It started with a small detail in someone else’s digest: one piece on the history of Unix mentioned that the first Unix system clocks counted pulses from the power grid — 60 per second, exactly matching the frequency of an American wall outlet. A nice detail, but it hides a thought that people usually skip over: the outlet in your wall is not just a source of voltage. It is a distributed timekeeper, one that has been running for more than a century and still serves stoves, microwaves, alarm clocks, and radio receivers.
And this timekeeper has a feature any system administrator would envy: it does not merely tick, it knows how to catch up. If clocks fall behind, dispatchers can briefly speed up the entire continental grid, and the clocks will correct themselves. And if the failure comes from politics, a continent can collectively lose several minutes — and nobody will notice for months.
That is exactly what happened in 2018. And the protagonist of that story was a dispute between two Balkan administrations over who should pay for electricity in four towns.
Synchronous electric clocks are indecently simple. They have neither pendulums nor quartz crystals: a tiny motor turns in sync with alternating current, and gears count its revolutions. Fifty current oscillations make one second of “grid time”; for continental Europe, that is 4 320 000 cycles a day, which operators are required to supply exactly. No more, no less.
That is where the whole time-service architecture comes from. Frequency deviations from nominal are paid for in seconds: the grid runs ten thousandths of a hertz fast, and over a day all connected clocks gain nearly seventeen seconds. European operators compare the accumulated deviation with UTC atomic time and, if it exceeds twenty seconds, shift the target frequency: 49.990 Hz, to slow the clocks down, or 50.010 Hz, to bring them up to speed.
The twenty-second threshold is maintained out of caution. Shifting all continental generation by ten thousandths of a hertz is an enormous collective effort: Europe’s grid has a frequency response of around 27 000 megawatts per hertz, and an imbalance of just three gigawatts is enough to pull frequency down by more than a hundred millihertz. Every “clock adjustment” is a coordinated movement by hundreds of power stations across twenty-five countries.
The amusing part is that this entire service was born from an engineering failure. In 1912, Henry Warren founded a Massachusetts company making battery-powered clocks — and they lied outrageously, because batteries ran down. Warren invented a self-starting synchronous motor, patented it in 1918, and immediately discovered something unpleasant: 60 hertz in the grid was not equal to 60 hertz. His own clocks were gaining or losing ten to fifteen minutes a day.
So he did something that now looks absurd: he installed a reference clock with an ordinary pendulum at a Boston Edison power station, checked twice daily against signals from the U.S. Naval Observatory, and hung another one beside it — powered by a synchronous motor from the grid. As long as their hands moved together, grid frequency was considered correct. That is how the “master clock” was born — along with the unified frequency standard without which no large power system can be assembled. By 1947, Warren’s master clocks regulated more than 95% of electrical lines in the United States.
The irony here is fairly dense. A pendulum from the age of steam became the tool that allowed power engineers to synchronize generators and stitch a country into unified grids. Power systems themselves grew out of this synchronization as a side effect, and from them came electric clocks, which became a mass-market product precisely because there was no longer any need to wind a spring. Warren’s company, Telechron, sold millions of clocks and died in 1992: first batteries became long-lasting, then quartz arrived, and dependence on the outlet turned from an advantage into a flaw.
From the beginning, the time service was an agreement. It had thresholds, protocols, and people to blame.
In North America, thresholds differed between grids: Wikipedia lists ten seconds for the Eastern Interconnection, three for Texas, and two for the Western Interconnection, while metrologists at NIST write of ten, five, and “at the discretion of the operator” in Texas. Both sets of numbers circulate in documents from different years, and I am leaving the discrepancy as it is. Once the limit was exceeded, dispatchers set targets of 59.980 or 60.020 Hz — a shift at which clocks gain or lose roughly 1.2 seconds per hour.
Then, a few years later, something happened that nobody had planned: a procedure that had been a voluntary formality for thirty years became a rule carrying a penalty of up to 325 000 dollars for noncompliance. The number of people eager to remain “voluntary keepers of time” instantly shrank. Sources also disagree on the date when correction became mandatory in the United States: Wikipedia gives 2009, while NIST metrologists write that standard BAL-004 became enforceable in 2007.
Then the logic I see in every infrastructure kicked in: if a service can no longer be an obligation, it gets declared obsolete. In 2015, NERC published a policy paper proposing that time correction be removed from the reliability standard and moved into the category of commercial practice — and in the same document called the practice harmful: shifting the operating point by 20 millihertz brings the system closer to the automatic load-shedding setting, reducing the safety margin from 300 to 280 millihertz. The argument sounded technically convincing, and people swallowed it. On November 2, 2016, the NERC board voted to retire standard BAL-004-0; a week later, an application was filed with FERC, and on January 18, 2017, the regulator approved it.
