Applications / Foundation papers
Entanglement as geometric correlation
The correlation of an entangled pair is the inner product of the two analysers’ axes, read over one shared helical phase: E = −a·b. An inner product is stored at neither vector and travels between neither. Two patterns of shared bivector grade carry phases locked half a turn apart; each wing’s record is the eigenvalue ±1 of its own clamp, read from its own clamp and the phase it receives; and the pair with its two clamps is one system whose record is one settlement, on the projection measure of the clamp that hosts it, |cos(φ − a)|/4 (I-4, 0.1a, P-2). The paper proves the converse. With the lock and two local sign records, that measure is the only one a host clamp can set that returns the table with even shares at each wing, and every other measure that turns with its clamp and returns the same correlation lets the far wing’s share read the host clamp. No communication is needed, and its absence is the condition that selects the measure. Bell’s inference stands; one premise is declined, measurement independence, at the exact price 2√2 − 2. The table of correlators is a Gram matrix of rank at most three, the way a sequence model stores a graph it has memorised, and a determinant separates it from the table of a clamp-independent phase with no inequality: 0 against −16/27. Under transport the correlation is −aTOrelb, and the mean relative rotation of the two wings’ frames fixes what a pooled pair shows: its trace the fidelity with the emitted pair, its signed singular values the purity, the negativity and every CHSH ceiling.
The record
The correlation, the tests and a second record from machine learning
Two patterns prepared together and read far apart return records whose correlation no list of answers assigned in advance, independently of the analysers, reproduces. Einstein, Podolsky and Rosen drew from the perfect correlations the conclusion that the quantum description is incomplete. Bell proved that a completion by a state whose statistics are independent of the analyser settings, with each record a function of its own setting, obeys an inequality that the quantum table violates.
The record is settled. Aspect, Dalibard and Roger changed the analysers while the pair was in flight. Hensen and colleagues found S = 2.42 ± 0.20 on 245 event-ready trials of two electron spins 1.3 kilometres apart, with the photonic tests of Shalm and of Giustina and their colleagues beside it. Yin and colleagues distributed photon pairs from orbit to receivers 1 203 kilometres apart and found S = 2.37 ± 0.09 on coincident pairs. Rauch and colleagues set their analysers by the colour of light from quasars: on coincident pairs, under fair sampling, the table exceeds the local bound by 9.3 standard deviations, and any common cause of the settings and the pair’s state is pushed back at least 7.8 billion years. FDT accepts this record in full.
A second record comes from machine learning. Noroozizadeh, Nagarajan, Rosenfeld and Kumar separate two ways a sequence model stores the edges of a graph it has memorised. In associative memory each entity has an arbitrary key and the relation sits in a weight matrix: a lookup of the pairs seen in training. In geometric memory the relation sits in the positions of the entities themselves, f(u)[v] = Φ(u)·Φ(v), and the inner product of two entities that never appeared together encodes the model’s own multi-hop distance between them. They report that Transformers and Mamba trained on local edges alone build the second kind of storage, on path-star graphs of tens of thousands of nodes, and they pose as an open question why gradient descent prefers it.
