How a timing pattern in public catalogs became First Principles V9.6.1 — what was withdrawn, what became theorem, and the prediction the theory now stakes itself on.
Z. Paz — EXISTS | HAPPENS — 2 September 2026
The STF programme did not begin as a theory. It began as a pattern in public data. A statistical programme over public catalogs of three messengers — the Pierre Auger Observatory’s ultra-high-energy cosmic rays from 2004 to 2018, cut at 20 EeV and zenith below 80°, 494 unique events; gamma-ray bursts; and gravitational-wave events — returned a temporal structure that had no business being there. In the triple coincidences the cosmic rays arrived before the gamma-ray bursts of the mergers they were paired with: 3.32 ± 0.89 years before, on average, with a separate 71-day window in the neutral channel; and a blind maximum-likelihood search over a family of exponents independently returned n = 11/8 with a mean period of 3.31 years. The observational manuscript, hosted on the companion site uhecrtoday.com, reports the pre-merger correlation at more than five sigma, and this site states no larger number for it.
Then came the part that turned a statistical anomaly into a physics question. Take the three timescales — 54 years, 3.32 years, and 71 days, the first obtained by closing the emission window in a closure the current paper grades as conditional, the other two measured — and run them through the Peters inspiral law for a compact binary. They land at 1466, 730, and 360 Schwarzschild radii of separation: successive near-halvings on one trajectory. General relativity relates the three numbers while marking none of them as special; if the pattern was real, something beyond GR had to be selecting them. And a period of 3.32 years, read as a field oscillation, corresponds to a scalar mass of 3.94 × 10⁻²³ eV. The first STF Lagrangian was written to express what those data contained: a scalar coupled to the rate of change of spacetime curvature, activating in the last years before a merger.
Those are the unexpected observations of the title — two measured, one closed from them. Everything since has been an attempt to reconstruct, from General Relativity, topology, compactification and causal response, the theory the data seemed to require — and to find out how much of the reconstruction could be made to stand. Two editions bracket that attempt.
The reconstruction’s last single-clock edition, and the one that claimed completeness, was V7.9. In March 2026, The STF from First Principles V7.9 ran to nineteen appendices and, printed, nearly three hundred pages. It described a derivation chain: from General Relativity and ghost-freedom constraints at the bottom, through a scalar field coupled to the rate of change of spacetime curvature, up through a ten-dimensional compactification to Calabi–Yau geometry and the Jarlskog invariant at the top. It reported a flyby-anomaly coefficient matching Anderson’s measured value to 99.99 percent, a chirp mass predicted from the fine-structure constant, and most Standard Model constants to within about a percent. It was confident, and it was long.
The current edition, V9.6.1, is twenty-nine pages. It claims less and proves more. Between the two lie a dozen intermediate releases, two frozen consolidations, a five-gate adversarial audit, a fifty-one-gate architecture programme, eleven comparative gravity studies, two retrospective data studies, and a machine-verified record of more than thirty-three million checker assertions. Several of V7.9’s headline results are gone — not deleted, withdrawn, with the withdrawal and the superseded statement both preserved. In their place stands something V7.9 did not have: a theorem where there had been an assumption, an audited gravitational architecture where there had been a claimed derivation, and a frozen, preregistered prediction that the theory can lose.
This is the delta, told in order.
The most important change between V7.9 and V9.6.1 is not a result. It is a rule. Every claim in the current edition carries a status — theorem, derived, conditional, retrospective, open, withdrawn — and a quantity counts as a prediction only if the calculation producing it never used the observation against which it is tested. If it did, it is a calibration. If an independent calculation reproduces an observation it did not use, that is a validation. And where a past claim was corrected, the correction and the superseded statement are both kept: in this programme, withdrawal is a status, never a deletion.
