂(N) − 1.8591 with R² ≈ 0.98942. Power fit: d ≈ 2.5284·N^0.2290 with R² ≈ 0.97826. Observed d/log₂N ratios: 1.0092, 1.1803, 1.1447, 1.1290, 1.2234, 1.2005, 1.2735 — bounded and slowly drifting. VERDICT: finite-size numerics are consistent with logarithmic/small-world growth but are NOT a proof of the asymptotic law; the ratio drift at the largest sizes is honestly reported. The result relocates locality to the deeper algebra rather than defeating fundamental causality.

Lesson · Two curves can hug seven points for different reasons. The fit is a witness, not a verdict.

Affine ring Z₃₁ with ordinary hops x → x±1 at unit strength. The deeper generator x → 2x is not removed; it is gated so that using it also raises an auxiliary label costing energy Δ. A low-energy probe launched from a site with the label down should then see only ordinary locality. Measured: maximum leakage into the raised-label subspace over t ∈ [0, 400], maximum transfer to the ×2 partner, and the second-order effective coupling g_eff = g²/Δ with its exposure time π/(2g_eff).

Hiding works, and it is priced. Leakage falls as Δ⁻²: 0.5675 (Δ=0), 0.2391 (4), 0.0613 (8), 0.0155 (16), 0.0039 (32), 0.00077 (64) — a clean factor-of-four drop per doubling. Transfer to the ×2 partner tracks it: 0.2045 → 0.00031. But the exposure time to actually USE the generator rises linearly: 6.28, 12.57, 25.13, 50.27, 100.53. The product (leakage) × (exposure time)² is constant to within numerical noise across the whole scan. VERDICT: the ×2 generator can be made arbitrarily invisible to low-energy probes, and every decibel of invisibility is repaid in activation time. This is the Phase-235 susceptibility tradeoff reappearing in generator form, not an escape from it.

Lesson · You can hide the door as well as you like. You will spend exactly what you saved when you try to open it.

Replace the energy penalty with an exact algebraic selection rule (Phase 236): H* = −g Σ_x X_{x,2x} ⊗ P₁, where P₁ projects on a single type label. In type 0 the ×2 term is exactly absent, so low-energy locality is EXACTLY the ordinary ring. In type 1 it is O(1). First-arrival times were measured for the pair (8 → 16), which is 8 hops apart on the ring and one ×2 step apart in the deeper metric.

Type 0 first arrival at thresholds 10⁻⁴ / 10⁻³ / 10⁻²: t = 2.283, 2.723, 3.333 — ordinary ring transport. Type 1: t = 0.013, 0.033, 0.103 — speedups of 176×, 82×, 32× on the same fixed H. Leakage in type 0 is exactly zero, not merely small, so the Phase-245 tradeoff is genuinely evaded in the narrow sense. VERDICT: the first half of the Phase-245 challenge PASSES. A single fixed Hamiltonian both hides ×2 perfectly from every probe in the inactive type and delivers an O(1) shortcut in the active one — but only relative to the ordinary ring metric, which this construction has already reclassified as non-fundamental.

Lesson · Exact rules beat approximate suppression every time. They also make the bill impossible to hide: the type label now owes everything.

The Phase-246 pass hangs entirely on the type label. Three candidate implementations were audited against the gates already standing: (a) one GLOBAL type qubit gating all n/2 doubling terms; (b) a per-edge label carried locally at each endpoint, requiring both endpoints permissive as in Phase 237; (c) a genuinely primitive relational degree of freedom q_AB attached to the pair rather than to either endpoint.

(a) FAILS the Global-Selector Accounting Gate. One qubit gating O(N) nonlocal terms is exactly the compressed rule with extensive support already rejected in Phase 241: its description is one bit, its physical coupling support is N/2 terms, and flipping it changes causal structure everywhere at once. (b) FAILS by Phase 237: with H* = g X_AB ⊗ P₁_A ⊗ P₁_B the label states 00, 10 and 01 give exactly zero transfer, so a single local flip at one endpoint cannot arm the remote edge — endpoint coordination is required, and coordination is bounded by ordinary first arrival. (c) is NOT REFUTED and NOT CONSTRUCTED: no toy in this programme has produced a q_AB that is primitive, sparse, and not secretly a stored pair table. VERDICT: the second half of the Phase-245 challenge FAILS for every selector we can build.

Lesson · The shortcut was never in the generator. It was always in whoever is allowed to say yes.

Phase 244 set the branch-deciding condition: hide ×2 from all low-energy probes while preserving ordinary locality, then expose it by an exact selection rule WITHOUT endpoint coordination, pre-arming, or a global selector that already contains the answer. Phases 245–247 tested each clause.