Most revealing of all was who objected. The objections came not from power engineers but from metrologists. Specialists at the National Institute of Standards and Technology analyzed five years of measurements from the power supply to their observatory and calculated something simple: if corrections had not been made, clocks set in March 2016 for the switch to daylight saving time would by November have drifted by seven and a half minutes. Precisely because grid frequency is a live turbine under load, one that someone has to guide by the hand.
Europe made the opposite decision at exactly the same moment. European Union Regulation 2017/1485, known as the System Operation Guideline, established a “time control process” as a separate chapter: the control target is to keep average frequency at nominal, monitoring grid-time deviation is assigned to a designated synchronous-area monitor, and frequency correction is explicitly included among actions involving active reserves. While Washington dismantled the time service as a relic, Brussels turned it into an article of law.
This service has its own geography too, because frequency is a border no less real than a state boundary. In Japan, the eastern half of the country lives at 50 hertz and the western half at 60. The division emerged at the very end of the nineteenth century and has a prosaic explanation: Tokyo imported generators from Germany, Osaka from the United States, and the Fuji River became the border between the two worlds. While the grids were local, nobody cared about compatibility; once they grew together, changing anything became too expensive — the idea of unification was rejected in both 1939 and 1945. The grid still lives with it: three converter stations, with a combined capacity of around one million kilowatts, stand between the two halves.
After the 2011 earthquake, it became clear that the country physically could not transfer surplus power from west to east — there were catastrophically too few converters. The difference between decisions made by two companies at the dawn of electrification was, a century later, measured in megawatts unavailable in an emergency.
Now we can return to those six minutes.
Since the 1990s, four Serb-majority towns in northern Kosovo have remained that do not recognize the government in Pristina and do not pay for electricity. The debts piled up, and by an informal arrangement they were spread across the bills of Kosovo’s other consumers. At the end of 2017, a Pristina court ruled that this could no longer continue and removed the burden from consumers; the regulator ordered grid operator KOSTT to cut its tariff by 44% at once. The operator replied that it had neither the legal nor financial ability to cover the losses of the northern towns, and according to Reuters, one energy official put the position with absolute honesty: there were two options — cut off people, hospitals, and water utilities and get protests, or sit back and watch. They chose to watch.
Then the allocation of responsibility enters the picture. Under the grid-operation agreement, Kosovo belonged to a balancing area for whose stability Serbia was responsible — meaning Serbian operator EMS was required to maintain the balance of the Kosovo portion. It did not happen. From mid-January, average frequency across continental Europe began to slide down, and by early March ENTSO-E, the association of European system operators, issued a press release in a genre I have encountered nowhere else: a technical document with a political ultimatum. The deviation was “unprecedented for the Continental European system,” the undelivered energy amounted to 113 gigawatt-hours, and the question of who would compensate for the loss had to be resolved.
Average frequency over that period was 49.996 Hz. The difference from nominal was four thousandths of a hertz, or 0.008%. Over three months, that was enough for synchronous clocks across Europe — from Spain to Turkey, from Poland to the Netherlands — to fall behind by nearly six minutes. Ovens, microwaves, radio alarm clocks, and heating timers in twenty-five countries all lied in unison, and nobody would have noticed if somebody had not checked against their phone.
The parties, naturally, tell different versions. ENTSO-E speaks of a political dispute and an unfulfilled obligation. The Serbian operator explains the shortfall by saying that Kosovo had “continuously and unauthorizedly withdrawn” unpaid energy from the system. The Kosovar operator answers with a number: by its calculation, the dispute cost it €9.6 million in 2017 alone, because the Serbian side controlled the transmission capacity of its cross-border lines. Both sides signed the 2015 agreement regulating grid operations, but it never entered into force — the stumbling block was ownership of the grid itself, built back when Kosovo was a province of Serbia.
The price of the matter, incidentally, turned out to be ridiculous against the scale of the consequences: ENTSO-E put compensation at €4.5–5.6 million, while the Kosovo government ultimately allocated a million for the northern towns. Six minutes of time across twenty-five countries cost less than one good transformer.
The compensation was collected from the entire continent. Starting March 1, 2018, system operators in twenty-five countries held average frequency at 50.01 hertz — one hundredth above nominal — and spent a month “catching up” the lost time until the clocks returned to normal. Formally, it looks like a brilliant engineering operation. In practice, like a sentence passed on the idea itself.
Because the clocks that had fallen behind, and the clocks whose owners manually adjusted them in March, were now separated by the same six minutes. You can manage time in only one direction at once: either you lead everyone, or everyone goes their own way. The official recommendations came out in exactly that spirit — if you adjusted your own clock, sorry, adjust it a second time.