| Quantity | Value | Kind |
|---|---|---|
| Sign records on the projection measure, CHSH | 2.828427 exactly; 2.828730 ± 0.001414 with wing A’s clamp as host, 2.829470 ± 0.001414 with wing B’s, 2.827638 ± 0.001414 with either | closed form; simulated, 4 000 000 records each, every record kept |
| Sign records on the clamp-independent measure, CHSH | 2.000000 exactly; 2.000188 ± 0.001732 | closed form; simulated, 4 000 000 records, every record kept |
| The two conditions solved for the measure on 360, 720 and 1 440 phase points | |cos|/4 to 3 × 10⁻⁶, 8 × 10⁻⁷ and 2 × 10⁻⁷; the correlation alone leaves 360 of 720 unknowns free | linear solve |
| The price of the clamp | 2√2 − 2 = 0.828427; 0.2088 bits | closed form and quadrature |
| The same weight used as a filter on a uniform phase | 2.828682 on accepted pairs at a wing efficiency of 0.636600; 1.800740 with every record counted | simulated |
| Determinant of the 3 × 3 table at clamps 0, 60° and 120° on both wings | 0 on the projection measure, −16/27 = −0.592593 on the clamp-independent one; −0.002 ± 0.036 and −0.592 ± 0.024 over 600 runs of 2 000 records per cell | closed form; sampled |
| Pooled pair, independent fields on the two arms at αn = 0.4 | negativity 0.309728; CHSH 1.750213 at the fixed quartet and 2.353295 ± 0.001082 at the axes that are optimal over the sphere, against 2.352637 | simulated, 1 536 paired histories; 2 000 000 records per pair of settings |
| Hensen and colleagues, 245 event-ready trials | S = 2.42 ± 0.20; the kernel at those axes gives 2.818997, a clamp-independent phase 2 | measured; closed form |
The FDT reading
I-4 the pair’s record, 0.1a one settlement, P-2 the clamps set the mode
The lock, the clamp and the record
FDT has one substance, the omnium, and every pattern in it moves as a helix. The wavefunction is the projection of the helix onto the plane normal to its axis, and superposition is ignorance of the helical phase at observation: the phase is definite at every instant (I-4). Two patterns share bivector grade when their helical phases are locked. The equal-weight pair of opposite sense carries φ and φ + π on the phase circle, the doubled analyser angle for a polarisation and the Bloch angle for a spin; in consensus names it is the singlet. The lock is a relation between two phases. It is a property of neither wing, and it has no dependence on where the wings are.
A clamp is an analyser, a boundary condition (P-2): the clamp at a is the cut cos(φ − a) = 0 of the phase circle. Each wing’s record is the eigenvalue of the channel it leaves by, +1 on one side of the cut and −1 on the other, and the correlation is the mean product of the two eigenvalues over every record:
One system, one settlement
Every step is a system-wide settlement: the omnium settles the whole system from the state it is in, and a response is never computed pattern by pattern or pair by pair (0.1a). The mode geometry of a system is set by its boundary conditions, the clamps (P-2). The pair and its two clamps are therefore one system, and the pair’s record is one settlement of it. For the equal-weight pair the settlement lands uniformly across the projected aperture of the clamp that hosts it:
The correlation is an inner product
On that measure the two records return the cosine on the phase circle and, on the sphere of the helix chart, the inner product of the two clamp axes:
Each wing’s share of +1 is 1/2 at every pair of clamps, for either host and for every mixture of hosts. On the sphere the measure is |a·η|/2π, uniform on the disc normal to the host’s axis, and the proof is one line: |a·η| sgn(a·η) = a·η, so the correlation is −a·∫η sgn(b·η) dΩ/2π, and by the symmetry about b the integral is 2πb. At equal clamps the two records are opposite at every phase.
The measure is forced
Those results start from the measure and return the table. The converse holds: the table returns the measure. Keep the lock and the two sign records, and let the shared phase have any probability density that the host clamp sets. Ask two things at every position of the remote clamp: that the correlation be the kernel, −cos(a − b), and that the remote wing’s share of +1 be 1/2. Then the density is |cos(φ − a)|/4. On the sphere the same two conditions at every remote axis give |a·η|/2π.
The proof differentiates. As the remote clamp turns, its record changes sign at two phases half a turn apart, so the remote share changes at a rate set by the difference of the density at those two phases. A flat remote share therefore makes the density repeat every half turn, and the kernel then fixes its value at every phase. On the sphere the map from a density to the remote wing’s mean record annihilates even functions and is one-to-one on odd ones, by the Funk–Hecke formula, and the same result follows.
