Quantum mechanics as usually understood requires every history in Feynman's sum to cross each instant of time exactly once, running forward; the direction of time thus enters as a condition on the histories rather than as a consequence of anything observed. In this paper we examine what follows when the condition is withdrawn. A history may turn in time, with an amplitude per unit time fixed by a constant $\Delta$ characteristic of the field, and the forward direction of time is defined, operationally, as the direction in which an experiment's absorbers absorb. From these two postulates, for the electron with its spin in three spatial dimensions, one conservation law follows -the signed pairing between the prepared state and the absorbed state is independent of the intermediate time -and from it the two-time probability rule, weak values, and the rule of Born are derived rather than postulated. The turn lowers the electron's mass and hides behind it exactly: the theory at fixed $\Delta$ is a constant rotation of a two-valued label, carrying the free electron and the electron in any electromagnetic field onto a Dirac electron of mass $\sqrt{m^2-\Delta^2}$, so that no free measurement bounds $\Delta$ and every ordinary intensity is even in it. The turn is exposed only to an instrument that is not in its ground state: a population-inverted screen behind a double slit, dark in the ordinary theory, on which the strand of the electron running against time interferes with itself. The correlations between distant measurements are carried by local histories fixed at both ends, which is not action at a distance. Part~I states the theory and its consequences in physical terms; Part~II proves them.
Quantum mechanics as usually understood requires every history in Feynman's sum to cross each instant of time exactly once, running forward; the direction of time thus enters as a condition on the histories rather than as a consequence of anything observed. In this paper we examine what follows when the condition is withdrawn. A history may turn in time, with an amplitude per unit time fixed by a constant Δ characteristic of the field, and the forward direction of time is defined, operationally, as the direction in which an experiment's absorbers absorb. From these two postulates, for the electron with its spin in three spatial dimensions, one conservation law follows -the signed pairing between the prepared state and the absorbed state is independent of the intermediate time -and from it the two-time probability rule, weak values, and the rule of Born are derived rather than postulated. The turn lowers the electron's mass and hides behind it exactly: the theory at fixed Δ is a constant rotation of a two-valued label, carrying the free electron and the electron in any electromagnetic field onto a Dirac electron of mass m2−Δ2, so that no free measurement bounds Δ and every ordinary intensity is even in it. The turn is exposed only to an instrument that is not in its ground state: a population-inverted screen behind a double slit, dark in the ordinary theory, on which the strand of the electron running against time interferes with itself. The correlations between distant measurements are carried by local histories fixed at both ends, which is not action at a distance. PartI states the theory and its consequences in physical terms; PartII proves them.
Feynman's sum admits only histories that cross each instant once, forward in time. We withdraw this condition. A history may turn in time, with an amplitude per unit time fixed by a constant Δ, and the forward direction is defined by the absorbers of each experiment. One conservation law follows. From it the arrow of time, the square in Born's rule, anomalous weak values and the correlations of Einstein, Podolsky and Rosen cease to be paradoxes. The turn hides exactly behind the electron's mass and shows itself only to a reader of the backward strand.
Three results are new. First, the two-state-vector formalism of Aharonov and collaborators is \emph{derived} rather than postulated: for a Feynman sum whose histories may fold backward in time, with both boundary conditions imposed, the unique slice-invariant pairing between the forward- and backward-evolved boundary data is fixed by a conserved topological charge --- the signed count of slice crossings --- and the two-state description is the native form of the path integral. Second, at nonzero fold coupling Δ the two-state vector acquires dynamics: the fold vertex exchanges amplitude between the two states, which in every existing time-symmetric formulation evolve independently. Third, the measurement calculus follows in sequence: weak values, the Aharonov--Bergmann--Lebowitz rule, and, upon summing the future boundary, the Born rule, whose first-quantized deficit is exactly the flux of created pairs. Conservation of the signed count forces the fold vertex to be a hyperbolic SU(1,1) mixing, so the coupling enters the dispersion as E2=c2p2+ℏ2a2−Δ2, reducing mass and forcing second quantization, under which the fold is a two-mode squeezing with conjugate-amplitude ratio R=tanh(21artanhΔ/E). The coupling hides behind the mass it shifts and is not yet bounded by experiment; with absorber-defined boundary data every strand-blind observable is even in Δ, so first-order access requires strand-odd probes or a reader of the backward strand --- a gain medium for a bosonic field, a post-selection for a fermion. The arrow of time appears nowhere in the equations; it enters through the readiness of absorbers.
