The anomaly that wasn't
In September 2011 the OPERA collaboration reported that neutrinos sent 730 km through the Earth from CERN to the Gran Sasso laboratory arrived roughly 60 nanoseconds early — as though they had outrun light. The claim was framed, correctly, as either an error or a revolution, and the collaboration did the honorable thing: they published the anomaly and asked the world to break it.
The world broke it. By early 2012 the 60 nanoseconds had been traced to two mundane faults in the timing chain, not the particle: a fiber-optic connector in the GPS synchronization link that was not fully seated, and a master clock oscillator running fast by a few parts per million. ICARUS, sharing the same baseline, measured the neutrinos arriving exactly on time. The neutrino had never been fast. The clock had been wrong.
This is the part everyone remembers as a story about scientific self-correction, and it is. But the location of the fault is the interesting thing. The error did not hide in the neutrino's flight. It hid in the act of deciding what "at the same time" means at two places 730 km apart. To see why that act is so fragile — why a loose connector can masquerade as a tear in relativity — we have to take seriously a fact that is usually mentioned and then dropped: the one-way speed of light cannot be measured.
The convention hidden in c
Every measurement we have ever made of the speed of light is, on inspection, a two-way measurement. A pulse leaves point A, reflects at point B, and returns to A. One clock, at A, times the round trip. The two-way speed is unambiguous and invariant:
$$ c_{\text{two-way}} = \frac{2\,L}{t_2 - t_1}. $$
But the one-way speed — A to B alone — requires a clock at B that already agrees with the clock at A about what time it is. And to make those two clocks agree, you must send a signal between them and assume how long it takes to arrive. You must already know a one-way speed to measure a one-way speed. The circle does not break.
Reichenbach made this precise in 1928. Assign the reflection event at B the time
$$ t_B = t_1 + \varepsilon\,(t_2 - t_1), \qquad \varepsilon \in (0, 1). $$
The standard Einstein choice is $\varepsilon = \tfrac{1}{2}$ — the assumption that light is isotropic, equally fast in both directions. But nothing in any experiment forces it. For any $\varepsilon$, the one-way speeds become
$$ c_{A \to B} = \frac{c}{2\varepsilon}, \qquad c_{B \to A} = \frac{c}{2(1 - \varepsilon)}, $$
and the only quantity the round-trip experiment ever constrains is their harmonic combination:
$$ \frac{1}{c_{A \to B}} + \frac{1}{c_{B \to A}} = \frac{2}{c}. $$
The admissible range is $c/2 \le c_{A \to B} \le \infty$: light may travel instantaneously one way provided it crawls at $c/2$ on the return. Every such world is empirically identical to ours. The isotropic universe and the anisotropic-but-self-consistent universe are the same world under two labelings.
So $\varepsilon$ is not a number waiting in nature to be read off. It is a gauge — a free convention we impose to make the bookkeeping consistent, the way a choice of zero on a thermometer is real engineering but not a discovered fact. The one-way direction of $c$ is the most fundamental example we have of a quantity that is constitutively unmeasurable: not hard to measure, but impossible in principle, because the measurement presupposes its own answer.
Synchronization is a connection
Here is where the abstraction earns its keep. A choice of $\varepsilon$ at every pair of points is a choice of what counts as "now" across all of space — a foliation of spacetime into simultaneity surfaces, a global section of the "time" coordinate. Carrying that definition of now from one place to another is not free transport; it is parallel transport under a connection.
Write the synchronization one-form induced by the spacetime metric as
$$ \omega_{\text{sync}} = \frac{g_{0i}}{g_{00}}\, dx^i. $$
To synchronize a chain of clocks A → B → C → … is to integrate this form along a path. The clocks agree at the end if and only if the integral is path-independent. And the question every global positioning system, every distributed timing chain, every "let us all agree on the time" protocol must answer is exactly:
$$ \text{Can the whole network be synchronized at once?} \quad\Longleftrightarrow\quad \oint_{\gamma} \omega_{\text{sync}} = 0 \ \text{ for every closed loop } \gamma. $$
That is a cohomological condition. Global synchronizability is the statement that the synchronization connection is flat — that its class vanishes in the first de Rham cohomology:
$$ \big[\omega_{\text{sync}}\big] = 0 \in H^1(\mathcal{M}). $$
When the class is trivial, you can lay a single consistent "now" over the entire manifold. When it is non-trivial, you cannot: transport your simultaneity surface around a loop and it comes back displaced from itself.
The obstruction lives in H¹
The displacement is not hypothetical. It has a name and a number. In any rotating or stationary-but-non-static spacetime the synchronization holonomy is non-zero, and the physical effect is the Sagnac effect:
$$ \Delta t_{\text{Sagnac}} = -\frac{1}{c^2}\oint \frac{g_{0i}}{g_{00}}\, dx^i \;=\; \frac{2\,\omega A}{c^2} \quad (\text{rotating frame, enclosed area } A). $$
Send two light pulses in opposite directions around a closed fiber loop on a spinning platform and they return at different times, by a gap proportional to the area enclosed and the rotation rate. The gap is the holonomy of $\omega_{\text{sync}}$ — the failure of "now" to close up. It is a non-trivial $H^1$ class made of glass and light.