Each condition does its own work. The kernel alone fixes the part of the density that is even under a half turn of the phase and leaves the odd part free. With u = φ − a and Δ = a − b, the measures
all return E = −cos Δ. For ε ≠ 0 the remote wing’s share depends on the host clamp through Δ: a signal passes. The same holds for every measure that turns with its clamp and has a part that is odd under a half turn. Among the measures that turn with a host clamp and return the kernel, the projection measure is the one through which no signal passes. The lock, two local sign records, a measure set by one host clamp and the table with even shares at each wing leave no other.
Why no communication is needed
Communication means a signal: a dependence of anything one wing records on the clamp of the other. Each record is local, a function of its own clamp and the phase its wing receives. No record carries a reading of the remote clamp: each wing’s share is 1/2 at every pair of clamps and for every host, and whatever is done to one clamp, the frequencies at the other wing are unchanged. That is the condition that selects the measure, never an assumption added to it. The table is the same whichever clamp hosts the settlement, and the statistics have no dependence on which record is made first or on the frame in which their order is judged.
The kernel contains the two clamp angles and nothing else, so it has no dependence on the separation of the wings. The record agrees: the local bound is exceeded with the wings 1.3 kilometres apart and, on coincident pairs, 1 203 kilometres apart. No force carries the correlation either. A wing’s own gravitational window at the other wing is αext = 4.1 × 10⁻⁶³ at one metre for a stroke of wavelength 810 nanometres. The small number is a correct reading (I-1); it falls with the separation, and the kernel does not.
The correlation is stored where an angle is stored. The angle between two directions is located at neither direction and travels between neither: it is a property of the pair of them. The kernel −a·b is that property for the two clamp axes, read over one shared phase.
The measure is forced. What each wing sees: the slider sets ε, the part of the measure that is odd under a half turn of the phase. The left panel is the correlation of the two records, computed by quadrature on the chosen measure with no closed form in the curve; it lies on the grey track, −cos Δ, at every ε. The right panel is the remote wing’s share of +1, computed the same way. At ε = 0, the projection measure, it is flat at 1/2. At any other ε it rises and falls as the host clamp turns, so the remote wing reads the host clamp from its own counts. The family: the chosen measure, solid, against the projection measure, dashed; the shaded area is the odd part, which the kernel leaves free and the flat share removes. Records: the page draws records one at a time from the chosen measure with a fixed seed, with the host clamp at a and at a + 180° in turn, and enters every record in the statistic. The bars are the host wing’s share and the remote wing’s share at the two positions of the host clamp; the ticks are the exact values.
I-4, 0.1a, P-2
Bell and EPR
Use 6 One premise declined, and what it costs
What stands
Let the measure of the shared phase be independent of the clamps, and let each record be a function of its own clamp and the phase. Read all four clamp positions on one phase: 𝒜₀(ℬ₀ + ℬ₁) + 𝒜₁(ℬ₀ − ℬ₁) is +2 or −2, because one bracket is 0 and the other ±2. Averaging over one measure keeps the mean between −2 and 2. That is the inference of Bell’s theorem, and it stands. On the uniform measure the same two sign records give the triangle, E = −1 + 2|Δ|/π, and a CHSH value of exactly 2. The projection measure does not have that form: its four correlators are averages over the measures of two positions of the host clamp, and the bound that needs one measure for all four does not constrain them.
The price, in three currencies
Distance between measures. Hall’s measure of the dependence of the measure on the host clamp is the largest distance between the measures two positions of the clamp give, 𝔇₁ = sup ∫ |p(φ | a) − p(φ | a′)| dφ, written M₁ on the Bell-correlations page. For host clamps an angle θ apart the distance is 2√(1 + sin θ) − 2, largest at a quarter turn: 𝔇₁ = 2√2 − 2 = 0.828427. For any records ±1 on a measure set by one wing’s clamp, S ≤ 2 + 𝔇₁, and the settlement meets the bound with equality at every pair of host clamps: written with the clamps as unit vectors on the circle, the largest CHSH value over wing B’s clamps is |a + a′| + |a − a′| = 2√(1 + sin θ). The excess of the settlement over the local bound is the distance between the measures of the two host clamps, never more and never less, and Tsirelson’s value is the ceiling.
