T Theory has spoken in instruments — twenty-three rooms where the move is made by hand: every t crossed out, the count ν written in its place. The instruments show; they do not yet prove. This page is the program that either turns the showing into physics or says exactly where it fails. It commits, in public, to the discipline the corpus already claims: every seam declared on the face of the instrument.
The rules of this document. Every claim wears one of seven tags. Where something is derived, the derivation is on the page. Where something is postulated, the postulate is numbered and admitted. Where the framework genuinely forks from general relativity, the fork is computed to the order where the two roads part. And it ends the only way a physical theory is allowed to end: with a number, and the name of the telescope that will check it. Every number below was verified by machine — series expansion, numerical geodesic integration, quadrature — before it was printed.
Postulate The fundamental object is not a particle at a time. It is a history with a phase. Let 𝒞 be the configuration space of a system. A history is a curve γ in 𝒞. The state assigns to every history a unit complex number,
where N[γ] ∈ ℝ is the count functional — the number of complete turns of the wheel along that history. The count functional is required to satisfy three axioms:
A1 · Additivity. N[γ₁ ∘ γ₂] = N[γ₁] + N[γ₂] — counts of successive histories add.
A2 · Locality. N is an integral of a local density along the history.
A3 · No t. N[γ] is invariant under every reparametrization of γ. The count of a history cannot depend on how the history is described.
A3 is the whole doctrine stated as one line of mathematics. There is no external clock anywhere in the definition: “time” never enters, only the histories and their counts. Evolution is relational — one subsystem’s count read against another’s. This is not an exotic position; it is precisely the structure general covariance forces on any parametrization-invariant theory, where the Hamiltonian is a constraint and time is internal. T Theory does not soften that structure. It takes it as the starting point and gives the invariant a name: the count.
Revision 2 called worldline uniqueness a theorem of A1–A3. Review found the hole, and it is conceded: A1–A3 force only N = ∫F(x, ẋ) ds with F homogeneous of degree one in the velocity — which is the definition of a Finsler structure, an infinite-dimensional class in which the proper line element is a single point. Finsler deviations are the standard parametrization of Lorentz violation: constrained by experiment, not eliminated by these axioms. The missing assumption is therefore stated instead of smuggled: A4 — the degree-one density is the square root of a quadratic form. Equivalently, local Lorentz invariance. Closing the Finsler class from A1–A3 alone would be a real theorem and is debt №1; until then A4 is an axiom.
Given A1–A4, and given the quadratic form itself: A1 and A2 make N the integral of a local density along the curve; A3 with A4 leave exactly one invariant measure — the proper line element of that form. Hence
with ν₀ a constant of the system, whose value is not derived — it is the number physics has always called mass. And one import is declared rather than hidden: ∫dτ presupposes the quadratic form on 𝒞, which A1–A4 do not supply. The program's logic is therefore: the axioms force the count to read a metric; what supplies the metric is the mutual-rate field of Step 6. The count and the geometry are not derived one from the other — they are the same postulated structure, seen from its two ends.
And one separation, kept deliberately. That we measure duration by counting is a fact of metrology. That nothing but counts is needed in the foundations is the postulate of this step. That count gradients are what gravity is — that is Fork B of Step 6. Three different statements; the first is never allowed to prove the third.
Measured The count must be defined with no reference to gravity, or the program is circular. Here is the definition, and it is operational:
The count between two configurations of a system is the number of interference fringes the system can be made to exhibit between them — an integer read off an interferometer. No clock is consulted. The count rate ν of a system, quoted against a reference cycle, is its count per cycle of the reference.
For a single particle of mass m at rest, the measured rate is
— the Compton frequency. Three facts make this metrology rather than metaphor:
One precision, adopted from review verbatim: at present, ν = mc²/h is an operational identification, not a derivation of mass from counting. The form N = ν₀∫dτ is forced by the axioms given A4 (Step 1); the value ν₀ is measured, not explained. T Theory has not explained mass. It has recognized that the present system of units already defines mass as a count rate — and it takes that as its anchor, not as its trophy. And a second precision of the same kind: an interferometer reads a phase difference. Only differences of count are observable; the absolute N of a single history is not. The fundamental object of this theory is relational to the bone, and the page should not pretend otherwise.
Nothing gravitational was used anywhere above. The count is anchored before the first word about falling is spoken. For the electron, ν = 1.24 × 1020 Hz; for one kilogram, ν = 1.36 × 1050 Hz. These are not small print — they are the largest well-measured numbers in physics, and they are about to do all the work.
Postulate One dynamical postulate is admitted, and it is the oldest one in quantum mechanics:
The amplitude for a transition is the sum of the state over all histories connecting the configurations: A = Σγ e2πi N[γ].
Derived From SUM, the selection rule is a theorem, not a further assumption.
Let histories be labeled by a parameter s with count N(s) ≫ 1. Where N′(s) ≠ 0, neighboring phases 2πN rotate through full turns and cancel; the sum is dominated by the bundle around N′(s₀) = 0, of width ΔN ∼ 1. Hence:
The histories that survive are the histories of stationary count. The classical world is the stationary-count skeleton of the sum, sharp to one count — which is to say, to ħ.