That rule is enforced mechanically. Every release from V8.2 onward ships with a deterministic checker that re-derives its quantitative claims; the recent releases assert two counters in every run — zero fits to observation, zero timing values used to solve a theoretical coefficient — and, from V9.6, a clean-room verifier that re-checks every hash and regenerates every result byte for byte. The chain from the frozen V8-series baseline to the current state is hash-pinned end to end and was independently re-verified on review, with the V9.5 and V9.6 archives rebuilt byte-for-byte. What follows is what that discipline did to the theory.
V7.9’s most quoted number was the flyby coefficient K = 2ωR/c, whose factor of two, sign, and amplification V7.9 derived from the Lagrangian, and which matched Anderson’s empirical value to 99.99 percent. Four no-go results in V8.1 — three against the mechanical reading, a fourth against the repaired rate interaction on a stationary Earth — showed the derivation could not stand. The interaction, pulled back to a spacecraft’s worldline, is a connection one-form whose curvature is antisymmetric: it does no work. It cannot transfer planetary rotational energy to anything. The force reading was withdrawn, and everything calibrated from it went with it — the flyby calibration of the coupling constant and every claim downstream of that calibration.
What survived is more interesting than what was lost. An interaction that does no work can still have a holonomy around a closed loop, and a coherent radio transaction can register one. The factor of two in Anderson’s formula turns out to be the vorticity identity ∇ × (ω × r) = 2ω of the Earth-fixed clock congruence relative to universal time; the equatorial radius and declination-only dependence are the operator norm of the rotational clock channel over the closed carrier; the absence of G and M is the signature of an observer-clock map rather than a curvature-sourced force. The flyby anomaly became a measurement of clocks, not a push on spacecraft. The later planetary nulls, which falsify the force reading, favor the observer-clock reading — Earth is source and clock carrier at once — with a per-configuration utilization coefficient that is still open, and the paper says so.
Also withdrawn: the DHOST Class Ia classification and the Horndeski route to ghost-freedom (ghost-freedom is established on exterior-vacuum and Kerr backgrounds and open in general); the chirp-mass relation M_c = 18.54 M☉ and its “99.9 percent LIGO” comparator (the relation was dimensionally inconsistent, and the comparator was not a catalog statistic); the claim that the threshold’s numerical value had been converted to SI rather than matched at the reference system; and the description of the field mass as reached by “independent paths,” which is now reported as what it is — a three-path convergence.
Every one of these is still readable, in full, on the archived V7.9 page under its notice. So is the part of V7.9 that was not withdrawn: the downstream derivations — Standard Model constants, the flavour manifold, CP violation from the phase-lag mechanism, the Weil–Petersson curvature — remain V7.9’s derivation record, delegated to it by the current edition and published as separate records with their own provenance and status notes.
The framework’s Theory of Time had always held two temporal notions apart: a universal ordering that every system references, and the local time each closed system creates for itself. At V7.9 that distinction lived in the framework’s ontology papers — Theory of Time and The Structure of What Happens — as ontology; V7.9’s own Lagrangian collapsed both clocks into one symbol. In V8.1 the distinction became a theorem, and the theorem is almost embarrassingly simple.
Suppose you try to use the oscillating scalar amplitude φ = A cos Θ as a clock, by taking its normalized gradient. At every turning point of the oscillation the gradient vanishes. A periodic amplitude therefore cannot carry a continuous global ordering. This is the gradient-clock obstruction, and it forces the theory’s central structural decision: the internal cyclic phase and the universal ordering cannot be the same object. STF must carry two clocks — a universal scalar T_U whose unit normal orients every causal response, and an internal phase Θ_I on the circle that records where the field lies in its cycle.
Everything downstream reorganized around that split. Because the unit normal depends on T_U only through its normalized gradient, the action knows which direction is the future but not the rate at which universal time advances against any particular clock — the Clock-Rate Invisibility Lemma. Every empirical statement in STF is therefore a comparison between clocks, and every sector of the current paper — the curvature response, the flyby record, the gravitational completion, the timing prediction — is a different face of that comparison. Readers of the framework’s companion papers will recognize the same separation doing the same job elsewhere; it entered the backbone as a theorem here.