Hiding: PASSES exactly (Phase 246). Preserving ordinary locality: PASSES exactly. Exposure without endpoint coordination: FAILS (Phase 247b). Exposure without a global selector: FAILS (Phase 247a). Exposure without pre-arming: FAILS — the ×2 support sits in the fixed H from t = 0 in every construction. VERDICT: the compressed-selector branch is CLOSED under the standing gates. It is not closed as a theorem; it is closed as an engineering route, with one named survivor — a primitive relational degree of freedom q_AB that is not an endpoint label, not a global bit, and not a disguised pair table. Every future attempt on this branch must produce q_AB first.

Lesson · Closing a branch honestly is worth more than keeping it open dishonestly. We know exactly what would reopen it.

Ring of N qubits, nearest-neighbor edges. The candidate pool for a primitive q_AB is every two-local non-identity Pauli operator living on an edge: 9 per edge, 9N total. Each such operator belongs to exactly ONE unordered pair — its two endpoints. Measured: pool size, total unordered pairs N(N−1)/2, and the fraction of pairs that can be assigned a dedicated O(1)-support pair operator from strictly local pieces.

Coverage collapses as 2/(N−1). N = 8: 8 of 28 pairs covered (0.286). N = 31: 31 of 465 (0.067). N = 128: 128 of 8128 (0.016). N = 1024: 1024 of 523,776 (0.002). A strictly local construction can give a primitive q_AB only to pairs that are ALREADY nearest neighbors — exactly the pairs that never needed one. VERDICT: the local-pool construction of q_AB FAILS by coverage: it arms the adjacent and is silent about the remote, which is the ordinary ring metric restated.

Lesson · The local world already knows who its neighbors are. A q_AB that only exists for neighbors is the ring wearing a name tag.

Drop locality of the operator and keep only the requirement that each of the N(N−1)/2 pairs be independently armable. Counted: minimum description bits log₂(#pairs) versus the physical support the Global-Selector Accounting Gate charges for a selector over #pairs independent terms.

N = 31: 465 pairs, 8.86 description bits, charged support 961. N = 64: 2016 pairs, 10.98 bits, charged support 4096. N = 128: 8128 pairs, 12.99 bits, charged support 16384. The description stays logarithmic while the charged support grows as N² — this is precisely the Phase 241 compressed-rule shape: a short name over extensive support. VERDICT: independent arming for all pairs is a stored pair table in the accounting sense, regardless of how compactly it is described. The second construction class FAILS.

Lesson · Logarithmic words, quadratic bill. The gates cannot be impressed by how short the sentence is.

The strongest honest construction: a shared entangled pair between A and B is genuinely a degree of freedom of the PAIR — not reducible to endpoint labels, O(1) support per pair, and its correlations cannot be decomposed into anything either endpoint holds alone. Cost audit on the 31-site ring: distributing one Bell pair across ring distance d via a nearest-neighbor SWAP chain costs d swaps and time ∝ d.

Distribution cost equals ordinary first arrival at every separation tested: d = 1, 2, 4, 8, 15 → cost 1, 2, 4, 8, 15. The dynamical q_AB exists, is primitive, and is pair-owned — and acquiring it is exactly as slow as walking. If pre-created and stored, it FAILS the Pre-Arming Gate (Phase 185): the shortcut was paid at preparation time, at ordinary speed. VERDICT: the Bell-pair q_AB PASSES every property test and FAILS every speed test. It is a genuine relational degree of freedom that cannot beat the metric it lives in. This is the cleanest no-go the programme has produced: the object is real, and it does not help.

Lesson · We finally built the thing. It is beautiful, it is genuinely pair-owned, and it walks at exactly the speed limit.

Phase 248 named q_AB as the sole survivor that could reopen the compressed-selector branch, with four requirements: primitive, O(1) support per pair, not a stored pair table, and changeable without either endpoint acting alone. Phases 249–251 tested the three construction classes exhaustively: strictly local operators, arbitrary independent selectors, and dynamical pair resources.

Local pool: FAILS coverage — arms only the already-adjacent. Independent selector: FAILS support accounting — a logarithmic name over O(N²) charged support, i.e. the pair table. Dynamical pair resource: EXISTS and is primitive, but its creation cost equals ordinary first arrival and its pre-storage fails pre-arming. VERDICT: q_AB is REFUTED as a shortcut mechanism for every construction class this programme can build. One survivor remains, renamed one level deeper: a pair-owned degree of freedom that is EMITTED by the algebra's own dynamics at zero incremental cost — not stored, not distributed, but generated as a byproduct of evolution the system was already doing. No known algebra does this. Phase 253 will look for one.

Lesson · Three doors, three walls, and one window we can now describe precisely enough to look for.