For a free relativistic particle the count along a worldline is N = (mc2/h)∫dτ — proper time in disguise — so stationary count is stationary action, S.3 is the principle of least action, and extremal aging is extremal counting. Nothing was smuggled: the classical variational principle falls out of counting turns.
Written without cushioning, as review demanded: N = S/h. Steps 1–3 are the Feynman path integral in changed variables, and claim no result beyond it. The quantum sector of this program has so far derived nothing that the path integral had not already derived; it is the gravitational sector, from Step 4 on, that carries every claim of new content.
Sketch Grouping. Two systems that interact exchange phase; after the exchange their joint stationary bundle is shared, and neither can be assigned a stationary history alone. Grouping — the persistence of composite things — is count-locking, and its loss to an environment is decoherence. This paragraph is the program’s sketch, not its theorem; the rigorous version (pointer states from stationary joint count) is listed in the ledger of what is owed.
Postulate One gravitational postulate is admitted. It is a statement about rates, and every clock experiment of the last sixty-five years has measured it directly:
The presence of counts modifies the rate at which every collocated counter runs, by one universal factor R(x) ≤ 1 — the local count per count of the same system far away. Modifications compose multiplicatively: two sources multiply their factors, so ln R adds. Define the potential as the additive quantity, Φ := c2 ln R.
Derived Now let a body move slowly through a region where the rate varies. Its count is
Apply S.3 — stationary count. The Euler–Lagrange equation of the integrand is
This is the equation the corpus has carried from the beginning — g = −c2∇ln ν — no longer a statement but a two-line consequence of SUM + RATE. Read it plainly: a body falls because falling is how it keeps the most of its count. Nothing pulls. The stationary bundle simply lies down the rate gradient.
PostulateDerived The source side — what mass does to the rate — was a raw postulate in the first revision of this page, and review demanded it be derived or admitted underivable. It can be half derived, and the half matters: one postulate remains, but the form of the law is then forced.
The rate field is physical, so it carries a count density of its own, entering the total count opposite in sign to matter’s — as field energy does in Newton. (The relative sign, attraction, is read from the sky here as it is there.)
Only rate ratios are observable, so Φ → Φ + const is gauge: the field’s density may depend on ∇Φ alone. RATE’s exact superposition forces the field equation to be linear, hence the density quadratic. Locality and isotropy at lowest derivative order leave exactly one candidate, |∇Φ|², with one coupling. Stationary total count — matter (S.4) plus field — then yields
with G the lone empirical constant converting count density to rate gradient. For a point mass, Φ = −GM/r, and S.5 returns a = GM/r2: Newton — obtained from the counting axioms plus one declared postulate, not inserted by hand.
Owed And a gap review exposed that Revision 2 did not see: S.6 carries a mass density and nothing else. In Einstein’s theory pressure gravitates — it is what stiffens the Tolman–Oppenheimer–Volkoff equation and sets the neutron-star maximum mass, and neutron stars are the second-best strong-field laboratory in the sky. Does pressure count? Does radiation — whose quanta carry zero count (Step 5)? The source assignment for a radiation-dominated or pressure-dominated body is currently undefined, and the ledger now carries it.
Measured RATE is not a hypothesis in want of data; it is among the best-verified statements in physics — and note what each experiment actually reads: a frequency ratio. A count against a count.
| Experiment | Height / scale | Effect Δν/ν | Status |
|---|---|---|---|
| Pound–Rebka–Snider 1960/64 | 22.6 m | 2.47 × 10⁻¹⁵ | ✓ to 1% |
| Gravity Probe A (Vessot) 1976 | 10⁴ km rocket | ~4 × 10⁻¹⁰ | ✓ to 7 × 10⁻⁵ |
| Chou et al., Al⁺ clocks 2010 | 33 cm | 4 × 10⁻¹⁷ | ✓ |
| Bothwell et al., Sr lattice 2022 | 1 mm | 1.1 × 10⁻¹⁹ | ✓ — one clock, one millimeter |
| COW neutron interferometry 1975→ | quantum phase | ΔN = ν·(gΔh/c²)·T | ✓ ~1% — gravity enters QM as a count-rate gradient |
| Müller–Peters–Chu reading of atom interferometry 2010 | Compton frequency | redshift at νC | disputed — Wolf et al. (Nature 467, E1, 2010) read it as a free-fall test, not a redshift; the exchange never resolved in the original authors’ favour. The redshift case rests on the clocks above, not on this row. |
Derived First, an honest failure, because it is the failure that forces the next move.
If the rate field touches clocks alone — ds² = R²c²dt² − dx² — then light, by stationary phase, obeys Fermat’s principle in an effective index n = 1/R ≈ 1 + GM/c²r, and the deflection at the solar limb comes out α = 2GM/c²b = 0.87″. This is Einstein’s own 1911 half-value. The sky says 1.75″. A rate field that touches only time is experimentally dead — killed in 1919 and re-killed by every VLBI season since.