V7.9 said the 3.32-year timescale followed from a cosmological threshold condition alone. The current edition states the dependency graph of every timing number and lets no number pretend to be what it is not.
Observation supplied the 3.32-year and 71-day anchors. The 3.32-year value is the mean separation between ultra-high-energy cosmic rays and gamma-ray bursts in the triple-coincidence events, and a blind likelihood independently returned the 11/8 exponent with a 3.31-year mean period. The emission-window closure supplied the approximately 54-year outer anchor — conditionally, on the observed centroid and one named boundary input; it is a closure, not an independent prediction. The Peters inspiral law then translates the three times into 1466, 730, and 360 Schwarzschild radii of binary separation — successive near-halvings, which remains the unexpected convergence — but the separations are GR images of the temporal anchors, not independently predicted radii.
None of these numbers is rigid; the 71-day and 1,212-day structures are observational associations with legitimate data and model freedom. And a standing firewall, asserted in every release checker, bars all of them from solving any theoretical coefficient or selecting any model element. They may serve as inputs to a declared forward likelihood, and as retrospective validation witnesses. Nothing else. That firewall is what makes the prediction at the end of this article a prediction.
Here the story turns from correction to construction, and it begins with the theory doing something theories rarely do in print. Version 8.2 superseded the claim that STF’s local interaction — the scalar coupled to the universal rate of the curvature norm — is a healthy fundamental metric action. Eliminate the curvature norm into a finite-order local metric theory and its Hessian generically produces a nondegenerate metric-acceleration block and a quartic pole of the wrong residue; none of five named repairs removes that physical rank with a constant constraint structure. The supersession is preserved in the record. What survived is a regulated readout — smooth through zero curvature, linear in the squared curvature norm at small argument — behind an exact causal high-pass memory with a single Drude turnover at 0.529 years. A statically curved universe produces no steady-state response; only change in the clock-relative curvature state is registered.
The gravitational question then became sharp: what parent theory carries this readout consistently? The answer was built adversarially, in layers, each frozen before the next began.
V9.0 attached a five-gate audit — open-operator classification, the covariant environment vertex, all-loop zero-DC protection, gravitational-wave emission, and merger-production activation — and four numeric acceptance gates that any coefficient-complete parent must pass at once: a Hulse–Taylor orbital-decay correction below 3.59 × 10⁻³, a PSR J1738+0333 flux correction below 0.181, a GW170817 chirp-rate envelope below 6.67 × 10⁻³, and a tensor-speed deviation below one part in 10¹⁵. All five gates opened; none closed at that layer.
V9.1 and V9.2 consolidated thirteen standalone calculation records. Among the permanent results: a no-go for the common-shift Stueckelberg completion; a compactification stop gate that closes the parameter-free route from the frozen compactification; the exact bookkeeping identity M_I = g_Q², which removes an adjustable amplitude from the pulsar sector; and two rejected shortcuts — ordinary neutron-star tides are not the universal scalar bath, and an isolated finite tube cannot supply the gapless continuum the memory needs.
V9.3 through V9.4.2 ran a fifty-one-gate covariant architecture programme. Its discipline record is the part I would ask a skeptical reader to look at first: four withdrawals, each by an explicit correcting gate — a premature no-parent-selected claim, a cubic window composition, a Carter-mobility identification, a one-datum-unlock implication — one declared tuning, labeled as such, and two branch suspensions with their structural results retained. The same programme produced a microscopic material-clock normalization from a neutron-superfluid phase cycle and retired the parent’s exact quadratic-DHOST label without claiming any universal DHOST no-go.
The comparative gravity lift studies — eleven by V9.5 — took the architecture to a compact-response boundary. A finite relational phase star was constructed as the frozen comparator; its radial operator was derived and its first eight eigenvalues computed positive, with fundamental ω₀R = 0.1889913902; a positivity theorem was proved for every closed Hermitian same-port response below the fundamental. And then the eleventh study proved the decisive negative, constructively: the parent as then written did not determine a unique compact response at all. Explicit families of arbitrarily small stable completions preserved every frozen datum while changing the modal residues and exterior projections. No unique response followed from the parent data alone.