And the dispute did not go anywhere. In July 2018, the deviation returned, and ENTSO-E introduced an automatic compensation program that activates when grid time drifts by more than sixty seconds; on July 6, it triggered for the first time. By then, that meant the time service was already operating in emergency mode: with an automatic fuse instead of planned adjustment. Kosovo’s full exit from the Serbian control block came only in 2020, when KOSTT signed an agreement to join ENTSO-E — two years after Europeans spent six minutes looking for someone to blame.
And in North America, the time service died beautifully and by the rules. Standard BAL-004-0 was retired in 2017, and with it went the procedure for manual correction across the continental system. An interesting detail: in the western United States, Automatic Time Error Correction not only survived but continues to develop — in March 2025, NERC and WECC filed a joint petition with FERC to approve its fourth version. Formally, the explanation is simple: the Western grid performed this procedure automatically and received a separate regional standard that did not get caught in the general cleanup.
But the result is still curious. On one continent, two philosophies emerged: where correction had to be done by hand, it was deemed obsolete; where it happened on its own, it was kept.
Nobody has measured what that means for those stoves and alarm clocks. In a 2017 study, authors from the University of Tennessee and Oak Ridge Laboratory — who gathered grid-frequency statistics from more than two hundred sensors in North America and several dozen around the world — wrote plainly: nobody knows exactly how many devices in the world get their time from the outlet, and nobody knows what will happen when the service disappears for good. The same study shows the difference between “knows how to catch up” and “does not”: in Brazil, where no correction is performed, clocks gain up to eighty seconds of drift a month — simply because average frequency stays two millihertz above nominal. Nothing dramatic. The system just does not think it owes anyone anything.
The most interesting part of this story begins in the graveyard of the time service.
While European dispatchers were wondering who should pay for 113 gigawatt-hours, Romanian audio engineer Cătălin Grigoraș formulated, in the mid-2000s, a principle that turned digital forensics upside down: grid frequency fluctuates differently at every moment, and this curve never repeats. Any recorder plugged into an outlet — or merely standing beside one — captures, along with the voice, a quiet 50- or 60-hertz hum modulated by precisely that curve. Compare the recorded hum with a frequency archive for the relevant day, and you get a timestamp sewn into the recording itself.
London police began maintaining such an archive in 2005, Bavarian police in 2010, and they still do it today. The frequency standard ceremonially discarded as an unnecessary formality is now used as a trap for editing: a splice assembled from two recordings produces two different frequency fingerprints in one file, while sound and picture must match in phase — because the flicker of incandescent bulbs and fluorescent lighting carries the same information as the hum. A month of recordings can be checked in roughly ten minutes of machine time, where Watergate tapes took months of work.
The method’s limits are honest too. Because of constant shifts in a recorder’s own oscillator and the self-similarity of the frequency curve itself, a recording shorter than ten minutes can match several different dates — and then the report says not “recorded on May 3,” but “the probability of this under one hypothesis is eleven times higher than under another.” It works beautifully in court, where two versions are being compared, and badly in movies, where people want to hear an unambiguous answer.
It remains to return to the line where all this began — because it is no accidental guest in this story.
The first versions of Unix counted time as a 32-bit integer incremented by one 60 times per second — exactly according to that machine’s system clock, which in turn counted pulses from the grid. That was enough for a little over two years, so the starting point was moved several times: in one of the first editions of the manual, dated November 1971, time was defined as “the number of sixtieths of a second since 00:00:00, January 1, 1971.” The epoch was later redefined twice, the increment changed to one second, and eventually the date became 1970.
Dennis Ritchie explained the choice without any sacred meaning: there were several file systems, the starting points kept changing places, and at some point they simply decided to choose “something that wouldn’t overflow in the foreseeable future”; 1970 was as good as any other. Hence the oddity every programmer has seen at least once in logs: -1 second is both an error and the perfectly valid date of December 31, 1969, because the signed type remained signed.
Unix’s outlet-powered past left traces elsewhere too, with no nostalgia involved. Fifteen seconds, sixty sixtieths, twelve hundred cycles a minute — all of it comes from current frequency. Once it was the most accessible time standard in the world; then it became a technical detail; then forensic evidence; and then quartz, which costs pennies and asks nobody about politics, pushed it out everywhere. The one thing quartz did not inherit was the ability to catch up. Outlet clocks knew how to catch up: the continent did it for them, shifting frequency by one ten-thousandth of a hertz. Then the continent decided it no longer had to — and that is perhaps the most precise answer to what happened in 2018 to Europe’s six minutes. Nobody broke the clocks. Someone just stopped keeping them.