Postulate The repair is one axiom, and it is the only further axiom the geometry needs:
The count field sets all measures. The factor that divides the clock multiplies the ruler: where the count rate is R, lengths carry counts as 1/R. One field, one seam, zero new constants:
The exponential is not decoration: it is forced by RATE’s multiplicative composition. Superposition is exact — Φ of many bodies adds — a property even Einstein’s theory does not keep.
Derived Expand S.7 to first order: g₀₀ = 1 + 2Φ/c², gij = −(1 − 2Φ/c²)δij — exactly the weak-field metric of general relativity. To the next order, e2Φ/c² = 1 + 2Φ/c² + 2Φ²/c⁴, which in the parametrized post-Newtonian bookkeeping is
— the same PPN point as GR. All four classical tests therefore come out at full value, and were re-derived here from S.7 directly, numerically, before being printed:
| Test | Count gravity | Observed | Status |
|---|---|---|---|
| Light deflection at the limb (index n = e−2Φ/c², ray-traced) | 4GM/c²b = 1.75″ | VLBI: γ−1 = (−0.8 ± 1.2)×10⁻⁴ | ✓ |
| Shapiro delay (Cassini 2003) | 1PN, γ = 1 | γ−1 = (2.1 ± 2.3)×10⁻⁵ | ✓ |
| Mercury perihelion (geodesics of S.7, quadrature) | 42.98″/century | 42.98 ± 0.04″/cy | ✓ |
| Gravitational redshift | RATE, directly | see Step 4 | ✓ |
| Lunar laser ranging (Nordtvedt η = 4β−γ−3) | η = 0 | η = (−0.2 ± 1.1)×10⁻⁴ | ✓ |
Take S.1′ seriously: along a null curve dτ ≡ 0, and ν₀ = mc²/h = 0 for a photon. The count of light is zero twice over — light has no phase-count in this framework. What light obeys is the metric. Every result on this page that rides on a photon — the 0.87″ failure, the 1.75″ success, the Shapiro delay, and the entire shadow prediction of Step 7 — is therefore a statement carried by S.7, which the count postulates select but do not constitute. Said at referee strength: strip the counting language out and the physics of Steps 5–8 is unchanged. The counting is the road by which the metric was found; the metric is the carrier of the predictions; and the page stops pretending otherwise. Extending the axioms to massless fields is open, and the ledger carries it.
Owed A static scalar rate field has no answer for moving sources. Gravity Probe B measured the Lense–Thirring drag of Earth’s rotation at 37.2 ± 7.2 mas/yr (GR: 39.2); LARES tightened it to a few percent. Frame dragging is real, and a scalar alone does not produce it. The program therefore owes the count current: promote the count density to a 4-current jμ = (ν, νv/c), whose curl drags the stationary bundle — gravitomagnetism in count language. Until that sector is derived, count gravity rides with the measured gravitomagnetic limit rather than predicting it. Declared, not hidden — and sharpened in Step 6 into a definite prescription, where Revision 2 marks this debt conditionally paid.
Derived Set the count metric beside Schwarzschild in isotropic coordinates and expand both in u = GM/c²r. The coefficients below were extracted numerically from both metrics:
| 1 | u | u² | u³ | |
|---|---|---|---|---|
| g₀₀ · count | 1 | −2 | +2 | −4/3 |
| g₀₀ · Schwarzschild | 1 | −2 | +2 | −3/2 |
| gsp · count | 1 | +2 | +2 | · |
| gsp · Schwarzschild | 1 | +2 | +3/2 | · |
The two theories are identical through first post-Newtonian order — everything the solar system can currently measure — and part company at 2PN: a coefficient of 1/6 u³ in time, 1/2 u² in space. At Mercury, u ≈ 2.6 × 10⁻⁸; the departure is a factor ~10⁻⁸ below the leading correction, invisible to Cassini-class precision — in the solar system. Revision 2 wrote “forever” here; review struck the word, correctly: compact binaries sweep u ∼ 0.1–0.3 straight through the interferometers’ band, and gravitational-wave phasing reads the conservative 2PN coefficients directly. The seam between count gravity and Einstein lives where u is of order one — and the sky has been listening there since 2015.
The seam now has a physical name. The exponential map reproduces the first nonlinear order of Einstein’s equations exactly — that is why β = 1 — and departs at the second. In general relativity the field’s own energy gravitates, and keeps gravitating; in count gravity the underlying field stays linear and all apparent nonlinearity is the multiplicative composition of rates. Through 1PN the two mechanisms are indistinguishable by mathematical identity; at 2PN they are different theories. The fork is, precisely: does gravity’s own count gravitate — or do rates merely compose?