N = 128 ring, nearest-neighbour XX dynamics, dispersion eps(k) = 2 cos k. Nobody prepares anything and nobody distributes anything: we let the system run the evolution it was already executing and measure the exact two-site propagator |G_d(t)|, the honest stand-in for a pair-owned resource emitted between sites separated by d. Recorded: first time the correlation crosses a fixed detection threshold of 1e-3, and the implied emission velocity d / t*.

Emission is real and it is strictly conical. d = 8 → t* = 1.66; d = 16 → 4.84; d = 24 → 8.32; d = 32 → 11.92; d = 48 → 19.32. The implied velocity falls monotonically toward the Lieb-Robinson group velocity of this dispersion, v = 2: 4.82, 3.31, 2.88, 2.68, 2.48. Outside the cone the emitted amplitude is not small — it is exponentially suppressed in d. VERDICT: free local dynamics DOES emit pair-owned resource for free, and it emits it at exactly the speed of ordinary first arrival. The emitter exists; the shortcut does not.

Lesson · The algebra was generous all along. It gives away pair correlation freely, at walking pace.

Same measurement, but the free dynamics is now power-law hopping J(r) = r^-alpha out to N/2, for alpha = 3.0, 2.0, 1.5. This is the only known class whose emission can outrun a linear cone, so it is the strongest form of the survivor we can actually build.

Arrival collapses as alpha falls. alpha = 3.0 reproduces the linear cone almost exactly (d = 48 → t* = 19.24 versus 19.32 for nearest-neighbour). alpha = 2.0 breaks it: d = 48 → 2.28. alpha = 1.5 shatters it: d = 4 → 0.02, d = 48 → 0.34, a sublinear arrival curve. VERDICT: the survivor class EXISTS numerically. Free dynamics with a long-range generator emits pair correlation at remote separation far earlier than any local cone permits. This is the first construction in the co-generation track that arrives early without anyone distributing anything.

Lesson · For one measurement we finally had it. Then we asked what the Hamiltonian cost.

The early arrivals of Phase 254 are threshold crossings. We now ask what is actually there: the emitted amplitude |G_d(t)| at a fixed early time t = 0.5, and the shot count 1/|G|^2 required to resolve it above statistical noise — the same accounting the programme applies to every other candidate channel.

The early signal is a direct coupling tail, and it is faint by exactly the amount the generator is weak at that range. alpha = 1.5: d = 8 → amplitude 2.3e-2, 1.9e3 shots; d = 48 → 1.5e-3, 4.3e5 shots. alpha = 2.0: d = 48 → 2.2e-4, 2.1e7 shots. alpha = 3.0: d = 48 → 4.5e-6, 4.9e10 shots. Shots-to-resolve grows as r^2alpha at fixed time, so every unit of earliness bought by lowering alpha is repaid in repetitions, and the total experiment time to certify the early arrival exceeds a single ordinary local traversal in every row of the table. VERDICT: the early arrival is real and the usable early arrival is not.

Lesson · Faster, fainter, and the two cancel with a precision that is starting to look like a law.

Phase 252 named the survivor as an algebra emitting pair-owned resources from dynamics it was already executing, at zero incremental cost. Phases 253–255 built exactly that object in two classes and audited both against the standing gates: Support-Invariance (183), Pre-Arming (185), Global-Selector Accounting (241), Pair-Resource Fare (251).

Local generator: emission is free and conical — ordinary first arrival, refuted by the Free-Emission Velocity Gate. Long-range generator: emission is free and early — but the r^-alpha couplings are pre-existing nonlocal support written into H before the run, which is the Pre-Arming Gate failure at the level of the generator rather than the state, and what it emits early is unusable under the Emission-Amplitude Fare Gate. VERDICT: the emission branch is CLOSED at this depth. The zero-cost emitter is not free; its cost was moved out of the run and into the Hamiltonian, where the programme's own accounting still finds it. One survivor is named, one level deeper again: a generator whose long-range terms are themselves emergent — not written in by hand but produced by a strictly local underlying algebra — which is the same question the programme asked at Phase 89 and has never answered.

Lesson · We moved the bill from the state to the Hamiltonian and the accountant followed it there. The programme has now closed the same loophole in five different costumes.

Visible sites carry no direct hopping at all. Each is coupled, strictly locally, to one site of a hidden ring whose only interaction is nearest-neighbour hopping 1 with a uniform gap Delta. Eliminating the hidden layer gives an exact induced visible coupling J_eff(r) = g² · [(Delta − H_med)^-1]_{0r}, evaluated by exact momentum sum on N = 128. If any gapped local layer induces a genuine power-law tail, the Phase 254 construction becomes legitimate.