Bits. The mutual information between the phase and a host clamp uniform on the circle is log₂(π/2) + 1 − 1/ln 2 = 0.2088 bits, and for two equally likely host clamps a quarter turn apart it is 0.2018 bits.
What a state fixed at emission needs in its place. An account that fixes the hidden state at emission, apart from the clamps, returns the same table only with a further resource. One bit of communication between the wings suffices. So does selection: the weight |cos(φ − a)| applied as a local acceptance on a uniform phase returns the kernel on the accepted pairs at a wing efficiency of 2/π = 0.6366. The simulation finds 2.828682 on accepted pairs at a wing efficiency of 0.636600, and 1.800740 on the same run once every record is counted. The loophole-free record excludes both for the laboratory table: the settings are space-like separated from the remote record, and every heralded trial is counted. The settlement uses neither. It keeps every record, passes no signal between the wings, and carries the same weight as the measure of one system.
A measure free to depend on both clamps returns the singlet’s table at a third of that distance, 𝔇₁/3 = 0.276, giving up 14 per cent of measurement independence, and reaches S = 4 at larger distances. The settlement has one host. Its measure is set by one clamp, it sits at the optimum for such measures, with 1 − 𝔇₁/2 = 2 − √2, 59 per cent, of measurement independence kept, and it stops at 2√2.
The measure’s one argument
The measure has the host’s clamp angle as its one argument, and no dependence on what chose the angle: a random-number generator, a switch driven while the pair is in flight, or the colour of light that left a quasar billions of years before. Those tests bound an account in which the angle and the state share a cause in their past. The measure is written on the angle itself and names no such cause.
EPR: two things asked, a third assumed
The argument of Einstein, Podolsky and Rosen asks two things of a complete description and assumes a third. It asks for a definite state behind the perfect correlations: the helical phase is that state, and wing B’s record at every clamp is the function −sgn cos(φ − b) of it. It asks that nothing done at one wing reach the other: each record reads its own clamp and its own phase, and no signal passes. It assumes, as Bell’s formulation makes explicit, that the state of the pair is independent of the analysers that will read it. That third premise is the one the chain does not grant: the pair and its clamps are one system (0.1a), and the measure of the shared phase is set by the host clamp. The first two are met together with the full quantum table. The phase is the completion Einstein asked for; Bell’s theorem fixes the form the completion takes, and the uniqueness result names it.
The clamp that hosts the settlement sets the measure of the shared phase. The record at that clamp then tells on which side of the cut the phase lies. Reading the record at wing A changes nothing at wing B. What changes is what the reader of wing A knows about a phase that was definite throughout.
Geometric memory
The table of correlators is a Gram matrix
The kernel E(a, b) = −a·b has the form f(u)[v] = Φ(u)·Φ(v). The entities are the clamp settings, the embedding of a setting is its axis on the sphere of the helix chart, and the relation of two settings is the correlation of the pair read at them. For any axes at wing A and any axes at wing B the table of correlators has rank at most three, and rank at most two when the axes lie on one phase circle. Every 4 × 4 table of correlators, and every 3 × 3 table on a phase circle, has determinant zero.
The table is fixed by positions, not by a list. With four axes per wing there are sixteen correlators, and at generic axes the sixteenth is fixed by the other fifteen. The correlator of two clamps that were never set together follows from the axes: the exact counterpart, for the pair’s record, of the inner product of two entities that never co-occurred. Tsirelson’s theorem states the general fact: the tables of correlators of two-valued quantum observables on two wings are exactly the tables of inner products of unit vectors.
