For all of human history, a second was a slice of a day — the Earth's own spin, divided. Then in 1967 we handed the job to an atom, and the planet started drifting away from its own clock. Now the Earth is running fast, and timekeepers are preparing to do something they have never done: take a second away.
Four centuries of precision, compressed. The scale below is logarithmic — each step is a different order of accuracy, not a different amount.
The 2024 strontium clock's error budget is so small that it would take roughly three times the current age of the universe to drift by a single second. That figure is derived from the published uncertainty, not a measured drift.
Fractional frequency uncertainty — how much of itself a clock can be wrong by. Light to dark is worse to better, across seventeen orders of magnitude.
Optical clocks now beat the caesium standard that defines the second by roughly two orders of magnitude — which is precisely why the definition is being rewritten. Era anchors are rounded to the nearest order of magnitude.
Atomic time runs at a constant rate; the Earth does not. Since 1972, leap seconds have been inserted into UTC to keep the two within 0.9 s of each other — 27 of them. And then, in December 2016, the staircase stopped.
All positive. The last was inserted at the end of 31 December 2016.
Ten seconds of head start in 1972, plus every leap second since.
The 2022 General Conference voted to stop inserting them by 2035.
Tides have been braking the Earth for billions of years. But since around 2020, the planet has repeatedly broken its own record for the shortest day of the atomic era — a short-term speed-up nobody fully explains.
Milliseconds below the nominal 86,400-second day, from IERS Earth-orientation measurements. A millisecond a day sounds trivial — but it accumulates, and it is measured against clocks that would take 40 billion years to lose a second.
The Earth's day has lengthened at this average rate since 720 BC, measured from ancient eclipse records — slower than the +2.3 ms/century that tidal friction alone predicts. The Moon retreats 3.83 cm a year, and takes some of our spin with it.
For about a billion years, roughly 2 to 1 billion years ago, day length appears to have stalled near 19 hours — solar atmospheric tides pushing the planet forward almost exactly as hard as lunar ocean tides held it back.
Timing is infrastructure. Power grids, mobile networks, stock exchanges and every satellite fix on Earth are all downstream of a small number of atomic clocks.
Orbiting clocks gain 45 µs/day from weaker gravity and lose 7 µs/day from their speed. Left uncorrected, position error would grow about 10 km per day.
Under MiFID II RTS 25, high-frequency trading clocks must stay within 100 microseconds of UTC, timestamped to 1 µs.
Estimated UK economic loss from a satellite navigation and timing outage — £5.2bn over five days. The authors call it a lower bound.
The next decade rewrites both ends of the problem: what a second is, and how we reconcile it with a planet that will not keep time.
If the current speed-up holds, UTC may need its first subtracted second around 2029 — projected, not scheduled. Polar ice melt, by redistributing mass toward the equator, has slowed the Earth just enough to push that date back about three years.
The preferred scenario in the international roadmap puts a new, optical definition of the second before the General Conference in 2030 — retiring caesium after 63 years.
From 2035, UTC is allowed to drift from the Earth's rotation by more than a second. The proposal is to leave it alone for at least a century.
The leap second exists because two definitions of "day" disagree, and for fifty years we papered over the gap by hand — one irregular, announced-six-months-ahead second at a time. That worked when time was a broadcast. It breaks when time is a global network of machines that assume every minute has sixty seconds.
A second used to be something the planet gave us. Now it is something we keep, and lend back.
Notes on figures. "~40 billion years" for the 2024 strontium clock is derived from its 8.1×10⁻¹⁹ fractional uncertainty, not an observed drift. The 1650s pendulum figure (~15 s/day, hence ~1.6 hours to a one-second error) is an approximate historical performance estimate, not a measured standard. Spectrum-strip era anchors are rounded to the nearest order of magnitude. The 2029 negative leap second is a projection from current Earth-rotation trends; no such adjustment has been scheduled.
Like this? Get a new data story in your inbox every morning.