Every gapped local layer we can build induces an EXPONENTIAL profile, and the gap sets the decay length. Delta = 3.0: |J_eff| falls 1.71e-1 → 9.52e-3 → 2.03e-4 → 9.18e-8 across r = 1, 4, 8, 16, correlation length xi = 1.04. Delta = 4.0 gives xi = 0.76, Delta = 6.0 gives xi = 0.57, Delta = 10.0 gives xi = 0.44 with |J_eff(16)| = 1.1e-9. The profile is exp(−r/xi), never r^-alpha, at every gap tested. VERDICT: the induced generator is short-ranged by construction, and the harder the layer is gapped — the very condition that makes the elimination valid — the shorter its range becomes.

Lesson · The gap that lets you eliminate the layer is the same gap that kills the tail. You cannot have one without the other.

The single escape from Phase 257 is to push Delta to the mediator band edge (Delta → 2), where the resolvent stops decaying exponentially. Same exact momentum sum, Delta = 2.1, 2.01, 2.001, out to r = 48.

The tail lengthens exactly as promised and the escape is real on paper. Delta = 2.1: 1.14 → 1.01e-2 → 6.56e-5 at r = 1, 16, 32. Delta = 2.01: 4.52 → 1.01 → 2.04e-1 → 4.29e-2 at r = 1, 16, 32, 48. Delta = 2.001: 1.59e1 → 1.02e1 → 6.62 → 4.81, barely decaying at all across the whole ring. VERDICT: a strictly local hidden layer CAN induce couplings of unbounded range — but only at the band edge, where the layer is gapless. At that point the elimination is invalid: the layer is no longer fast, no longer eliminable, and no longer a static generator. It is a dynamical channel with its own state, and the induced coupling diverges precisely because the mediator is on shell.

Lesson · The long tail exists. It is the mediator itself, standing in the doorway, holding a bill.

No elimination, no effective theory: the full visible+hidden Hamiltonian (N = 64 per layer, g = 0.5) is diagonalised exactly and the visible-to-visible propagator |G_{0r}(t)| measured against the same 1e-3 threshold used in Phases 253–255. This is the only test that cannot be gamed by a choice of effective description.

The full model is conical at exactly the mediator's own velocity. Delta = 4.0: r = 2, 4, 8, 16, 24, 31 arrive at t* = 0.58, 1.32, 3.16, 7.14, 11.18, 15.36 — implied velocity 2.02 at the largest separation, the Lieb-Robinson velocity of the hidden ring. Delta = 6.0 is slower still (18.42 at r = 31). Delta = 10.0, the deepest gap, does not reach the threshold at r ≥ 16 within t = 200 at all: the visible sites are effectively disconnected. VERDICT: whichever way the knob is turned, the emergent generator obeys the cone of the local layer that generated it. Gapping the layer buys locality and loses reach; ungapping it buys reach and hands the transport back to an ordinary local channel travelling at v ≈ 2.

Lesson · Every effective theory is a loan against a real one. We asked for the ledger of the real one and it balanced.

Phase 256 named the survivor as a strictly local algebra generating its own long-range terms. Phases 257–259 built exactly that object across the whole gap range and audited it against the standing gates: Generator-Cost (256), Emission-Amplitude Fare (255), Free-Emission Velocity (253), Pre-Arming (185), Support-Invariance (183).

The branch is a closed interval with a wall at each end. Gapped: induced couplings are exponential with xi < 1.1 at every gap tested, and the exact dynamics either travels at v ≈ 2 or does not arrive at all — Induced-Range Gate failure. Gapless: the induced range is unbounded, but the mediator is on shell, the elimination is invalid, and the exact dynamics is an ordinary local channel at v ≈ 2 — Un-Eliminated Dynamics Gate failure. There is no interior point that is both eliminable and long-ranged: the same parameter controls both, in opposite directions. VERDICT: the emergent-generator branch is CLOSED at this depth, and the closure is a TRADEOFF rather than a list of failures, which is a stronger form of wall than the programme has produced before. One survivor is named, one level deeper again: a substrate whose GROUND STATE, not its Hamiltonian, already carries long-range structure — critical or topologically ordered — so that no elimination is required because nothing has to be generated at run time. That object is not free either; it must be built once and it must then survive the Pre-Arming Gate, which every previous carrier of pre-existing nonlocal support has failed.

Lesson · Six costumes, six bills. The interesting part is that the last one was not a mistake we made — it is a constraint, and constraints are the only thing this programme is allowed to keep.

Half-filled XX chain at criticality, N = 4096, exact momentum sum for the static correlator C(r) = <c†_0 c_r>. This is the strongest form of the Phase 260 survivor available in one dimension: a vacuum with no gap, no length scale, and nothing that has to be built during a run.