Where the parallel stops
The embedding belongs to the clamp setting, not to the pair. Give instead each pair one fixed vector, uniform on the sphere and independent of the clamps, and let each wing record the sign of its projection on its own axis. The records then round the inner product to its arcsine, E = −(2/π) arcsin(a·b), which is Grothendieck’s identity and, on a phase circle, the triangle. The rounding destroys the low rank. At the clamps 0, 60° and 120° on both wings, analysers 30° apart for a polarisation, the two tables are
| −1 | −1/2 | 1/2 |
| −1/2 | −1 | −1/2 |
| 1/2 | −1/2 | −1 |
| −1 | −1/3 | 1/3 |
| −1/3 | −1 | −1/3 |
| 1/3 | −1/3 | −1 |
At the four axes of a regular tetrahedron on both wings the kernel’s table has determinant 0 and the arcsine table 0.6316. On the projection measure the whole table, at any number of clamps, is three coordinates per clamp. On a clamp-independent phase the table has no finite-dimensional inner-product form, because the arcsine carries every odd power of a·b, and at generic clamps its rank grows with the number of clamps: nothing is compressed. In the terms of the memory result, the first is geometric memory, and the second is a table in which each new clamp adds a row that the others do not fix.
A test with no inequality
The zero of the determinant holds at any three clamps per wing on a phase circle, with no calibration of their angles, and it survives every reduction of visibility that leaves the table bilinear in the clamp axes: white noise, and the unresolved transport of the next section. With N records per cell the counting standard error of the determinant on the unit-visibility kernel at these clamps is 1.59/√N, so 181 records per cell separate the two values by five standard errors. In 600 sampled runs of 2 000 records per cell the determinant is −0.002 ± 0.036 on the projection measure and −0.592 ± 0.024 on the clamp-independent one.
The determinant test. Two tables: each cell is the correlation of the pair at one pair of clamps, with the clamps at 0, θ and 2θ on each wing and θ set by the slider. On the projection measure the table is a Gram matrix of three unit vectors in one plane, so its determinant is 0 at every spacing. On a clamp-independent phase the same records give the triangle’s table, of rank three. The spacing: the two determinants against θ, computed from the tables. The clamp-independent determinant is largest in size at 60°, where it is −16/27; the band is five counting standard errors about zero at 181 records per cell, and it reaches that value there. Records: the page draws records one at a time into each of the nine cells, on each measure, with a fixed seed, and enters every record in the statistic. The points are the determinants of the two sampled tables with their counting standard errors; the solid line is 0 and the dashed line the clamp-independent value.
I-4
Shared density gradients
I-4, P-4 Conservation, ignorance, fragility
Only the relative rotation enters
Between the origin and the clamps each wing crosses the density gradients of its own path, and its transported frame arrives turned: on the sphere of the helix chart by a rotation OA for wing A and OB for wing B. Each wing’s record reads its clamp axis in the frame it arrives with, and the kernel becomes
with both shares at 1/2, for either host. A turn common to both wings leaves every cell of the table unchanged. The generator is the gradient. The transport is that of ray optics, computed in the companion paper: in a medium of measured index n the twist-free frame along a ray turns with B = t ∧ ∇ln n, later increments multiplying on the left. Density alone rotates nothing, because multiplying the index by a constant leaves the gradient of its logarithm, the rays, the rotors and the records unchanged. Order matters: rotors about different planes do not commute (0.1a), so two paths with equal histograms of the density, equal fluctuation amplitude and equal optical length end with different frames when the order differs. Two wings that leave a shared origin with matched frames and cross gradients that coincide pointwise and in order arrive turned alike, and the pair’s table at the clamps is the emitted table exactly.