Review demanded a computation of the 2PN phase coefficient, not an assertion. It was run before this revision was written. For circular orbits of a test mass, the count metric’s energy has a closed form:
Its gauge-invariant expansion E(x) = 1 − (x/2)(1 + e₁x + e₂x² + …), with x = (MΩ)2/3, gives e₁ = −3/4 — identical to GR, as β = γ = 1 requires — and
Propagated through stationary-phase inspiral phasing with the GR flux held fixed, the shift is δφ̂₄ = −0.33 in the test-mass limit. (Not an interval: the comparable-mass value requires the theory’s two-body dynamics and is not computed — in GR the same coefficient moves 18% between mass ratios.)
The comparator, checked rather than asserted — Revision 3 claimed “bounds at O(10%)” without a citation; Revision 4 checked; Revision 5 corrects Revision 4’s reading of what it checked. The current measurement is GWTC-4.0 (arXiv:2603.19020, v2 of 20 July 2026; 42 O4a events + 49 earlier; PN bounds improved 1.2–5.5× over GWTC-3). Its Table 2 gives δφ̂₄ at 90%: hierarchical −0.02 +0.45/−0.43 (FTI·EOB), +0.07 ±0.45 (TIGER·Phenom); shared-value “restricted” −0.03 +0.16/−0.15 and +0.06 +0.24/−0.25. At the mass ratios that actually set these bounds (η ≈ ¼), the prediction is δφ̂₄ = −10/α₂(¼) = −0.216 — on the restricted edges (−0.18/−0.19), outside by roughly 15% of the interval half-width, with the declared numerator uncertainty (±18%, the size of e₂’s own η-running in GR) straddling both edges.
And the framing must be inverted — Revision 4’s verdict sentence comes out. The theory predicts an approximately absolute shift Δαᵢ; δφ̂ᵢ is a fractional deviation of an η-dependent GR coefficient. A universal theory therefore maps to a per-event-varying δφ̂ pattern — which is what the hierarchical reading permits and the shared-value reading forbids. “The reading closest to a universal theory” was exactly backwards. Neither published analysis tests the actual prediction — a correlated, η-dependent, sign-structured pattern across PN orders — and no verdict is drawn here from either. The only correct instrument is a likelihood-level comparison against the event posteriors; with exact Δe₂, Δe₃ and transfer weights in hand and the posteriors public, it is the fastest route in this program to a real result.
The flux escape, quantified. The SPA algebra gives α₂ = 10(3e₂ − 2e₁f₁ + f₁² − f₂), verified against TaylorF2 to all digits; hence ∂α₂/∂e₂ = +30 and ∂α₂/∂f₂ = −10. Since 3Δe₂ = −1 exactly, Δα₂ = −10 − 10Δf₂: a theory 2PN flux shift of Δf₂ = −1 cancels the entire tension identically. The dissipative sector enters at the same weight as the conservative one, and only one of the two has been computed. Δf₂ is now the sharpest single number the program owes.
The tower, recomputed after review corrected itself. The fourth review found its own earlier 3PN transfer incomplete — and this page had printed it. Both re-derive the full decomposition now, and the shipped script covers it. Δe₂ also feeds the 3PN coefficient, through the −3e₂f₁ cross term: ∂α₆/∂e₂ = 60f₁ = −6235/28 alongside ∂α₆/∂e₃ = −80, so the flux-fixed shift is Δα₆ = +155.6 + 74.2 = +229.8, not +155.6. And Revision 4’s fractional comparison was ill-posed: α₆(η) crosses zero at η ≈ 0.053 (q ≈ 17), inside the catalogue, so δφ̂₆ runs from +1.3 through a pole to −0.35 at equal mass — “≈ +0.9” described a mass ratio no event occupies. In the units that stay meaningful — absolute coefficient shifts at η = ¼ — the restricted bounds are ≈ 7.2 at 2PN and ≈ 39 at 3PN, against |Δα₂| = 10 and |Δα₆| = 230: excesses of 1.4× and 5.9×. 3PN is the harder constraint; Revision 4 undersold it at the right η while overselling it at the wrong one.
The escape and the escalation are one unknown. ∂α₆/∂f₂ = 4975/42 = +118.5: the same Δf₂ = −1 that zeroes the 2PN tension takes Δα₆ from +229.8 to +111.3 — the flux shift that saves 2PN halves the 3PN excess to ≈ 2.8× before Δf₃ is even computed. There is no independent escalation argument. There is one pair of numbers, Δf₂ and Δf₃, and they are the most consequential computation in the program.
And the tower extends beyond the inspiral. While auditing the ray tracer, review derived — and this revision verified by 30-digit quadrature of the exact Fermat integral — that the count metric’s second-order light-bending coefficient is exact and closed form: α = 4GM/c²b + 4π G²M²/c⁴b², against general relativity’s 15π/4 — a ratio of exactly 16/15, a 6.7% excess in second-order bending. (Richardson limit of the quadrature: 12.56632 against 4π = 12.56637.) Unmeasurable at the solar limb, where M/b ∼ 2×10⁻⁶; exact, and in the same family as Δe₂ = −1/3 and Δe₃ = −35/18.