This is the moment I want a reader to sit with. The theory proved that it did not yet predict. It could have fitted a response to the timing anchors — they were on the record, and every session that built the comparators knew them. Instead the release converted the stop into an exact, machine-readable completion contract and a version policy: no successor could advance by fitting timing or choosing a convenient projection, only by new action-level information that passed the contract.
One did. The Compact CRGC Port Completion Amendment — the programme’s one new theory choice since the freeze — replaced the symbolic compact map by an explicit eleven-component map, derived the mixed constraint block rather than dropping it (the naive unity coupling fails its own stability test once that block is included), obtained a radial stability window |g_R| < 0.4312478776, and froze the dyadic value g_R = 1/4 inside it with zero timing input. Its self-grade is exact and the paper carries it verbatim: a radial-classical completion pass — not a full nonlinear, nonradial, or quantum completion. On that scope, and only on it, the long-paused compact-port gate closed, while the five audit gates remain formally open. What changed nonetheless is categorical: a normalized response functional now exists on one explicit two-clock branch, so the question “what does STF predict?” stopped being ill-posed and became a calculation with named missing inputs. The controlling grade has not moved since the gravitational audit began, and it is boxed on the first page of the current edition:
STF is a coherent gravitational candidate — not a completed gravity theory.
With a normalized port in hand, the source question became concrete: what drives the compact response of a merging binary, and when? Four derived steps later, the answer is a source-clock law. Particles of magnetic rigidity ℛ — energy divided by nuclear charge — can first be produced at a time before merger that falls as ℛ^(−8/5); the exponent is derived, the coefficient C is a normalization with physical content. Charged transport through magnetic fields adds a delay that falls as ℛ^(−2), with a coefficient A that is ordinary propagation physics, fixable from independent measurements.
The detector lead L = Cℛ^(−8/5) − Aℛ^(−2) looks nonlinear, and it has an exact affine representation:
\[ X=\mathcal R^{2/5},\qquad Y=\mathcal R^{2}L,\qquad Y=CX-A. \]
Two rigidity-resolved events from one source, one emission episode, and one line of sight determine both coefficients exactly. Every third event is then a prediction with no refit permitted. There is even a parameter-free physicality test on any pair, and one of the first things the machinery does is refuse the framework’s own anchors: forcing the known 71-day and 1,212-day structures into it as a same-source pair yields a negative magnetic-delay coefficient and fails the test. That is the calculation’s own way of saying what the record already declared — the two structures belong to different association populations. They are witnesses, not calibration data.
The test is frozen before any qualifying data exists. The eligibility contract, the immutable calibration split (events ordered by dataset identifier, the first two eligible calibrate, all later events are held out), the rejection thresholds, and a forbidden-refit list that bars every escape route — the source exponent, the transport exponent, the frozen normalization band, membership changes after residual inspection, magnetic models selected by timing residuals, and any use of the legacy anchors as calibration — are published, with an executable evaluator, so the test can be run by anyone on the day a qualifying multiplet exists, without contacting the author.
The V9.6.1 closure then derived both coefficients from measurable inputs — source capacity physics for C, with a lift factor ℱ for production in a remote magnetic bubble rather than at the compact object, and line-of-sight magnetics for A — so that a source with an independently measured dossier yields a predicted lead for every rigidity with no timing calibration at all. This is also where the theory forks. The local capacity bands contain the observed structures — 1,212 days inside the 1 EV band, 71 days inside the 5 EV band — but that containment is graded retrospective, not blind, because the comparator inputs were chosen by sessions that knew the targets; and the local route then fails its own production audit, so the surviving route is the remote one — a conditional existence result, not a unique selection — whose reference lift ℱ = √5 moves the bands and assigns the 1,212-day structure a different rigidity band. The arithmetic on transport is stated without softening: the reference Galactic line of sight alone delays a 1 EV particle by about 874 years, which erases any premerger lead; even at 20 EV the lead survives only for lines of sight some forty times cleaner than the Galactic reference. The neutral channel has no such delay, which is why the gamma-ray-burst association is the clean one.