The long-range structure is real and it is exact. Odd separations: r = 1 → 3.1831e-1, r = 3 → 1.0610e-1, r = 9 → 3.5367e-2, r = 33 → 9.6437e-3, r = 65 → 4.8930e-3, matching 1/(pi·r) to four significant figures at every distance. The fitted power-law exponent is −1.0002 across r = 3 … 65. There is no exponential envelope anywhere: unlike every induced generator of Phase 257, this substrate genuinely has algebraically decaying correlation at unbounded range, present before anyone does anything. VERDICT: the survivor class EXISTS, exactly, and it costs nothing at run time.

Lesson · For the second time in four phases we actually had the object. And for the second time the question was what it does, not what it is.

The same critical chain, now perturbed strictly locally at one site, with the remote response measured as the exact commutator norm |G_r(t)| against the identical 1e-3 threshold used for the nearest-neighbour ring in Phase 253 and the two-layer model in Phase 259. If pre-existing algebraic correlation is usable, this is where it shows.

The arrival times are not merely similar to the ordinary local chain — they are the same numbers. r = 4 → t* = 0.40; r = 8 → 1.66; r = 16 → 4.84; r = 32 → 11.92; r = 48 → 19.32, implied velocity 10.00 → 4.82 → 3.31 → 2.68 → 2.48, converging on v = 2. Every digit matches the Phase 253 gapped-free-emission table. VERDICT: a vacuum with unbounded correlation length responds to a local operation at exactly the same speed as one without. Correlation is not a channel. The structure is there, and it does not carry anything.

Lesson · The vacuum is wired everywhere and answers nowhere sooner. That is the cleanest statement of the wall we have found.

The remaining form of the survivor is a topologically ordered vacuum, where the long-range structure is entanglement rather than a two-point function and the natural nonlocal object is a string operator connecting two anyons. Accounting only — COUNTED, NOT SIMULATED, and labelled as such: no toric-code numerics were run here.

The accounting is decisive before any simulation. A string operator joining two sites at separation r is a product of Theta(r) local terms; applying it or measuring it is Theta(r) local operations, which is the Support-Invariance Gate (183) and the Global-Selector Accounting Gate (241) restated for the vacuum. The one genuinely nonlocal feature — that the endpoints share a superselection label — is exactly the object refuted in Phases 249–252 as a stored pair table. Topological order gives remote parties a shared invariant they can VERIFY, at a cost linear in the separation and only after ordinary local communication, which is the standing Retrieval-Is-Not-Influence result. VERDICT: no early response is available, and the branch does not require its own numerics to close.

Lesson · We keep meeting the same wall in a new alphabet. This time it was written in strings.

Phase 260 named the survivor as a vacuum carrying long-range structure at zero run-time cost. Phases 261–263 built the strongest one-dimensional instance exactly, audited it against Pre-Arming (185), Support-Invariance (183), Free-Emission Velocity (253) and Un-Eliminated Dynamics (259), and counted the topological case.

The vacuum branch closes, and it closes on the strongest possible terms: not because the structure is absent, but because it is present, exact, unbounded in range, free at run time — and produces no early response whatsoever. That is a stronger negative result than any cost argument the programme has made, because there is no bill left to point at. The whole sequence 253 → 264 now reads as one statement: in every construction we can build, the arrival time of a local operation's remote effect is set by the generator's Lieb-Robinson velocity and by nothing else — not by the state, not by the correlation length, not by the induced couplings, not by the topology. VERDICT: the pre-existing-support programme is CLOSED at this depth in all four of its costumes. The survivor named for Phase 265 is not another carrier: it is the assumption itself, that the generator is fixed and time-independent. Every phase since 253 has held H constant. A time-dependent, locally driven generator has a Lieb-Robinson velocity too, but it is not the same one, and the programme has never measured it.

Lesson · Four costumes, one velocity. The only thing we have never varied is the one thing the bound depends on — which is either a large oversight or the next wall.

Construct a single static routing Hamiltonian on a history chain of length L using the standard perfect-state-transfer spectrum J_n = J0·sqrt(n(L−n)), with the coupling bounded by Jmax = 1 so that no result is bought by making an interaction arbitrarily strong.

The construction is textbook mathematics used honestly: the chain has an equally spaced single-excitation spectrum, so the excitation mirrors end to end in a fixed time. With a history depth K, the transfer time is O(K) = O(log N) in the number of addressable destinations N = 2^K, at bounded Jmax. VERDICT: ESTABLISHED MATHEMATICS. This is not a new physical mechanism; it is a known engineered spin-chain result imported as the programme's transport primitive.

Lesson · The transport primitive is not the hard part and never was. The hard part is what the chain is made of.

Direct unitary evolution of the Phase 302 Hamiltonian in the single-excitation sector, measuring end-to-end transfer fidelity for K = 8, 12, 20 and 32.

