The wave that came back
On 14 January 2025 a pair of black holes merged, and on 25 June 2026 the collaboration reported what the ringdown carried. The headline result is not the masses or the spins. It is that a single post-merger wave encodes a direct measurement of the remnant's horizon: anything falling toward a rotating horizon is dragged into orbit at the horizon's own angular velocity, while the signal it sends outward decays exponentially at a rate fixed by the surface gravity, because the same gravity redshifts it away. For the first time the near-horizon connection was read off from outside the boundary, without crossing it.
DRK-168 made a claim that this observation now anchors. It argued that clock synchronization is not a fact about spacetime but a connection on it; that the question of whether the whole universe can be synchronized at once is the question of whether that connection is flat; and that the residual desync around a closed loop is a holonomy — a class in $H^1$. The one-way speed of light cannot be measured because there is no Archimedean point outside spacetime from which to set the gauge. OPERA was that gauge failing in miniature: a loose fiber read off as a property of a neutrino.
This post completes the argument into the sector where the connection becomes violent. The thesis in one line: frame-dragging is the holonomy of the synchronization connection, the horizon is that connection's extremum, and the information paradox is a statement about how the cost of inverting it scales.
1. Frame-dragging is the gravitational Sagnac gap
Write the stationary metric in the weak-field gauge ($c = 1$):
$$ ds^2 = -(1 + 2\Phi)\,dt^2 + 2\,A_i\,dx^i\,dt + (1 - 2\Phi)\,\delta_{ij}\,dx^i dx^j, $$
where $\Phi$ is the gravitoelectric (Newtonian) potential and $A_i = g_{0i}$ is the gravitomagnetic potential. For a mass with angular momentum $\mathbf{J}$ the Lense–Thirring solution gives
$$ \mathbf{A} = -\,2G\,\frac{\mathbf{J}\times\mathbf{r}}{r^3}, \qquad \mathbf{B}_g = \nabla\times\mathbf{A}, $$
and the dragging of inertial frames is precisely the curl $\mathbf{B}_g$.
Now recall the synchronization connection of DRK-168. Einstein synchronization assigns a global time coordinate by transporting clock offsets; the offset accumulated along a spatial path is the line integral of the one-form
$$ \omega_{\text{sync}} \;=\; \frac{g_{0i}}{g_{00}}\,dx^i \;=\; -\frac{A_i}{\,1 + 2\Phi\,}\,dx^i \;\approx\; -A_i\,dx^i . $$
Transport it around a closed spatial loop $\partial\Sigma$ and apply Stokes:
$$ \boxed{\; \Theta \;\equiv\; \oint_{\partial\Sigma}\omega_{\text{sync}} \;=\; -\!\int_\Sigma (\nabla\times\mathbf{A})\cdot d\boldsymbol\Sigma \;=\; -\!\int_\Sigma \mathbf{B}_g\cdot d\boldsymbol\Sigma \;} $$
The desynchronization holonomy $\Theta$ equals the gravitomagnetic flux through the loop. It is nonzero exactly when the loop encircles angular momentum. This is the gravitational completion of the Sagnac source listed in DRK-168: the Sagnac effect is the special-relativistic shadow of $\Theta$ for a rotating frame; frame-dragging is its general-relativistic body. GW250114 did not measure a metric coefficient in the abstract — it measured $\Theta$ at the place where it is largest, the horizon, where the dragging rate saturates at the horizon angular velocity $\Omega_H$.
The desync gap of DRK-168 was therefore never merely an engineering artifact of loose fibers. Around a spinning mass it is a feature of spacetime, and it has now been seen.