Sorkin defined a quantum measure as a set function on histories that is nonnegative, permits two-way interference, and satisfies the sum rule that forbids three-way interference. We show that the two-boundary measure of the folded-histories theory (the theory obtained from Feynman's sum by allowing a history to turn in time with an amplitude per unit time Delta) is a quantum measure in this sense for every value of Delta: it arises from a Hermitian, biadditive, strongly positive decoherence functional, and its measure of the whole history space exceeds unity by exactly the squared norm of the backward-in-time strand, which the second-quantized theory identifies as the flux of created pairs. The measure is Sorkin's double sum over histories with one added condition: histories interfere only if they end on the same strand, and only the forward strand is counted. The signed crossing count, which the fold conserves, also satisfies the sum rule but is not positive; Sorkin's positivity axiom excludes it; among the positive pairings that remain, the absorber selects the one restricted to the strand it reads. In unitary quantum measure theory a serial event, one defined by a sequence of projectors, cannot exceed unity; in the folded theory the whole history space, a serial event specified by identity projectors, does whenever the backward strand is populated. A serial event whose measure exceeds unity is therefore the signature of evolution that is not unitary; for the free electron, the only such mechanism in the theory is the turn. The fold introduces two-way interference between strand-resolved classes, while preserving the absence of three-way interference at every real Delta. We prove the measure identities, outline an optical test, and report a finite-system numerical audit.
Quantum mechanics is taught from five postulates: a state is a vector in a complex space; left alone it turns by a unitary step that an energy operator generates; a measurement returns an outcome with a probability equal to a squared length; the state afterwards is the part that matches the outcome; and two systems together live in the product of their spaces. The textbook does not say where the five come from. We read each of them as a statement about one object, the recognition ledger, a record of events that is written in postings and never erased, and we say for each how much of it is then a theorem. The guiding sentence is short: histories that leave no record add, and what is on the record is counted. The record's clock makes a complex unit. The tick of the eight-position register has no complete set of real stationary states, and on the states that four ticks reverse, two ticks square to minus one. For a law on a finite-dimensional space whose factors all lie above zero, the unitary step is fixed up to its unit: a linear rule for the step that uses only the law between two records, keeps every length, regroups and is continuous is exp(−icH) with H = −log T, where T is the law itself. The square follows, for reads of a register's positions, from two principal properties and three mild ones: a read that weighs each position by its own value and adds over positions, is silent on an empty position and is never less for a larger value, and whose total is blind to a common turn of the register, is the sum of squares. For reads along other splittings of a space the square is a hypothesis. Two acts can be read together exactly when they commute, and reading one first moves the other's weights by an amount we give in closed form, never more than one half. Where a crossing takes held states off the held part, a contact divides a held state into a part left exactly as it was and a part with one crossing applied, with squared lengths cos²θ and sin²θ; where it keeps them held, the contact reads one sign of the register and nothing else. Two blocks of one cube compose by the product, their acts commute, and their correlations have three ceilings, 2, 2√2 and 4, for acts that only read, acts that write, and tables that merely do not signal; the cube reaches 2√2 exactly. Each theorem is proved on the page from premises we name, and we say postulate by postulate what that gives. The second follows from four statements about a linear step. The third follows for reads of positions, given that a read is taken position by position, which is owed. For the first, the clock makes a complex unit with no premise, the one such unit that is a whole number of ticks; that states are complex rests on a premise about the clock that is close to the conclusion, that a held register is among the states the unit acts on is a working statement, and from the second postulate on the spaces are taken complex, as in the textbook. Of the fourth the algebra is proved; the weights and the state afterwards are the textbook's, and that one outcome occurs is owed. The fifth is proved for two blocks of one cube and owed for two laboratories, where published network experiments exclude a real product whose sources share nothing. Several of the theorems are known results of the textbook's own mathematics, and the ledger enters through what their objects are identified with. What is owed is named in four statements, and we say how exact each is. The paper closes with twelve paradoxes of quantum mechanics, from the double slit to Wigner's friend, each in three lines: the puzzle, what the record says, and what is owed. Eleven of the twelve wait on the same owed statement, and every number they return is the textbook's own.