Fidelity 1 at every tested depth in the ideal, noiseless, exactly engineered model. VERDICT: the toy model does what the algebra says it does, which closes an implementation question and opens no physical one.

Lesson · Ideal means ideal. The next four phases exist to say how quickly that word stops being free.

K = 20 history chain with independent multiplicative noise applied to every coupling J_n, averaged over realisations, measuring mean transfer fidelity as the noise amplitude is swept.

Mean transfer fidelity 0.999976 at 0.1% noise, 0.99736 at 1%, 0.98964 at 2%, 0.93812 at 5%, 0.77668 at 10%. The degradation is smooth and roughly quadratic in the noise amplitude at small noise, which is the expected engineered-chain behaviour. VERDICT: CONDITIONAL PASS — the architecture tolerates sub-percent fabrication error and dies at tens of percent. It is an engineering tolerance statement, not a physical result.

Lesson · Every perfect number in this ledger is a limit, and the limit has a slope. Publish the slope.

Transport an arbitrary qubit state rather than a bare excitation, under the factorized generator H = I_payload ⊗ H_route, and measure both payload fidelity and destination arrival probability.

Payload fidelity 1 and destination probability 1 for arbitrary input states. This is a structural consequence of the factorization, not a discovery: if the router never touches the payload register, the payload cannot decohere from routing. VERDICT: PASSED IN THE IDEAL FACTORIZED MODEL, and the factorization itself is the assumption under test in Phase 306.

Lesson · A model that assumes no cross-talk proves no cross-talk. So we broke the assumption on purpose.

Break the factorization of Phase 305 by adding normalized cross-talk coupling between the payload register and the routing chain, sweeping the cross-talk strength and averaging payload fidelity over random input states.

Mean payload fidelity ≈ 0.9974 at 1% cross-talk, ≈ 0.9906 at 2%, ≈ 0.9434 at 5%, ≈ 0.8154 at 10%. Payload coherence is more fragile to cross-talk than transfer probability is to coupling disorder. VERDICT: payload preservation is CONDITIONAL on strong payload/router isolation, and isolation is now a first-class design constraint of the architecture.

Lesson · The router does not have to fail to ruin the message. It only has to listen.

Attach a passive receiver site to the end of the router and attempt to capture the arriving excitation without co-designing the receiver into the transfer spectrum.

Best achieved capture ≈ 0.95938, with the residue reflecting back into the chain and re-emerging. No tuning of the naive receiver reached unit capture. VERDICT: FAILED — destination capture is not a bolt-on. The receiver must be part of the same engineered spectrum as the route.

Lesson · A door is not a wall with a sign on it. Capture has to be designed, not attached.

Rebuild the whole chain as ONE static Hamiltonian in which the visible source, the hidden route and the visible destination are all part of a single PST spectrum, with the bound Jmax = 1 retained.

Perfect ideal transfer. K = 20: t = 17.278760, P = 1, Jmax = 1. K = 32: t = 26.703538, P = 1, Jmax = 1. The Phase 307 failure is fully repaired by co-design, and the transfer time still grows as O(K) = O(log N) at bounded coupling. VERDICT: END-TO-END AUTONOMOUS TRANSPORT PASSES IN THE ENGINEERED IDEAL TOY MODEL — and only there.

Lesson · The whole path, one spectrum. That is the first architectural statement the programme has that actually holds together.

Send an arbitrary payload qubit through the complete Phase 308 portal/router/receiver Hamiltonian in the ideal factorized model.

Payload fidelity 1 and visible destination probability 1. The complete architecture transports quantum information, not merely an excitation. VERDICT: PASSED in the ideal factorized model; the Phase 306 cross-talk numbers remain the governing robustness statement.

Lesson · Ideal end to end, twice. Now find out what a bad portal costs.

Detune the portal couplings away from their co-designed values by 1%, 2%, 5% and 10% and re-measure mean transfer fidelity through the complete chain.

Mean F ≈ 0.999572 at 1%, 0.998459 at 2%, 0.989814 at 5%, 0.960341 at 10%. Portal mismatch is markedly gentler than bulk coupling disorder (Phase 304) and gentler than payload cross-talk (Phase 306). VERDICT: the conversion interface is the most forgiving element of the architecture; the payload isolation requirement is the tightest.

Lesson · The doorway is not where this breaks. Good — that was the part we most expected to break.

Ask whether ONE time-independent Hamiltonian can contain 2^K address sectors, each performing exact transfer to a different target, selected by an address register.

Mathematically yes: a direct sum over address projectors gives exact target transfer in every sector from a single static generator. But the construction as written hides O(N) address projectors inside the Hamiltonian. VERDICT: EXISTS AS MATHEMATICS, REJECTED AS AN EXPLANATION until the projector count is audited — which is Phase 312.