2. Frame-dragging bounds coherence (a gauge-invariant inequality)
Discretize. Place $N$ zero-angular-momentum observers (ZAMOs) at the vertices of a cycle graph $C_N$ encircling the mass, each holding a local synchronization section $x_v$ (a phase offset). The edge restriction maps are parallel transport of sync: edge $e=(v,v')$ carries the connection weight
$$ \theta_e \;=\; \int_e \omega_{\text{sync}}, \qquad \sum_{e} \theta_e \;=\; \Theta . $$
The twisted coboundary is $(\delta^0 x)_e = x_{v'} - x_v - \theta_e$, and the optimal global section $x^\star$ minimizes the Dirichlet energy $\lVert \delta^0 x\rVert^2 = \sum_e (\Delta_e - \theta_e)^2$ subject to the cycle constraint $\sum_e \Delta_e = 0$ (offsets must close around the loop). The Lagrange solution is $\Delta_e = \theta_e - \Theta/N$, and the irreducible residual is
$$ \lVert \delta^0 x^\star\rVert^2 = \sum_{e}\Big(\tfrac{\Theta}{N}\Big)^2 = \frac{\Theta^2}{N}. $$
The per-edge frustration $\Theta^2/N$ falls with refinement, which is why one could mistake it for a discretization artifact. But the total frustration energy is invariant:
$$ \boxed{\; N\,\lVert\delta^0 x^\star\rVert^2 \;=\; \Theta^2 \;=\;\Big(\textstyle\int_\Sigma \mathbf{B}_g\cdot d\boldsymbol\Sigma\Big)^{2}. \;} $$
This is the new result of the section: the irreducible incoherence of any clock network encircling a rotating mass is gauge-invariant and equals the squared gravitomagnetic flux it encloses. No synchronization convention, no choice of $N$, can remove it; it can only be spread thinner. In the coherence normalization of the framework, $\Gamma = 1 - \lVert\delta^0 x^\star\rVert^2/\lVert x^\star\rVert^2$, this gives the monotone bound
$$ \Gamma_{\text{loop}} \;\le\; 1 - \frac{\Theta^2}{N\,\lVert x^\star\rVert^2}, $$
so frame-dragging upper-bounds the achievable sheaf coherence of the network. As the loop is pushed toward the horizon, $\Theta \to \Theta_H$ set by $\Omega_H$, and $\Gamma_{\text{loop}}$ is driven to its floor. Coherence and frame-dragging are antagonists, and GW250114 fixed the exchange rate at the one place it is extremal.
Status flag. The invariant $N\lVert\delta^0 x^\star\rVert^2=\Theta^2$ is derived and solid. The step to $\Gamma_{\text{loop}}$ inherits the same normalization debt flagged for $\lVert\eta_d\rVert = 1-\Gamma(d)$ in DRK-170: it is exact only under the harmonicity normalization $\Gamma = 1-\lVert\delta^0 x^\star\rVert^2/\lVert x^\star\rVert^2$, and is a bound, not an identity, until $\lVert x^\star\rVert$ is gauge-fixed.
3. The horizon as a convergence obstruction
DRK-168 ended at the impossibility of measuring the one-way gauge. A horizon is where that impossibility becomes causal rather than conventional.
Model exterior reconstruction as an inference. Let $\sigma$ be the section carried by infalling matter, and let $R_T$ be the operator that takes the ringdown observed on $[0,T]$ and returns an estimate of $\sigma$. The late-time field at the observer is a sum of quasinormal modes,
$$ \psi(t) \;\sim\; \sum_n A_n\, e^{-\kappa_n t}\, e^{i\omega_n t}, \qquad \kappa_n \gtrsim \kappa = \tfrac{1}{2}\,\text{(surface gravity)}, $$
and this exponential decay rate $\kappa$ is exactly what GW250114 reports as the horizon-set damping. Decompose $\sigma = P_{H^0}\sigma \oplus P_{H^1}\sigma$ into the part the horizon re-radiates (soft-hair / harmonic) and the part it traps. Then:
$$ \lVert R_T\,P_{H^0}\sigma - P_{H^0}\sigma\rVert \;\le\; C\,e^{-\kappa T} \qquad\text{(geometric convergence)}, $$
so the $H^0$ part is computable to precision $\varepsilon$ in finite observer time $T = \kappa^{-1}\ln(C/\varepsilon)$. But
$$ \lim_{T\to\infty} R_T\,P_{H^1}\sigma \;\ne\; P_{H^1}\sigma, \qquad \lVert R_\infty\sigma - \sigma\rVert \;=\; \lVert P_{H^1}\sigma\rVert . $$
The obstructed part never converges. By the redshift, an infalling worldline reaches the horizon at finite proper time but infinite coordinate time; the exterior observer is forever in the pre-image's $T\to\infty$ tail. The interior residue $\lVert P_{H^1}\sigma\rVert$ is undecidable from outside in finite proper time — the causal analogue of a non-halting computation, the one-way membrane of DRK-168 taken to extremum. The langolier reading is exact here: what crosses the horizon is the past the exterior can no longer taste, the section eaten behind the boundary.
This also closes the loop with DRK-170. The unit $\eta_d : d \to FG(d)$ is the abductive guess at a pre-image; its defect is the un-invertible part of the restriction. The horizon is the physical extremization of that defect: $\lVert P_{H^1}\sigma\rVert$ is $\lVert\eta\rVert$ made gravitational, the guessed section that no amount of exterior data can confirm.