For the record: the ISCO sits at u = (3−√5)/4, x = 0.1589 against Schwarzschild’s 1/6 — a ~5% earlier termination in x, hence a shifted merger frequency: itself a prediction, with the inspiral releasing 5.5% of Mc² against GR’s 5.7%. No conflict with observed peak frequencies is claimed or checked here.
Schwarzschild has a horizon: g₀₀ = 1 − 2u reaches zero at r = 2GM/c². The count metric’s g₀₀ = e−2u is positive for every r > 0. The count never reaches zero; it is only exponentially suppressed. No horizon. Instead, computing the areal radius 𝓡(r) = r eu and minimizing:
— a throat. No sphere of circumference smaller than 2πe GM/c² exists in the geometry. This structure has a literature: the exponential metric is the Yilmaz metric (1958), and Boonserm, Ngampitipan, Simpson & Visser (2018) showed it describes a horizonless, traversable throat. The program claims no priority for the metric — only the counting road that leads to it.
Folklore says “no horizon ⇒ gravitational-wave echoes.” The count metric refuses its own folklore, and the computation is two integrals. Inbound light obeys dt = e2udr: the coordinate time from 3M down to 0.1M is already 2.7 × 10⁶ M, to 0.05M it is 3.1 × 10¹⁴ M, and it diverges as r → 0 — the throat is infinitely deep in exterior time, so nothing returns to bounce: operationally it swallows like a horizon. Yet a falling body’s proper time to the bottom, ∫dr/√(1 − e−2u), is finite (≈ 3.53 M from 3M). Infinitely deep from outside, a short walk from inside. The honest consequence: no clean echo prediction — the discriminator must be sought elsewhere, and Step 7 names it. The same depth is the beginning of an answer to the surface-emission constraints on horizonless objects (Broderick & Narayan 2006): a surface infinitely deep in exterior time thermalizes nothing outward on any astrophysical timescale. The beginning only — the dynamical statement, that collapse actually forms such a throat, appears nowhere on this page and is owed.
There are exactly two consistent readings of T Theory from here, and the program must choose:
Fork A — interpretation. Adopt Einstein’s field equations as the law of the count field. Then T Theory is a language for general relativity — proper time renamed the count, geodesics renamed stationary counting. Every prediction agrees with GR forever. Safe, and unfalsifiable against it.
Fork B — dynamics. Keep the composition axiom of RATE — exact superposition, the exponential map — as the deeper truth. Then the field equation is S.6 with S.7, not Einstein’s: identical through 1PN, horizonless at strong field, and in open disagreement with GR at every compact object in the sky.
This program takes Fork B, because only Fork B can be killed — and a theory that cannot be killed cannot be confirmed either. What follows is Fork B’s exposure, computed.
A scalar rate cannot say how moving counts drag, nor what radiates — and a vector cannot rescue it: spin-1 exchange makes like masses repel. The completion the structure itself demands is the count of each direction of measure against each other — a symmetric tensor of mutual rates Kμν. Ten counts, not one. Fork B then has a definite prescription:
— the linear field forced by exact superposition and the ratios-only gauge (they admit exactly one massless spin-2 structure: linearized Einstein), the observable metric its exponential map. What that one sentence buys, and what it owes:
Review is right that “unproven” was too gentle by a category. There is a standing argument that the sharpened Fork B cannot be done. A free massless spin-2 field on flat space obeys a linearized Bianchi identity forcing ∂μTμν = 0 with respect to the background η — but matter in this theory moves on g = η e2K, whose stress tensor is not η-conserved. The mismatch appears at second order, and iterating the repair is the classical bootstrap that terminates in Einstein’s equations (Gupta 1954; Feynman’s lectures on gravitation; Deser, Gen. Rel. Grav. 1, 9 (1970); Wald’s uniqueness theorems point the same way). Misner’s specific attack on the Yilmaz theory (gr-qc/9904052) argues it admits no consistent nontrivial vacuum and requires a distributional source at the origin — at the very throat this page relies on.
The single candidate escape is the composition postulate itself: if the observable metric is η e2K while K stays linear, the bootstrap’s premise — that the field must couple to the total, g-conserved stress tensor — is refused rather than iterated. Whether that refusal can be made into consistent energy bookkeeping is the open problem of this program, and it gates everything: the metric of Step 6, both kill tests, the fork itself. If Deser’s argument cannot be evaded, Fork B is not wounded but dead, and this page will say so in one line at the top.
Two further structural admissions, demanded by review and owed to the reader. First: Φ is defined against “the same system far away” — an asymptotically flat, quasi-static reference. Where no such region exists, in cosmology or dynamical collapse, the central variable of this theory is undefined. The quarantine of cosmology in the ledger is therefore not scheduling but structure: as formulated, this is a theory of asymptotically flat systems, and it stops claiming to be a foundation of everything. Second: the sharpened Fork B carries a fixed background η, and a fixed background is exactly the kind of description-dependence A3 forbids. The tension is real, and it is recorded here rather than resolved.