Then, for the first time since the freeze, the frozen law was confronted with an external measurement it had never used: Auger’s published composition. The 494-event population behind the observational programme — the public Auger Open Data release, cut at 20 EeV, provenance-pinned by file hash and reproduced independently from the raw files — was propagated event by event through both normalization branches. Rigidity is charge-normalized energy, so each branch’s rigidity band for the 1,212-day structure maps, at each event’s energy, to a window of nuclear charges. The two branches predict different nuclear physics. The local branch demands charges predominantly above iron; for 185 of the 494 events no physical nucleus lies in its window at all. The remote branch, at its reference lift, implies an intermediate window around oxygen through silicon.
The Pierre Auger Collaboration’s preliminary four-group mass-fraction analysis says this population is nitrogen-dominated, with a small iron fraction. Confronting likelihood with windows, with energy errors, the energy-scale systematic, and hadronic-model spread all propagated, the in-band consistency is at most a fraction of a percent for the local branch against eight to twenty-three percent for the remote branch; the completion of the 86 highest-energy events in Study 20 leaves the verdict unchanged, with seventeen events carried as honest bounds rather than fitted fractions. Conservative envelopes overlap, so no branch is excluded. The recorded verdict is exact: remote favored, with no branch exclusion.
What gives this weight is convergence rather than any single number. The framework’s own energy-stratified timing tests, run on the discovery side before any branch existed, found the timing signal carried by the sub-75 EeV population. Auger’s composition says that population is light-to-intermediate. And the remote branch — derived from capacity physics and bubble geometry with no timing and no composition input — independently demands the same intermediate charge range for the 1,212-day carriers. Three separately obtained relations select one population — population-level, with no joint likelihood claimed.
The paper is equally exact about what this is not. It is retrospective, not blind; the known anchors preceded every comparator. It is conditional on a within-group interpolation whose literal four-charge alternative leans the other way, disclosed and retained. And it rests on a single unmeasured number: the lift ℱ. The composition consistency holds robustly for ℱ between about 1.6 and 29 and breaks robustly below 0.7 or above about 67; the frozen reference √5 sits inside the band, and the closest simulation-leaning values straddle the broken and indeterminate zones. ℱ is now, by the paper’s own account, the single most consequential unmeasured quantity in the programme — and the paper forbids itself from setting it by fit. It must come from source observation or a simulation ensemble meeting a published precision specification.
Because the programme can now be wrong in more ways than at any earlier version, and the paper calls that its principal achievement. The standing falsifiers are listed — the half-cycle timing test, the rank and hyperbolicity and preferred-frame falsifiers of the frozen architecture, the four numeric gravitational gates, the new response-law rejections, and the composition axis: an independently determined ℱ in the falsification zones, or a measured heavy composition for the signal-carrying population, would each break the retrospective structure. The programme has bound itself, by frozen protocol and version policy, to be judged by an independently determined ℱ and one qualifying held-out multiplet.
And because the record is public. The backbone edition states each result at its exact grade in twenty-nine pages. Beside it, the Verification Record publishes the complete technical stack — every release with its historical layers, every calculation record, every checker, every negative result, every withdrawal — hash-pinned, and runnable by anyone with Python and NumPy. The paper is the face; the record is the proof. A reader who wants to check a single number can find the document that establishes it and run that document’s package.
V7.9 asked to be believed. V9.6.1 asks to be checked. The observations were unexpected; what grew from them is now a coherent gravitational candidate with a frozen prediction — one that can be tested once a qualifying sample or an independently measured ℱ exists and, by its own rules, lost. That is the delta.