4. The information paradox in cost language
Now the sharpest claim, and the most falsifiable. Borrow the cost geometry of coherent inference (Li–Theil–Harrow–Chuang): reconstructing a $d$-dimensional signal through a measurement bottleneck costs $\Omega(d/\varepsilon)$ samples, while a coherent (measurement-free) channel costs only $O(1/\varepsilon)$. The linear-in-$d$ overhead is the price of premature projection — the cost of collapsing phase before the section is glued.
Apply it at the horizon. To resolve the $d$-th layer of infalling structure to precision $\varepsilon$, its outgoing amplitude must clear the redshift floor:
$$ e^{-\kappa\,t_d} \;\gtrsim\; \frac{\varepsilon}{d} \quad\Longrightarrow\quad t_d \;\lesssim\; \kappa^{-1}\ln\!\frac{d}{\varepsilon}, $$
and the signal-to-noise budget needed to hold that resolution scales as
$$ \text{SNR}_d \;\propto\; \frac{d}{\varepsilon}. $$
The exponent on $d$ is the diagnostic. If the horizon decoheres infalling information — a measurement, a firewall, a hard boundary — it forces the incoherent branch and the exterior pays $d/\varepsilon$: reconstruction cost grows linearly with mode depth. If the horizon is unitary — soft hair, a fuzzball surface, an information-preserving membrane — it permits the coherent branch and the cost is $1/\varepsilon$, independent of depth.
$$ \boxed{\; \text{firewall / decohering horizon} : \Omega(d/\varepsilon) \qquad \text{unitary / soft-hair horizon} : O(1/\varepsilon) \;} $$
So the information paradox, restated: does exterior reconstruction cost scale with depth, or not? The two horizons of the last fifty years of dispute are two points in the cost geometry of inference, separated by a single power of $d$. This reframing does not resolve the paradox, but it converts a metaphysical question ("is information lost?") into a scaling question about a measurable exponent — and ringdown spectroscopy of the kind GW250114 inaugurated is, in principle, where that exponent would be read.
5. Falsification (per DRK-131 protocol)
This post is wrong, in whole or in part, if any of the following hold.
The holonomy identity fails. If $\oint\omega_{\text{sync}}$ is not equal to the enclosed gravitomagnetic flux in the stationary weak-field limit — i.e. if the Stokes step in §1 is gauge-dependent in a way that survives the closed-loop integral — the central claim collapses to analogy. (This is the load-bearing step and the most secure; it is textbook gravitomagnetism re-read, not new physics.)
The invariant is normalization-dependent. If $N\lVert\delta^0 x^\star\rVert^2$ can be made to differ from $\Theta^2$ by a legitimate choice of restriction maps or vertex weights, §2's boxed result is false. The cycle-graph derivation says it cannot, but a counterexample on a non-uniform loop would kill it.
No depth-scaling exists, even in principle. If quasinormal reconstruction cost is provably independent of infalling mode depth for both unitary and non-unitary horizons, then §4's distinction is empty and the paradox does not live in the cost exponent.
The $H^0/H^1$ split is not causal. If soft-hair proposals are correct and all of $\sigma$ is eventually re-radiated, then $P_{H^1}\sigma = 0$, the "undecidable residue" of §3 is empty, and the horizon is a convergence delay, not a convergence obstruction. (Note: this is not a refutation of the framework but a determination of which branch §4 is in — the post is built to survive it, and indeed §4 makes that determination the whole content.)
The post commits to: §1 as established re-reading, §2 as a derived inequality with a flagged normalization debt, §3 as a model whose key claim is the causal non-convergence of the $H^1$ part, and §4 as a conjecture reframing the paradox as a scaling law.
Coda: the frame is dragged, the gauge is one-way
DRK-168 said the universe cannot be synchronized from outside because there is no outside. GW250114 has now shown us the place where that fact stops being a philosopher's caution and becomes a wall: the horizon, where the synchronization connection's holonomy saturates, where infalling clocks are dragged to the boundary's own rotation, and where their last light reaches us redshifted to a whisper that decays at the surface gravity. The frame is dragged; the gauge is one-way; and what falls past $\Omega_H$ is the section we can guess at but never glue.
The dragon counts the moves it can still make. The horizon is where the count runs out.
Khrug Engineering · draken.info · DOI 10.5281/zenodo.19273483 · ORCID 0009-0003-8049-7167 · CC BY-SA 4.0
Cross-references: DRK-168 (The One-Way Gauge), DRK-170 (The Guessed Section).