Derived Null orbits of S.7 conserve E = e−2uṫ and L = e2ur²φ̇; the radial equation is ṙ² = E² − L²V(r) with effective potential
Schwarzschild’s critical impact parameter is 3√3 GM/c² = 5.196. The shadow a distant observer sees is the disk of diameter 2bcrit. Therefore:
The shadow of every compact object is 4.63% larger than the Schwarzschild shadow of the same mass — where θg = GM/c²D is fixed independently by stellar orbits. And the sign is the weapon: in general relativity, spin only ever shrinks the shadow (Kerr: 0 to −7.5%), while the static count solution pushes it the other way. Stated at exactly its evidential strength, no more: the static solution predicts a positive 4.63% shift; whether this remains a robust discriminator after inclusion of rotation is an open calculation — debt №3.
Owed The rotating count metric is not yet derived; the prediction above is the static one, and the seam is declared: if count-spin corrections proved large and negative, the sign argument would weaken. The object to compute is now definite — the exponential-map Kerr analogue of the sharpened Fork B (Step 6) and its shadow. Until that calculation exists, the static number stands as the program’s stake, and the sign argument is a conjecture about it, not a theorem.
| Source | θg (prior) | GR / Kerr shadow | Count shadow | EHT measured ring |
|---|---|---|---|---|
| Sgr A* (M = 4.297×10⁶ M☉, D = 8.28 kpc) | 5.125 μas | 49.3 – 53.3 μas | 55.7 μas | 51.8 ± 2.3 μas |
| M87* — retained for scale only. Its measured quantity is a ring, not a calibrated shadow, and its mass prior is not independent: stellar dynamics gives ≈6.2×10⁹, gas dynamics ≈3.5×10⁹, and the 6.5×10⁹ used here was adopted within EHT’s own GR-based analysis. No lean in either direction is claimed from this row. | 3.82 μas | 36.7 – 39.7 μas | 41.5 μas | ring 42 ± 3 μas |
The Event Horizon Telescope measured both shadows. Its deviation parameter δ is exactly the quantity S.11 predicts. The present standing:
| Observation | Result | GR | Count gravity |
|---|---|---|---|
| Clocks vs height (P–R–S → GPA → mm optical) | to 10⁻¹⁹/mm | ✓ | ✓ RATE, directly |
| Compton clock; SI mass↔frequency | ν = mc²/h realized | silent | ✓ its anchor |
| COW neutron phase; atom interferometry | ~1%; 7×10⁻⁹ | ✓ | ✓ count coupling in QM |
| Mercury perihelion 42.98″/cy | ±0.04″ | ✓ | ✓ (β=γ=1) |
| Solar light bending 1.75″ | γ−1 ~ 10⁻⁴ | ✓ | ✓ full value |
| Cassini Shapiro delay | γ−1 = (2.1±2.3)×10⁻⁵ | ✓ | ✓ |
| Lunar laser ranging (β) | |β−1| < 10⁻⁴ | ✓ | ✓ |
| Frame dragging (GP-B, LARES) | 39.2 mas/yr ±19%, ±2% | ✓ | ✓ under sharpened Fork B — linear tensor sector ≡ GR’s; consistency proof owed |
| Binary-pulsar decay (Hulse–Taylor; double pulsar) | quadrupole to 6×10⁻⁵ | ✓ | ✓ quadrupole from the linear tensor sector (1PN flux corrections ∼10⁻⁶, below test); energy-bookkeeping proof owed |
| GW speed (GW170817) | |c_gw/c − 1| < 10⁻¹⁵ | ✓ | ✓ null structure shared |
| GW inspiral phasing — 2PN/3PN conservative (GWTC-4.0) · kill test I | absolute restricted bounds at η=¼: ≈7.2 (2PN), ≈39 (3PN) | ✓ | |Δα₂| = 10 (1.4×), |Δα₆| = 230 (5.9×) — pressed; Δf₂ = −1 zeroes 2PN and halves 3PN to 2.8×; likelihood-level test owed |
| Ringdown spectra (LIGO/Virgo) | Kerr-consistent, ~10% | ✓ | QNMs shift ~5% — below current resolution; XG/LISA territory |
| EHT M87* shadow | ring 42 ± 3 μas | ✓ | not evaluated — ring ≠ shadow; mass prior not independent |
| EHT Sgr A* shadow | δ = −0.08 ± 0.09 / −0.04 +0.09/−0.10 | ✓ | +4.6% at ~1σ–1.4σ · the kill test |
Standing: pressed at the edge, pending its own flux. A compatibility window at the EHT is not evidence, and this page says so in those words. Every experiment that decides at 1PN the theory passes at full value, because at 1PN it is general relativity. The moving-source and radiative sectors pass conditionally under the sharpened Fork B — conditionally, because the whole construction sits under a named obstruction (Step 6).
