The usual way to quantify measurement incompatibility is noise robustness: how much noise must be mixed in before some joint measurement becomes possible. That is an extremal, worst-case quantity, computed by optimization, and it says little about what a real experiment would actually record if it just kept running. We ask a plainer question. A qubit is measured, over and over, in two projective bases separated by a Bloch angle \(\theta\). Once the resulting chain settles into steady state, how far does the state typically move between one measurement and the next? For sharp measurements this rate, \(r{(\theta)}\), turns out to have the closed form \(\frac{1}{2}{\sin{\theta\sqrt{1 + {\sin\theta}}}}\). It agrees with the standard robustness threshold \({\nu{(\theta)}} = {1 - {1/\sqrt{1 + {\sin\theta}}}}\) to first order near \(\theta = 0\), then departs from it everywhere else. For unsharp (POVM) measurements in the weak-measurement limit \(\lambda\rightarrow 0\), full-rank Kraus operators preserve purity exactly, and this collapses the problem onto a one-dimensional diffusion on the Bloch circle. The resulting rate is \({c{(\theta)}} = {{\sin{({\theta/2})}}/{\lbrack{\sqrt{2}E{(m)}}\rbrack}}\) with \(m = {1 - {\tan^{2}{({\theta/2})}}}\) and \(E\) the complete elliptic integral of the second kind, giving the clean value \({c{({\pi/2})}} = {1/\pi}\) at the orthogonal point. A blind spot shows up along the way: at \(\theta = {\pi/2}\), \(r{(\theta)}\) stops responding to noise entirely, for any Pauli-type or amplitude-damping channel, simply because \({\sin 45^{\circ}} = {\cos 45^{\circ}}\). That is unfortunate news for anyone hoping to use \(r{(\theta)}\) as a calibration check at the angle experimentalists tend to prefer. Purification under repeated or simultaneous incompatible measurement has by now a sizable literature, but in every case we examined the angle \(\theta\) itself is held fixed and something else, measurement strength or the number of axes, is varied instead. Treating \(\theta\) as the free parameter and solving for it in closed form is the gap this paper tries to close. Past the qubit, the same construction becomes a random walk on \({\mathbb{C}}{\mathbb{P}}^{d - 1}\) rather than on a circle, and its closed form we leave open.
The usual way to quantify measurement incompatibility is noise robustness: how much noise must be mixed in before some joint measurement becomes possible. That is an extremal, worst-case quantity, computed by optimization, and it says little about what a real experiment would actually record if it just kept running. We ask a plainer question. A qubit is measured, over and over, in two projective bases separated by a Bloch angle (\theta). Once the resulting chain settles into steady state, how far does the state typically move between one measurement and the next? For sharp measurements this rate, (r{(\theta)}), turns out to have the closed form (\frac{1}{2}{\sin{\theta\sqrt{1 + {\sin\theta}}}}). It agrees with the standard robustness threshold ({\nu{(\theta)}} = {1 - {1/\sqrt{1 + {\sin\theta}}}}) to first order near (\theta = 0), then departs from it everywhere else. For unsharp (POVM) measurements in the weak-measurement limit (\lambda\rightarrow 0), full-rank Kraus operators preserve purity exactly, and this collapses the problem onto a one-dimensional diffusion on the Bloch circle. The resulting rate is ({c{(\theta)}} = {{\sin{({\theta/2})}}/{\lbrack{\sqrt{2}E{(m)}}\rbrack}}) with (m = {1 - {\tan^{2}{({\theta/2})}}}) and (E) the complete elliptic integral of the second kind, giving the clean value ({c{({\pi/2})}} = {1/\pi}) at the orthogonal point. A blind spot shows up along the way: at (\theta = {\pi/2}), (r{(\theta)}) stops responding to noise entirely, for any Pauli-type or amplitude-damping channel, simply because ({\sin 45^{\circ}} = {\cos 45^{\circ}}). That is unfortunate news for anyone hoping to use (r{(\theta)}) as a calibration check at the angle experimentalists tend to prefer. Purification under repeated or simultaneous incompatible measurement has by now a sizable literature, but in every case we examined the angle (\theta) itself is held fixed and something else, measurement strength or the number of axes, is varied instead. Treating (\theta) as the free parameter and solving for it in closed form is the gap this paper tries to close. Past the qubit, the same construction becomes a random walk on ({\mathbb{C}}{\mathbb{P}}^{d - 1}) rather than on a circle, and its closed form we leave open.