The pooled pair
When the two histories are not recorded, the records are pooled over an ensemble of relative rotations. One matrix then carries everything, the mean relative rotation M = ⟨Orel⟩, and the pooled table is E = −aTMb with even shares. Let s₁ ≥ s₂ ≥ |s₃| be the signed singular values of M, with s₃ carrying the sign of its determinant. The pooled table is that of a source emitting four equal-weight pairs, the emitted pair and its three images under a half-turn of one frame about a coordinate axis, in frames turned by two fixed rotations, with frequencies
They are the mean squared components of the unit quaternion of each history’s relative rotation, so they are non-negative and sum to one, and w₀ is the largest fidelity of the pooled pair with any equal-weight pair. The trace of M fixes the fidelity with the emitted pair; its signed singular values fix the frequencies, the purity, the negativity and the largest CHSH value over all axes:
For turns about the ray one complex number carries M: γ = ⟨e2iδ⟩, with δ the difference of the two frames’ turns. M has the singular values (1, |γ|, |γ|) and the trace 1 + 2 Re γ. The frequencies are (1 ± |γ|)/2 on two equal-weight pairs, the fidelity with the emitted pair is (1 + Re γ)/2, the negativity is |γ|/2, and the CHSH value is 2√2 Re γ at the Tsirelson quartet aligned to the emitted pair, 2√2 |γ| over all clamps on the phase circle and 2√(1 + |γ|²) over all axes. For a Gaussian relative turn of variance ς², |γ| = e−2ς²: the pooled pair exceeds the local bound on the phase circle exactly when ς² < (ln 2)/4 = 0.1733 rad², a spread under 23.9°, and over all axes for every |γ| above zero. For independent increments of the relative turn, one per correlation time τc with characteristic function p̂, |γ(t)| = e−Γt with Γ = −ln |p̂(2)|/τc.
Decoherence is ignorance of the relative history
Every resolved history preserves the pair exactly: one history has one relative rotation, with singular values (1, 1, 1), whatever its gradients. The reductions appear only when the histories are discarded and the records pooled. Applying the measured inverse of each arm’s map before pooling restores the emitted table. Nothing physical is lost; the record of the history is. This is I-4’s reading of superposition as ignorance of the helical phase, extended to the pair: deterministic, and local to each arm.
The simulated ensembles
Stages v19 and v20 of the helix-projection suite draw differentiable three-dimensional fields, trace each wing through its own realised field over one metre, and read the records on the projection measure, with 1 536 paired histories per scenario. In all nine scenarios the settlement’s table equals the consensus table of the same maps in all sixteen cells, the largest difference 7.2 × 10⁻¹⁵.
| The two arms’ field histories | Negativity, |γ|/2 | CHSH | CHSH, uniform phase |
|---|---|---|---|
| Uniform fluctuating density, no gradient | 0.500000 | 2.828427 | 2.000000 |
| Identical fields and bases | 0.500000 | 2.828427 | 2.000000 |
| Independent fields, αn = 0.4 on both arms | 0.309728 | 1.750213 | 1.364052 |
| Coefficient correlation 0.5, same marginals | 0.332705 | 1.882047 | 1.447986 |
| Shared coefficients, αn = 0.2 and 0.8 | 0.256879 | 1.450123 | 1.127653 |
| Independent fields, αn = 0.2 on both arms | 0.453294 | 2.563461 | 1.887369 |
| Independent fields, αn = 0.8 on both arms | 0.115175 | 0.651283 | 0.529043 |
| Matched order of gradient regions | 0.500000 | 2.828427 | 2.000000 |
| One arm’s regions reordered | 0.488372 | 2.759170 | 1.961892 |
Each row is a statement of this section. A fluctuating density with no gradient rotates nothing. Identical ordered histories conserve the pair. Records pooled over independent histories are reduced. So are records pooled under a common drive on unequal baselines, and records pooled with the regions of one arm reordered, although every scalar summary of the two arms is equal.