Quantum mechanics is taught from five postulates: a state is a vector in a complex space; left alone it turns by a unitary step that an energy operator generates; a measurement returns an outcome with a probability equal to a squared length; the state afterwards is the part that matches the outcome; and two systems together live in the product of their spaces. The textbook does not say where the five come from. We read each of them as a statement about one object, the recognition ledger, a record of events that is written in postings and never erased, and we say for each how much of it is then a theorem. The guiding sentence is short: histories that leave no record add, and what is on the record is counted. The record's clock makes a complex unit. The tick of the eight-position register has no complete set of real stationary states, and on the states that four ticks reverse, two ticks square to minus one. For a law on a finite-dimensional space whose factors all lie above zero, the unitary step is fixed up to its unit: a linear rule for the step that uses only the law between two records, keeps every length, regroups and is continuous is exp(−icH) with H = −log T, where T is the law itself. The square follows, for reads of a register's positions, from two principal properties and three mild ones: a read that weighs each position by its own value and adds over positions, is silent on an empty position and is never less for a larger value, and whose total is blind to a common turn of the register, is the sum of squares. For reads along other splittings of a space the square is a hypothesis. Two acts can be read together exactly when they commute, and reading one first moves the other's weights by an amount we give in closed form, never more than one half. Where a crossing takes held states off the held part, a contact divides a held state into a part left exactly as it was and a part with one crossing applied, with squared lengths cos²θ and sin²θ; where it keeps them held, the contact reads one sign of the register and nothing else. Two blocks of one cube compose by the product, their acts commute, and their correlations have three ceilings, 2, 2√2 and 4, for acts that only read, acts that write, and tables that merely do not signal; the cube reaches 2√2 exactly. Each theorem is proved on the page from premises we name, and we say postulate by postulate what that gives. The second follows from four statements about a linear step. The third follows for reads of positions, given that a read is taken position by position, which is owed. For the first, the clock makes a complex unit with no premise, the one such unit that is a whole number of ticks; that states are complex rests on a premise about the clock that is close to the conclusion, that a held register is among the states the unit acts on is a working statement, and from the second postulate on the spaces are taken complex, as in the textbook. Of the fourth the algebra is proved; the weights and the state afterwards are the textbook's, and that one outcome occurs is owed. The fifth is proved for two blocks of one cube and owed for two laboratories, where published network experiments exclude a real product whose sources share nothing. Several of the theorems are known results of the textbook's own mathematics, and the ledger enters through what their objects are identified with. What is owed is named in four statements, and we say how exact each is. The paper closes with twelve paradoxes of quantum mechanics, from the double slit to Wigner's friend, each in three lines: the puzzle, what the record says, and what is owed. Eleven of the twelve wait on the same owed statement, and every number they return is the textbook's own.
Textbook quantum mechanics postulates its state space, its observables, a commutation relation, a Hamiltonian for each problem, the Born rule and a projection on measurement, and introduces spin as a separate two-component structure. Complex Mechanics (ℂM) keeps none of these as a law. It starts from one container—real time and three complex dimensions carrying the Kähler triad (g, J, ω) as one object—and a base of eight items stated once: the container with its triad; null matter; matter as sections of the square-root bundle K^(1/2) that superpose; two demands on the field, that its conservation law be that of the count (the fraction of runs that land in a region) and that its modulus stand in one relation to the count's density for every state, its phase each particle's action; and three named posits—a particle's two alignments, its moment along an axis or against it; where the runs start; and the frame in which a pair's joint outcome is resolved. Beyond the base the paper takes in what its Table 1 lists: two named constructions for pairs; one named premise, that the magnetic moment reads the angular momentum of the J-rotation, the J-action; three readings, one that an apparatus reads the visible sector only and two in the bound electron's moment; and results, statements and inputs of the framework, two of them numbers solved once on a measurement. On that footing the imaginary unit is the complex structure J, and a field's phase is the angle of its rotation, the J-angle. A superposition of null phases, the phases of free null matter, obeys i∂ψ/∂t = √(m² − ∇²) ψ, whose leading term, with the phase's rate at rest removed, is the free Schrödinger equation; a pull enters as an input. The Born rule's weight is derived: |ψ|² is the only density the paths carry. Where the runs start stays a posit, an initial condition; in the one case computed, a particle in a box, the paths carry a start off the density onto it at each cell size read, never point by point. Spin-½'s half rate is that of K^(1/2), whose sections change sign under one full turn of the J-angle; with a named construction for pairs it gives the exchange sign −1, and with the moment law's premise and the lepton's unit of J-action, g = 2. A particle's two spin states are its two alignments, and its Stern–Gerlach odds are cos²(θ_field/2) at the angle θ_field between its moment and the field. The theory is deterministic, with hidden variables: each particle's velocity is, at the leading order, the gradient of its own J-angle over m, the velocity law of de Broglie's and Bohm's theory, here one step from the base's reading of the J-angle as each particle's action. Where quantum mechanics gives the odds, the path decides which particle goes where: the particles that start in the top cos²(θ_field/2) of a Stern–Gerlach beam land in the + lobe and the rest in the − lobe, and behind a double slit the paths bend with the state and land as the fringes. The opposite-alignment pair's correlation is −cos Δ at the angle Δ between the two analyzers, every pair state's CHSH value is at most 2√2, and nothing signals. At the order computed no prediction differs from quantum mechanics, and the one place a difference could live is where the runs start. Told with these: the moment beyond g = 2 is not yet computed, and the bound electron's leading-order moment stands +2.0 of the measurement's uncertainty from hydrogen's and about 400 and 1060 of theirs from two hydrogenlike ions'. Light as a quantized field and systems of many particles, from helium to condensates, are not yet derived. A companion to the framework paper Complex Mechanics, version 4.30 (doi:10.5281/zenodo.19349816). The script qm_repro.py, posted with this record, recomputes 110 of the paper's printed values from its printed equations.