The usual way to quantify measurement incompatibility is noise robustness: how much noise must be mixed in before some joint measurement becomes possible. That is an extremal, worst-case quantity, computed by optimization, and it says little about what a real experiment would actually record if it just kept running. We ask a plainer question. A qubit is measured, over and over, in two projective bases separated by a Bloch angle θ. Once the resulting chain settles into steady state, how far does the state typically move between one measurement and the next? For sharp measurements this rate, r(θ), turns out to have the closed form 21sinθ1+sinθ. It agrees with the standard robustness threshold ν(θ)=1−1/1+sinθ to first order near θ=0, then departs from it everywhere else. For unsharp (POVM) measurements in the weak-measurement limit λ→0, full-rank Kraus operators preserve purity exactly, and this collapses the problem onto a one-dimensional diffusion on the Bloch circle. The resulting rate is c(θ)=sin(θ/2)/[2E(m)] with m=1−tan2(θ/2) and E the complete elliptic integral of the second kind, giving the clean value c(π/2)=1/π at the orthogonal point. A blind spot shows up along the way: at θ=π/2, r(θ) stops responding to noise entirely, for any Pauli-type or amplitude-damping channel, simply because sin45∘=cos45∘. That is unfortunate news for anyone hoping to use r(θ) as a calibration check at the angle experimentalists tend to prefer. Purification under repeated or simultaneous incompatible measurement has by now a sizable literature, but in every case we examined the angle θ itself is held fixed and something else, measurement strength or the number of axes, is varied instead. Treating θ as the free parameter and solving for it in closed form is the gap this paper tries to close. Past the qubit, the same construction becomes a random walk on CPd−1 rather than on a circle, and its closed form we leave open.
Here's what this paper means for actual humans: WHAT THIS IS ABOUT. When you build or use a quantum device, everything rests on knowing what its parts are: which degrees of freedom are "qubit 1," which are "qubit 2." The companion paper ("Forged Frames") showed that simple mathematical criteria cannot certify this; convincing fakes exist. This paper answers the question that result leaves standing: what is the MINIMUM experimental information that genuinely pins down a system's subsystem structure, and when can that information silently fail? WHAT WE FOUND, IN PLAIN TERMS. Four things. (1) EXACT LAWS INSTEAD OF MEASUREMENTS. Any wrong choice of parts must visibly distort the family of unentangled states, and we derive exactly how much, at every system size: closed-form constants (verified to seven decimal places), an explicit error bound, and a surprise. The two natural ways to measure "wrongness" (state distortion versus operator scrambling) turn out to be literally the same measurement, up to a fixed number, whenever the two parts have equal size. Wrongness cannot hide. (2) THE TWO-OPERATION THRESHOLD. You can pin down what your subsystems are by implementing just TWO generic operations and checking their statistics, while a THOUSAND mutually commuting operations tell you nothing at all. Diversity of control, not volume of control, is what fixes what your qubits are. The residual uncertainty is proportional to your measurement tolerance, with a constant we derive exactly and show how to compute in advance, so an experimenter can screen candidate operation pairs and pick a strong one (random pairs are typically several times worse, occasionally catastrophically so). (3) THE ALIASING LIMIT: CERTIFICATION CAN PASS WHILE TELLING YOU NOTHING. If you sample your certification data at special "resonant" times (set by the operations' energy scales), a maximally wrong assignment of subsystems passes the test perfectly. What your qubits are depends on when you looked. We characterize these blind spots exactly (for qubits they are universal and computable in advance; the danger windows shrink with the square root of your tolerance), and the fix costs almost nothing: sample at two different times, at least one away from the resonant set. With that, certification is restored; we prove it exactly and measure