The fluctuation of a medium is written on the odds of αn = 1 − n⁻²: with q the logarithm of the odds, ∇ln n = ½ αn ∇q. With the fluctuation fixed on the odds, the drive grows with the baseline, and the coherence under independent fluctuations falls with it: the negativity is 0.4925, 0.4533, 0.3829, 0.3097, 0.2368, 0.1843, 0.1432 and 0.1152 at nominal αn from 0.1 to 0.8. The pair is more fragile in a denser medium. Doubling the index at a fixed gradient changes no digit of that row. On the phase circle the pooled pair of this field law passes under the local bound between αn of 0.3 and 0.4; over all axes its ceiling stays above 2 at every point of the sweep.
The pooled pair. Turns about the ray: the relative turn of the two frames is Gaussian, with the spread the first slider sets; the second slider sets the turn arg γ and leaves it in. The solid curves are the largest CHSH value of the pooled pair over the phase circle and over all axes; the dashed curve is the value at the quartet aligned to the emitted pair, which the turn arg γ lowers and a turn of one clamp restores. The fidelity with the emitted pair follows the trace, (1 + Re γ)/2, while the frequencies, the negativity and both ceilings stay where |γ| puts them. The dotted vertical line is the spread at which the phase circle falls under the local bound. No preferred axis: the bars are the four frequencies of the pooled pair, the emitted pair and its three images, against the mean relative rotation; the strip marks where the negativity vanishes and where the local bound is passed. The sweep: the simulated ensembles of the helix-projection suite under independent fluctuations of an amplitude fixed on the odds, from the paper’s table; the slider picks a point.
I-4, I-6, P-4
The tests
Use 2 What would end each result
Each statement names its observable, its comparator and what falsifies it. E1 to E4 and E6 are shared with the consensus account for the same maps: they test the measure and the transport, and they are convergent readings, never counted as predictions of FDT alone. E5 carries FDT’s constitutive input.
- E1Kernel and ceiling. On every all-record ledger of the equal-weight pair the kernel is −cos Δ at unit visibility and |S| ≤ 2√2. With N records per pair of settings the counting standard error of the CHSH value at the Tsirelson quartet is √(2/N). An all-record CHSH value above 2√2 by more than the stated number of standard errors, or a unit-visibility kernel other than −cos Δ, falsifies the projection measure.
- E2The determinant. At three clamps per wing a sixth of a turn apart on a phase circle the determinant of the 3 × 3 table of correlators is 0 on the projection measure and −16/27 on a clamp-independent uniform phase, and 181 records per cell separate the two by five standard errors. A determinant away from zero by more than the stated number of standard errors falsifies the Gram form of the table. No inequality enters.
- E3The spectral identities. With each arm’s turn measured on classical light through the same medium, the mean relative rotation is known before any pair is read; for turns about the ray it is the one number γ. The identities then fix the negativity, the fidelity and the CHSH values with no free number. A pooled pair off any one of them falsifies the premise that the unresolved channel is the relative rotation of the two frames.
- E4Common mode and the gradient null. Identical maps on the two arms conserve the pair at any noise amplitude, and multiplying the index by a constant at a fixed gradient of its logarithm changes nothing. A reduction of the pooled coherence under either falsifies the gradient as the generator.
- E5The fragility sign. Under independent fluctuations of an amplitude fixed on the odds, the coherence falls as the baseline αn rises: negativity 0.453, 0.310 and 0.115 at αn of 0.2, 0.4 and 0.8 in the simulation. The arm whose medium carries the higher index loses agreement faster. The comparison is built between two arms that differ only by the hand that raises the index, with E4’s null as the comparator’s zero. A coherence that rises with the index under those conditions falsifies the load map on αn.
- E6Order. Reordering the gradient regions of one arm, with every scalar summary of the arm unchanged, lowers the pooled value: 2.7592 against 2.8284 in the simulated example. An arm whose records are blind to the order falsifies the noncommuting transport.