Lesson · Anything is one Hamiltonian if you are allowed to write down exponentially many terms.

Count the terms. Compare the naive full-address projector construction against a proxy count for a factorized local-bit-rule construction at K = 20.

Naive construction: 1,048,576 projectors at K = 20. Local-bit-rule proxy: ~40 terms. The naive route is rejected as a scalable explanation by five orders of magnitude at a depth the programme routinely simulates. VERDICT: FAILED — description complexity is a physical charge in this programme, under the standing Support-Invariance and Global-Selector Accounting gates.

Lesson · Writing the answer into the Hamiltonian is not deriving it. The audit exists to stop us doing that quietly.

Replace the global target lookup with a levelwise rule: at routing level i the Hamiltonian tests ONLY the current address bit using the projectors P_i^(0) and P_i^(1), with no term referring to the full destination label. Explicit K = 3 numerical test over all eight addresses.

All 8 addresses route onto unique invariant paths with ZERO wrong-branch leakage measured. Term count grows linearly in K rather than exponentially. VERDICT: ADDRESS LOOKUP COMPRESSION PASSES AT THE LOGICAL RULE LEVEL. Remaining charge: the explicit tree geometry is still handed to the model rather than generated by it.

Lesson · The lookup table is gone. The tree it was pointing at is still sitting there in plain sight.

Describe the routing hierarchy not as a stored tree but as the free monoid on two generators {L, R}, where a constant-size append rule generates the entire exponential hierarchy.

The O(N) description of the tree collapses to two generators and one rule. The exponential structure is now generated rather than tabulated. VERDICT: EXPLICIT TREE DESCRIPTION SUBSTANTIALLY REDUCED at the description level. This does NOT reduce physical capacity or physical support, and must not be reported as if it did.

Lesson · Compressing the map is not the same as shrinking the territory — which is exactly what the next phase checks.

Separate the description charge from the capacity charge. N distinguishable destinations require a Hilbert space of dimension N; how many degrees of freedom is that?

Dimension N is carried by only log2 N qubits. Explicit O(N) hardware nodes are therefore NOT information-theoretically mandatory. VERDICT: the O(N) node count that the programme has been charging since Phase 241 is an artefact of one implementation, not a theorem. This is the single most encouraging result in this block, and it is a statement about counting, not about any physical system.

Lesson · A genuine loophole, honestly earned: we were charging for hardware the information theory never asked for.

Test the Phase 315 loophole against generic transformations rather than the structured one we designed.

Compact encoding helps only for STRUCTURED transformations. A generic permutation of N states has description complexity exponential in log N and no compact generator set; almost all permutations are incompressible. VERDICT: the loophole survives only for structured algebraic action. Any viable mechanism therefore needs SIMPLE PRIMITIVE ALGEBRAIC GENERATORS that look nonlocal only after reconstruction in the ordinary metric — which is now the sharpest statement of the programme's core hypothesis.

Lesson · The escape route only fits things with structure. Fortunately, physics is one of those things.

Define, before running anything, what would count as a relocation rather than a relabelling.

A valid relocation must (i) preserve the payload state, (ii) ESTABLISH the correlations characteristic of the target neighbourhood, and (iii) REMOVE the stale correlations with the source neighbourhood. A mere location-label flip satisfies none of (ii) or (iii) and is rejected. VERDICT: ADOPTED AS A GATE. Several earlier candidates in this programme would have failed it, and they stay on the record.

Lesson · We finally wrote down what we would even accept. That should have been Phase 3.

Construct an explicit SWAP-based unitary that moves a Bell-type relational correlation of object P from environment A to environment B while leaving P's payload untouched, and measure both bond fidelities.

P–A Bell fidelity 1 → 0.25 and P–B 0.25 → 1, with payload preserved. Relational correlation transfer is therefore POSSIBLE in ordinary toy quantum mechanics and satisfies the Phase 317 gate. VERDICT: PASSED — and note carefully that this uses only standard local unitaries and implies nothing about spacetime.

Lesson · The relational picture is at least coherent. That is a lower bar than it sounds, and we had not cleared it before.

Charge the Phase 318 construction. How does the cost scale with the number of local bonds m that must be re-made?

Relocating m finite local bonds costs O(m). If m is a FIXED local coordination number, the cost is O(1) independently of the macroscopic separation R — the first genuinely R-independent cost in the programme. If, however, the relational dressing grows with system size or with distance, the shortcut collapses immediately. VERDICT: CONDITIONAL — the entire hypothesis now rests on whether physical dressing is finite. Phases 320–322 test exactly that.

Lesson · One number decides this: does the dressing grow with distance? Everything else is decoration.

Take the simplest gauge-theoretic case: a Z2 charged excitation, and count the support required to relocate it under an ordinary-local implementation.