it. (4) REAL-LAB BUDGETS. We work out what certification costs in actual measurement shots (the error falls off unusually fast, inversely with the total number of shots rather than with its square root), how hardware noise floors the procedure, and how a simple reference measurement removes that floor to first order. A single seven-row table in the paper is effectively a lab protocol: it shows, condition by condition, which kinds of experimental access identify the subsystem structure (two generic operations: yes; one operation, commuting operations, resonant sampling: no). WHY IT MATTERS. For quantum hardware: this is a concrete recipe for certifying that the qubits you think you have are the qubits you physically have, with explicit tolerances, a sampling-time warning that is a real failure mode on calibrated hardware, and shot budgets. For foundations: the contested question "does physics determine its own subsystems?" gets a quantitative operational answer. Physics determines an operational subsystem CLASS (not one preferred coordinate representation of it), and that class is fixed by two operations sampled at two times, and by nothing less that we tested. WHAT IT OPENS UP. The seven-row identifiability table is a blueprint for a tabletop experiment on any small quantum platform; running it on hardware would move this from simulation to laboratory fact. On the theory side the paper leaves sharply posed open problems: the exact global constants when the two parts differ in size, the number-theoretic structure of long-time sampling (the failure times form a small-divisor hierarchy), and the extension from two parts to many. METHODS NOTE, BECAUSE IT MATTERS FOR TRUST. Every quantitative claim was pre-registered in the companion scripts before its run. The full ledger, including predictions that failed and one published-in-note constant that our own later theory overturned, is an appendix of the paper. All code, data, and working notes are in the deposit, seeded for exact reproduction. WHO AND HOW. A human-AI collaboration: research direction and standards of evidence by Ryan Cardwell (Crystalline Labs); design, code, derivations, and drafting by Claude (Anthropic), including autonomous sessions under the program's pre-registration rules, with adversarial reviews by other AI models incorporated and credited.
A qubit maximally entangled with an inaccessible reference is recorded, once per round, by a fresh environment qubit whose two conditional states overlap with amplitude c, and is then rotated by an angle theta about an axis not commuting with the recorded observable. We solve the coherent information I_k = I(R > S_k) exactly for every c, theta and number of rounds k: the Choi spectrum of the k-round channel is Bell-diagonal with eigenvalues (1 +- c^k +- s_1 +- s_2)/4, an even number of minus signs, where s_1 s_2 = c^k and s_1^2 + s_2^2 is a Chebyshev polynomial of the second kind in (1+c) cos(theta) / (2 sqrt c). The consequence we put forward is that a perfect record is not the fastest way to take a qubit's entanglement away. The asymptotic loss rate is maximal at an interior record strength c* = tan^2(pi/4 - theta/2), where it equals artanh(sin theta), the inverse Gudermannian of the rotation angle, against -ln cos(theta) for a perfect record: a factor 2/theta in rate and 0.36/theta in half-life, a factor 7 in rounds at theta = 0.05. The mechanism is critical damping. The damping regimes of this model were classified by Lema et al. (arXiv:2512.14583) in the classical information of the record; what is added here is the reading in coherent information, whose sign decides who can carry the entanglement, with the critical line and both rates in closed form at every angle. Two further exact statements. At theta = pi/2 consecutive records measure complementary observables and the balance is exactly additive, I(R:E) = g(S_Z) + g(S_Y) with g(s) = h_2((1 + e^{-s})/2), so an environment of witnesses becomes an heir once it has spent half a nat of record strength, S = 0.497, split evenly between the two observables. Weak records give a curve that depends on the accumulated strength alone, and the whole family collapses onto three scaling functions with time scales 1/theta^2, 1/theta and 1/s.