On the inspiral, precisely, against GWTC-4.0, at the mass ratios that set the bounds: δφ̂₄ = −0.216, on the restricted edges, with an uncontrolled numerator uncertainty comparable to the margin — and Revision 4’s restricted-vs-hierarchical verdict is withdrawn as structurally backwards, because an absolute-shift theory maps to per-event-varying fractional deviations that neither published reading tests. In the units that stay meaningful, the conservative tower exceeds the restricted bounds by 1.4× at 2PN and 5.9× at 3PN in absolute coefficient terms — and the single flux number Δf₂ = −1 that erases the 2PN excess also halves the 3PN one to ≈ 2.8×. Not one conspiracy per order: one pair of numbers, Δf₂ and Δf₃, plus the likelihood-level comparison. All three are computable, and they decide.
Three exits, numbered. The flux and two-body computations confirm the tension — Fork B dies at the interferometers, without a telescope, and this page is revised one last time to say so. They cancel it — Fork B has survived its sharpest test and earned the right to the shadow: δ = +4.63% against Kerr’s 0 to −7.5%, still conditional on the rotating solution (debt №3). Or the obstruction of Step 6 resolves against the program first, and the question is moot. The nearest executioner is unchanged: the interferometers.
This page is itself Step 9: standalone, self-contained, no external scripts, no library — every figure hand-set, every number machine-verified before printing (series coefficients; ray-traced deflection converging to 4GM/c²b, with a 1/b residual that is this metric’s own second-order bending term, exactly 4πG²M²/c⁴b² — 16/15 of GR’s; perihelion quadrature matching 1PN at the 10⁻³–10⁻⁴ level over its converged range a ≤ 10⁴ M; the throat integrals; the closed form S.12 checked against direct geodesic computation to 4×10⁻¹⁶ and e₂ = −89/24 confirmed by Richardson extrapolation). Review pointed out that asserted verification is itself an undeclared seam — so the code ships beside the page, and in Revision 6 it has survived its first hostile audit:
verify/verify-math.js — series, photon sphere, deflection ray-tracer (domain-corrected,
compared against the exact 4π second-order term), perihelion quadrature, throat integrals, shadow numbers
sha256 276dd7a70a4e595beb9c088e5c5c4d5814d16b8d1b9834e7d208554d51584fd6
verify/gw-2pn.js — kill test I: E(x) coefficients, phase-transfer factor (asserting), δφ̂₄, ISCO · unchanged since audit
sha256 e9a7f92f3c6544c5969dc1e204769ed2d0f1c89a30ff9a6489cac79f66ca6096
verify/pn-tower.js — Revs. 4–7: the PN tower and full SPA decomposition — e₃ = −7195/576,
Δe₃ = −35/18, α₂ against TaylorF2, the ∂α₆/∂e₂ = 60f₁ cross term, ∂α₆/∂f₂ = 4975/42,
α₆(η) with its zero at η ≈ 0.053, the absolute-unit comparison, the Δf₂ coupling; plus
e₂ by Richardson, the closed-form E(u) check to 4×10⁻¹⁶, and the 4π deflection tower
sha256 29e49fda14469183da3dba3e035e3d5e9970820bcba3ffc1fbf3a0470653f234
Plain Node.js, no dependencies. Run each with node verify/<script>; every number printed above falls out. A hash without the file is a commitment device, not reproducibility — the scripts travel in the same folder as this page. “Trust the page” is hereby converted to “check the page.”
What remains open is stated with the same bluntness as what is closed:
Review observed that Fork A is a guaranteed retreat, and that courage about Fork B is cheap while the retreat stands open. So, on the record: if the obstruction (debt №2) resolves against the exponential map, and the Finsler class closes onto the quadratic form (debt №1), then the axioms’ only consistent completion is Einstein’s theory, the count adds no derivational content beyond N = S/h, and this program — not just its fork — ends. The page will be revised one final time to say exactly that, and the corpus will carry it.
count axioms → source law ✓ → metric ✓ → 1PN ✓ → dragging ✓* → waves ✓* → 2PN inspiral ⚑ → rotating solution → shadow
the golden path · ✓* conditional on debt №2 · ⚑ kill test I pressed at the edge — the flux tower decides
The second referee review was answered the same way as the first: by computation where computation was possible, by concession where it was not. The Finsler gap in the worldline theorem was conceded and A4 stated (Step 1). The metrology was cut back to what was actually built, and two contested citations are now flagged as contested (Steps 2, 4). N = S/h is written plainly (Step 3). The count’s silence about light is conceded: the metric carries the photon predictions (Step 5). Debt №2 was promoted from ledger line to named obstruction — Deser and Misner, on the face of the page (Step 6). “Invisible forever” was struck, and the referee’s free kill test was run before the revision was written: Δe₂ = −1/3 exactly, δφ̂₄ ≈ −(22–33)% against O(10%) bounds — Fork B is wounded (Step 6, kill test I). The M87* lean was withdrawn and the δ-calibration assumption declared (Steps 7–8). The verification code ships beside the page, hashed. And the ledger grew from five debts to nine, because that is the direction honesty moved.