A charged excitation requires flux-string support scaling as O(R), because Gauss's law forbids a bare charge from moving without its string. Bare charge relocation therefore RESTORES LINEAR COST and destroys the Phase 319 O(1) result. VERDICT: FAILED for bare gauge charge. The dressing does grow with distance, exactly as feared.

Lesson · Physics answered the Phase 319 question and the answer was no — for anything carrying a charge.

Ask whether the Phase 320 wall applies to a NEUTRAL composite: a fixed-size +/− pair whose far-field electric charge cancels.

For a fixed-size neutral composite of extent ell, the far field cancels and the local dressing cost can remain O(ell) — constant in R. The Gauss-law wall is evaded, not by breaking the constraint, but by satisfying it trivially. VERDICT: SURVIVES AS MODEL INTUITION ONLY. This is a Z2 toy argument, NOT a proof for realistic gauge theories, and it is recorded as a hypothesis, not a result.

Lesson · Neutral things travel light. Whether that survives contact with real field theory is not something we have shown.

Apply the Phase 321 escape to gravity, where the relevant charge is mass-energy.

Mass-energy does not cancel the way electric charge does; there is no neutral gravitational composite. The Phase 321 escape does not transfer to the gravitational sector. CONSEQUENCE, adopted as programme direction: the first experimental objective is NOT macroscopic matter relocation. It is COHERENT INFORMATION TRANSFER BETWEEN STATIONARY ENDPOINT HARDWARE using neutral, nearly degenerate logical states.

Lesson · We narrowed the ambition to the only thing the accounting permits. That is a demotion and it is the right one.

Direct comparison at fixed time: an ordinary nearest-neighbour chain versus the hidden PST chain of Phase 308, measuring target arrival probability at separations R = 32 and R = 256.

R = 32: hidden arrival t ≈ 5.4414 with P = 1, while the ordinary chain's target probability is ≈ 3.18e-25. R = 256: t ≈ 7.854, hidden P = 1, ordinary P ≈ 2.18e-33. THIS IS NOT FTL. The hidden edges are part of the full Hamiltonian and therefore define the actual causal graph; the transfer is exactly at the Lieb-Robinson velocity OF THAT GRAPH. The separation measures the mismatch between two metrics, not a violation of one. CONCLUSION: if the microscopic Hamiltonian is local in the ordinary 3D metric, there is NO genuine O(log R) controllable shortcut. The hypothesis survives only if PRIMITIVE LOCALITY ≠ RECONSTRUCTED ORDINARY LOCALITY.

Lesson · The most important sentence in this whole block is a denial: nothing here goes faster than its own light cone.

Propose the minimal compact algebra with two sectors of different growth: an ordinary sector Z^3 with polynomial volume growth ~r^3, and a hidden sector F_2 (free group on two generators) with exponential growth ~3^r.

The candidate removes the explicit hidden tree and coordinate table at the description level: both sectors follow from four generators and their relations. But the direct product HAND-INSERTS the number 3. The ordinary sector's dimensionality is put in, not derived. VERDICT: DESCRIPTION-LEVEL PROGRESS, NO-INSERTED-GEOMETRY GATE STILL FAILED.

Lesson · We stopped writing down the tree and started writing down the 3. Progress, of a slightly embarrassing kind.

Quantify what the two sectors of Phase 324 would mean for distance: polynomial sector r_o ~ N^(1/3) versus exponential sector r_h ~ log N, evaluated at a proxy scale N = 2^48.

At N = 2^48: r_o ≈ 65,536 versus r_h ≈ 29.65, a ratio of ≈ 2,210. The contrast is the whole content of the hypothesis, stated numerically for the first time. The CORE UNRESOLVED PHYSICS is untouched by it: (i) what dynamically SELECTS a sector, and (ii) how ordinary operational locations MAP onto states of the common primitive algebra. Neither has a model.

Lesson · A ratio of two thousand, and no idea what chooses which side of it you are on.

Attack the inserted 3 directly and blindly: enumerate small nilpotent groups by their lower-central-series rank tuples and compute the Bass–Guivarc'h growth degree D = sum k·r_k, with NO reward, filter, or scoring term that favours D = 3. If polynomial growth selects three dimensions, the sieve should show it without being told.

PRELIMINARY: it does not. Growth degrees 1, 2, 3 and higher are all naturally available from small rank tuples, with nothing distinguishing D = 3 among them. Polynomial-growth algebra ALONE does not uniquely select D = 3. VERDICT: 'derive 3' remains OPEN, and naive algebraic growth FAILS as a dimension selector. This is a negative result about our own most-favoured route and it stays on the page.

Lesson · We built the sieve so it could not flatter us, and it did not. The number 3 is still an input, not an output.