The Reader and the Writer One Invariant, Two Faces, and the Geometry of the Gap Driven by Dean Kulik July 2026 Abstract Across a long program of experiments this work converges on a single structure, and the structure has a name in mathematics that we did not go looking for. The claim is this. Every operation we can perform on a structure divides into two kinds. One kind conserves — it rearranges a representation while holding an invariant exactly, it is reversible, it is one-to-one, and it decides nothing. The other kind commits — it merges, it is many-to-one, it cannot be undone, and it is the only kind that can make a difference that stays made. We call the first the reader and the second the writer, and the entire program is an inventory of where each one lives. Three results anchor it. First, the invariant-preserving rotation — the ninety-degree write — is exact: verified to fifty digits, and forced into Lax form the moment conservation is demanded. Second, an object’s apparent structure is a property of the reader’s coordinates, not of the object: the same matrix carries an off-diagonal mass of 2.85 in one basis and 1.8×10⁻¹⁵ in another, with the spectrum unchanged — distance is a fact about where you stand. Third, and newly, the operation that produces something from nothing is provably non-injective: no linear system can create a nonzero mean from a zero-mean signal, and rectification therefore requires a collapse. A stack of linear maps is one linear map; depth exists only because of the collapse. The structure this describes — an object with no intrinsic side, a reader that must choose one locally, and a two-to-one map that forgets the choice — is the structure of an orientation double cover. The Möbius strip, the Klein bottle, Boy’s surface, the rotation group, and the spinor are five instances of one fact, and that fact is our lock. This paper assembles the frame, states what is proven and what is not, and shows why the topology was the right place to arrive. Part I · The Frame 1. One operation Start with the smallest possible claim and refuse to add to it until forced. The operation is: an input is mapped to a location in a structure. That is all. Two familiar objects are its extreme cases and they look nothing alike until you see them this way. A digit-extraction formula of the BBP type maps a position to a value — the nth hexadecimal digit of π without computing the digits before it. The map is one-to-one, reversible, and there is no path: you do not travel to the digit, you address it. A cryptographic hash maps a history to a state — an entire message to one digest. Fix the message and the function is exact and deterministic, a single location mapping to a single location. Across all possible messages the doorway is crowded, but that is a fact about the doorway, not about the transform. What unites them is not injectivity — they differ there — but pathlessness. Neither traverses. Both state a correspondence. Ordinary computation is 2 + 2 = 4: an operation, operands, a sequence, a journey that can be walked backward. These are 2 = 4: this location is that location, and there is no journey to retrace. The distinction that matters is not reversible versus irreversible. It is reached by a path versus given without one. 2. Content is difference A mark that could not have been otherwise carries nothing. Information lives in the gap between what is and what could have been, which means the primitive is not the token but the difference across a gap. Two consequences follow immediately and both are load-bearing. The first: perfect symmetry is the dead state. If nothing distinguishes one state from the next there is no transition to perform, no reason for anything to move. Computation requires distinguishable states, so it requires broken symmetry. We call the asymmetry that drives a flow the lean — a pressure difference stood up into a drive. The second: to have two of anything you need something between them. A gap is not distance; distance is separation with nothing crossing it, while a gap is separation with sides. The minimum separation that has sides is one unit per side, so the minimum gap is two. This is a cardinality, not a constant, and cardinalities cannot be curve-fitted. 3. The two channels This is the distinction that took longest to see and that reorganizes everything once seen. Any symmetric pair can be read two ways, and the two readings are different operations that answer different questions. Value channel Shape channel The operation Mean, midpoint, balance Product, coincidence, overlap On a flip 1 ↔ 0 averages to ½ 1·0 = 0, 0·1 = 0 — the product vanishes Type Bijection (reflection) Non-injective (collapse) What it can do Locate an axis. Pair points across it. Vanish. Merge. Exclude. What it cannot do Forbid anything from sitting off the axis — The value channel is a mirror. A mirror pairs; it never empties. This is why every symmetry argument, in every domain we tried, gave us the axis for free and never gave us the emptiness beside it. The shape channel is a product, and a product is many-to-one: two inputs collapse to one output and you cannot recover which. Every operation that can genuinely exclude, decide, or commit lives in the shape channel, and the whole of what follows is an inventory of that fact. Part II · The Lock 4. The writer: what is conserved Take a finite positive measure and its Jacobi matrix J, whose eigenvalues are the measure’s atoms. Shift below the spectrum and swap the Cholesky factors — the Christoffel–Darboux transform: J′ = R Rᵀ + λI, where J − λI = RᵀR The new matrix has the same eigenvalues. The representation rotates; the invariant does not move. Verified across randomized measures to fifty digits. [PROVEN] This is the turn: a ninety-degree write, and