The third review verified every number in kill test I and then asked the question this page should have asked itself: against what? Revision 3’s “bounds at O(10%)” was uncited, and announcing a tension without checking the comparator is the same species of error as announcing a survival without checking it — the failure mode had inverted, not disappeared. Revision 4 checked. The comparator is GWTC-4.0 Table 2, published nine days before this page: δφ̂₄ = −0.33 is inside the hierarchical intervals (±0.45), outside the restricted ones (±0.16/±0.25). “Wounded” became “pressed at the edge.” Review also supplied, and this revision independently re-derived, the two results that reframe the test: the exact cancellation Δα₂ = −10 − 10Δf₂ (the theory’s own uncomputed 2PN flux can erase the tension at Δf₂ = −1), and the next rung of the tower, Δe₃ = −35/18 exactly, whose 3PN shift ≈ +0.9 escalates the stakes unless the flux compensates order by order. One conspiracy per order, or a real exclusion — the flux tower decides, and it is computable. The tower script ships hashed beside the others.
The fourth review did the rarest thing in the genre: it found its own error propagating through this page, and said so. Revision 4’s 3PN number was wrong three ways. The numerator was incomplete — Δe₂ also feeds α₆ through the −3e₂f₁ cross term (∂α₆/∂e₂ = 60f₁), so the flux-fixed shift is Δα₆ = +229.8, not +155.6. The fractional denominator changes sign inside the catalogue — α₆(η) = 0 at η ≈ 0.053 — so δφ̂₆ has a pole there and “≈ +0.9” described a mass ratio no event occupies. And the restricted-vs- hierarchical verdict was structurally backwards: an absolute-shift theory maps to per-event-varying fractional deviations, so the shared-value reading Revision 4 called “closest to a universal theory” is the one reading such a theory cannot inhabit. All three are fixed, in absolute units at the η where events live: 1.4× the restricted bound at 2PN, 5.9× at 3PN — 3PN was undersold at the right mass ratio while being oversold at the wrong one. And the escape and the escalation collapsed into one unknown: ∂α₆/∂f₂ = +118.5 means the Δf₂ = −1 that zeroes 2PN also halves 3PN to ≈ 2.8×. Every corrected number was re-derived here before printing; the shipped tower script now carries the full decomposition. Three successive retreats on one test is the record working as designed — each forced by arithmetic, each on the page — and what is owed next is not another comparison but a computation: Δf₂, Δf₃, and the likelihood-level test.
The fifth review round did what no earlier round could: it ran the folder. The physics text passed untouched — every number in §Rev-5 re-verified — but the audit found that gw-2pn.js, the script named as verification of the flagship number, disagreed with it by 7.6% and printed a failing self-check (K = 32.2 against “analytic 30”) without flagging it. The published −0.33 was right — exact algebra, thrice derived — but a folder that contradicts its page is worse than no folder. Both bugs were fitting artifacts, and the fixes were tested by the auditor before being demanded: the E(x) fit truncated too early over too wide a range (now six terms, u ≤ 0.01: every coefficient exact to the printed digits), and the phase fit’s alignment basis was wrong — tc enters as x3/2, not x, and the missing x1/2 term leaked into the measured coefficient (now the correct basis, window x ≤ 0.02: K = 30.097, and the self-check asserts — a check that reports failure and continues is not a check). Also corrected: the perihelion quadrature’s cancellation-limited row is dropped and its converged range stated; the deflection residual is now correctly described as the physical second-order bending term ∼15πM/16b — a stronger result than the 1/b² convergence the page had wrongly claimed; dead code deleted. Scripts re-hashed and re-issued. Three wrong numbers out of thirty printed, all fitting choices, none behind a published claim — and none findable by anything but running the code. That is what shipping it is for.
The sixth round closed the audit: hashes match, scripts run clean, the folder reproduces the page. It also returned two results, offered for checking, and both were verified here before printing. First, a gift: while checking the ray tracer’s residual, review derived that the count metric’s second-order light-bending coefficient is exactly 4π — against GR’s 15π/4, a ratio of exactly 16/15 — confirmed here by 30-digit quadrature of the Fermat integral, reproducing review’s table to every digit. A new exact entry in the tower, found because the code was shipped and someone hostile ran it. Second, the ray tracer’s last 0.03-percentage-point residual is the finite integration domain, analytically b²/2X² — applying it reproduces all three printed rows to the digit; the script now corrects for it and compares against the exact 4π term. Two prose–code gaps were also closed: e₂ by Richardson and the closed-form E(u) check are now printed by the shipped script, not merely claimed. Where the corrections have landed over six rounds: axioms → comparator → partial numerator → two fitting bugs → a domain truncation worth 0.03 points → a new exact result. That is what convergence looks like. The theory queue is unchanged: Δf₂ and Δf₃, the two-body dynamics, the likelihood test, and Deser — still gating everything.
Continuity of the corpus: the move itself at The Checkpoint, the equations at Unified Gravity · The Equations, the first numbers-in-numbers-out at The Bridge, the solar system counted at The Unified Orrery. This page is the spine those rooms hang from — and the first one that names its own executioner.