Run the checks
One script, nine sections, fixed seeds
The paper’s derived statements are recomputed by one script that uses none of the simulation’s code. It checks the kernel and the shares by quadrature; the sphere’s kernel on two million records per pair of axes; the uniqueness of the measure as a linear system on grids of phase points, with the family that signals beside it and the multipliers of the sphere’s map up to degree 201; the price, the ceiling and the two mutual informations; an event-by-event run with every record kept, and the same weight used as a filter; the rank and the two determinants, exactly and from sampled records; the kernel under transport; the pooled pair on four ensembles of 3 000 histories, one of them with a mean rotation of negative determinant; and the numbers quoted in the text.
It prints S = 2.8281 ± 0.0014 on the projection measure and 2.0024 ± 0.0017 on the uniform one, with all 4 000 000 records of each kept, the measure recovered to 2 × 10⁻⁷ on 1 440 phase points, and the two determinants as 0 and −0.592593. It needs NumPy and runs in a few minutes.
Readings declined
Use 6 Data yes, interpretation not necessarily
Standard reading
The correlation is carried by an action that crosses the space between the wings.
FDT reading
The correlation belongs to the configuration of one system: one shared phase and two clamp axes. The kernel −a·b is a property of that configuration and of nothing smaller, and each wing’s own table, taken alone, contains no trace of it (I-4, 0.1a).
The first record collapses a wavefunction, and the far wing’s state changes with it.
The clamp that hosts the settlement sets the measure of the shared phase, and the record at that clamp tells on which side of the cut the phase lies: projection and phase fixation (I-4). Reading the record at one wing changes nothing at the other.
The state of the pair is independent of the analysers that will read it.
The pair and its clamps are one system, and the measure of the shared phase is set by the host clamp (0.1a, P-2). With the lock and two local sign records, the table and the even shares leave no other measure for one host clamp to set, and the price is exact: 2√2 − 2.
Decoherence is the loss of the pair to its surroundings.
Every resolved history preserves the pair exactly. The reductions appear only when the histories are discarded and the records pooled, and the measured inverse of each arm’s map restores the emitted table.
A black hole that evaporates destroys the record of the wing lost behind its horizon.
A black hole loses no mass and grows only by merging: the matter accreting onto it is converted to spacetime and energy rather than falling into it (0.8). No wing’s record is drawn from behind a fold.
Convergence
Use 2 Same source, same data
Where FDT, quantum mechanics and general relativity converge, the paper recovers the record for the same source and data. The sign records of the helix chart on the projection measure of the host clamp return the consensus table of the equal-weight pair cell by cell: −cos Δ on the phase circle and −a·b on the sphere, even shares at each wing, 2√2 at the Tsirelson quartet with every record kept, and Tsirelson’s value as the ceiling. For the same maps the transported ensemble has the consensus table: the four frequencies from the mean relative rotation of the two frames, (1 ± |γ|)/2 for turns about the ray, with the consensus fidelity, purity, negativity and CHSH values. In the gravitational window the transport is general relativity’s. The laboratory record is accepted in full: Aspect and colleagues; Hensen and colleagues, 2.42 ± 0.20; Shalm and colleagues; Giustina and colleagues; Yin and colleagues, 2.37 ± 0.09 on coincident pairs.
FDT claims no divergence from the consensus table of the equal-weight pair. The one signed prediction carried here, the fragility of the pair in a denser medium, rests on a constitutive input: the load map on the odds of αn under the aperture assignment (I-6, P-4). FDT declines the consensus readings, action across the space between the wings, collapse, probability amplitudes and a hidden state independent of the clamps, and reads the same record as shared bivector grade with a deterministic phase, the pair’s record one settlement of the pair and its clamps (I-4, 0.1a).
Correlation is geometry in the exact meaning of a Gram matrix, and the pair’s record is that geometry read on the helix chart. The table of the equal-weight pair, its ceiling, its behaviour under transport and its fragility in a medium all follow from one measure, one lock and one generator. With the lock, two local sign records and one host clamp, the measure is the only one the table and its even shares admit.