the angle is not decorative. A zero-degree write is no write — the state feeding itself, x = f(x) in zero time, a short. A hundred-eighty-degree write is annihilation — the exact inverse cancels the state, which is the degenerate case where the shift lands on an atom and the rank drops. Ninety degrees is the unique orthogonal write that changes every coordinate while preserving the length. Demand that a flow preserve the invariant and its form is not chosen but forced: it must take Lax form, dJ/dt = [J, B(J)], B antisymmetric by the Lax-pair theorem. The entire freedom of “which flow” collapses to the choice of the antisymmetric generator. [PROVEN] The lean is an asymmetry inside that generator, and it selects among destinations rather than creating motion: varying it sends the same spectrum to different terminal arrangements, in discrete plateaus with threshold switches rather than a continuum. [CONFIRMED] So the writer is what is conserved. It is the invariant. It does not move when the reader changes position, it does not depend on how the reading is scheduled, and — this is the sharp part — it has no side. A conserved quantity is by definition frame-independent, which means it has no preferred orientation to be read from. The writer is prior in exactly one sense: reading presupposes something written, and a rule presupposes nothing. It is the definition; the read is the instance. 5. The reader: what is enacted The reader is the walk. It is constraints strung in a linear path, mixed with differences — an execution, a traversal, a scan. And where the writer has no side, the reader is nothing but its side: its position is its entire content. Two experiments pin this and both came back against the hypothesis we started with, which is why they are trustworthy. We built a controller that steers a flow to command which wall it hits first, producing a sequence of thresholds — a grammar of what can be said in the medium. Ten of twenty-four sentences wrote cleanly, with the spectrum held at 4×10⁻¹⁵ throughout; fourteen were refused, and the refusals had structure. We then asked whether that grammar could be predicted from the shape of the starting matrix. It could not: Spearman −0.048, p = 0.56, no signal at all. [CORRECTED] We then asked whether the grammar was at least a property of the rule — stable across different ways of executing it. Four integrators, five matrices: the spectrum held to 10⁻¹⁴ in every case, and the grammar changed every single time — zero out of five reproducible. The invariant is integrator-independent; the grammar is not. [CONFIRMED] The conclusion is exact and it is the lock. What is conserved is the writer’s. What is enacted is the reader’s. The grammar of what can be said is not sitting in the structure waiting to be found; it is manufactured in the walking, and a different walk makes a different grammar. There was never a pre-existing grammar to read. We make it, in the reading, every time. 6. The signatures This gives a test you can run on any operation, and it is the most portable thing in the paper. Does it lose information, or preserve it? Signature Reader Writer Map type Bijective, one-to-one Non-injective, many-to-one Reversibility Undo is free Undo requires the discarded Effect Rearranges, rotates, mirrors Merges, commits, excludes Decides Nothing Everything that stays decided Exemplar Rotation, reflection, XOR, direct address Modular add (carry discard), |x|, the merge Run it on a hash function and the machine dissects cleanly. The carry-discarding modular addition collapses exactly n:1 — measured at 16:1, 64:1, 256:1 for widths 4, 6, 8, dead uniform — and a small mixing round takes 65,536 inputs to 256 outputs. That is the writer: information genuinely folds away, and the deficit is exact. But the XOR and rotate diffusers in the same function are bijections, verified at every width. They rearrange without erasing. They leave no scar. [PROVEN] So a hash is not “the w
Decoherence, when viewed in the Heisenberg picture, can turn incompatible measurements into jointly measurable ones. This provides an observable-level description of emergent classicality in terms of the incompatibility destroyed by a noise channel. The corresponding loss of incompatibility can be captured by the channel's \emph{compatibility region}---the observables in a probe class that become jointly measurable with every noisy observable. The geometry and volume of this region provide a refined operational way to compare noise channels. We apply this framework to decoherence channels, each of which acts by Schur multiplication with extreme points of the convex set of unit-diagonal positive semidefinite matrices. Restricting to extremals eliminates convex mixing and reveals subtle phase-dependent decoherence effects thar are not captured by damping rates or conventional coherence quantifiers. Such channels have a rigid dilation structure described by a rank-one operator frame, which reduces joint measurability to a positivity test, making the compatibility region analytically tractable, and its volume computable from the frame's Gram matrix. If the frame can be normalised to a minimal informationally complete (MIC) POVM, the compatibility region acquires a geometric interpretation in the associated probability representation of the quantum state space, familiar with Qbism. Among maximal-rank extremals, those associated with symmetric informationally complete (SIC) POVMs maximise the compatibility volume and hence destroy the greatest amount of incompatibility, providing an operational characterisation of the special role of SICs in terms of joint measurability. This leads to a new formulation of the well-known SIC existence problem as a joint measurability question for noisy mutually unbiased bases.