This is not exotic apparatus. The Earth rotates. The GPS constellation — the very system whose timing chain OPERA was using — must correct for this holonomy continuously; without the Sagnac and relativistic corrections, GPS positions drift by kilometers per day. The protocol that tells the planet what time it is survives only by accounting, every second, for the fact that global simultaneity has a non-trivial $H^1$ obstruction.
Now the OPERA story snaps into focus. The chain was:
loose fiber connector → corrupted realization of the synchronization gauge → a 60 ns offset in the agreed-upon "now" → apparent faster-than-light neutrino.
The neutrino was never the carrier of the anomaly. The anomaly lived in a broken implementation of the simultaneity convention — a fault in the physical machinery that holds $\big[\omega_{\text{sync}}\big]$ pinned to zero. Finger that machinery with a glass connector that is one millimeter loose and the obstruction leaks — and what leaks out, read naively, looks like a particle outrunning light. OPERA was an H¹ error in disguise. The deepest objection raised in the original forum thread that seeded this post — that the relevant problem was clock synchronization, not the particle — pointed straight into the cohomology without yet having the operator's name.
The endogenous clock
There is a reason the one-way speed is unmeasurable, and it is not a limitation of our instruments. It is that there is no view from outside spacetime. To read both clocks "at once" and settle $\varepsilon$ empirically, you would need to stand at an Archimedean point external to the manifold and observe A and B simultaneously — but "simultaneously" is the very thing you are trying to define. The measuring apparatus sits inside the system it measures. The gauge must be chosen from within.
This is, in its cleanest physical form, the endogenous measurement problem — the same structure that runs through the whole Draken architecture and through the open correspondence on coherence-versus-soundness. Simultaneity is not a fact lying in the world waiting to be read; it is constituted by the observer's gauge choice. Reichenbach's $\varepsilon$ is the free parameter of an endogenous measurer who cannot step outside to check her own clock. Observers are not a convenience here; they are load-bearing. The anti-totalization principle has a physics prototype, and it is one millimeter wide: you cannot totalize "now" over the whole manifold without a choice that no measurement can adjudicate.
The feather of Ma'at returns as the same object. Judgment requires a defined zero on the scale; simultaneity requires a defined $\varepsilon$. Neither zero is discovered. Both are weighed against a calibration that the system imposes on itself. Remove the calibration and the verdict carries no information — every heart lighter than a weightless feather, every direction of light as fast as any other.
Two carriers, one cargo
A natural objection: but we have seen neutrinos arrive before light — from supernova SN 1987A, the neutrinos reached Earth some three hours ahead of the optical flash, across 168,000 light-years. Doesn't that vindicate the fast neutrino?
It vindicates the opposite, and the inversion is the lesson. The neutrinos did not travel faster; they were emitted earlier. They stream straight out of the collapsing stellar core at the instant of collapse, while the photons must wait hours for the shock wave to break through the star's outer envelope. Same event, two emission times, two coupling strengths. The three-hour spread over 168,000 light-years actually bounds the neutrino speed to equal $c$ to better than
$$ \frac{|v - c|}{c} \lesssim 10^{-9}. $$
SN 1987A is one of the strongest pieces of evidence against faster-than-light neutrinos, not a clue for them.
In carrier/cargo terms the structure is exact. The cargo is the event: the star collapsed. The neutrino and the photon are two carriers — two substrates transporting the same protocol-event across space, with different couplings and different emission moments. The naive inference "the neutrino was first, therefore it was faster" attributes to the cargo a property (superluminality) that in fact belongs to the carrier's emission mechanics. It reads a quirk of the vehicle's transit as a property of the truth-value of the message. That is precisely the failure mode the signalspaning taxonomy is built to catch — and here it is, written across 168,000 light-years instead of across a social feed. The protocol is substrate-invariant; the arrival times are not. The protocol is the agent; the particle is the substrate.
Coda: the thread that closed up
This post has a personal origin. In September 2011 a forum thread on the OPERA result ran for weeks; somewhere in it a Göteborg poster left a Schrödinger's-cat joke — if information could outrun the neutrino, the neutrino would have gone to the bar across the street instead, the question would never have been asked, and the bartender's whole future would branch into a version where the bar closes and a version where everything carries on as normal. In 2023 the same poster returned to the thread and wrote, under the joke, lol><.
The joke was a tachyon antitelephone. Faster-than-light signaling plus relativity yields closed causal loops, where self-consistency replaces chronology and the universe must select the global section whose holonomy closes. The punchline arriving before the setup is a miniature of retrocausality: the narrative protocol is invariant under time-reversal of its substrate event sequence. A small Draken model, built as a bar joke, holding up thirteen years later — and only legible now that the operator has a name. The thread, like the synchronization surface, came back around and almost closed.
Riktningen går inte att mäta. Den måste väljas.
Filed under: L01 · L03 · L05 · L11 · L12 · L17 — the measurement seam Operators: $H^1$ (the synchronization obstruction) · $K(t)$ (the chosen now) · the restriction morphism (label → boolean, against a calibrated zero) Cross-references: the Anubis / restriction-morphism sequence (the feather as defined zero) · the carrier/cargo taxonomy and Kuramoto-synchrony posts · The Sonder Egg (DRK-166) · the open Flygare Kinne correspondence on coherence vs. soundness and the endogenous measurement problem
ORCID: 0009-0003-8049-7167 · DOI: 10.5281/zenodo.19273483 (Draken thesis umbrella) · CC BY-SA 4.0