Every mechanical watch gains or loses time. A few hundred parts, driven by a wound spring, are being asked to divide a day evenly, and they manage it to within a few seconds. Drift is ordinary. Doing it to within seconds is not.
The number that describes this is the rate, given in seconds per day and usually written spd. A watch that gains four seconds a day has a rate of +4. One that loses two has a rate of −2. Plus means running fast, minus means running slow.
What “fast” and “slow” are actually measuring
A rate is not how wrong your watch is right now. It is how much wrongness it adds each day.
The difference matters. A watch you set a month ago and never touched might be four minutes ahead, which sounds alarming. Spread across thirty days it is eight seconds a day, which for a great many mechanical watches is unremarkable. A single glance at a dial cannot tell you which, because that needs two points in time and the gap between them.
The ordinary reasons
Most of the time a rate is what it is for reasons that are not faults at all.
How you leave it overnight. Gravity pulls on the balance wheel differently depending on which way the watch is lying. Dial up and dial down are not the same as resting on an edge with the crown up or down, and on many watches the spread between positions is wider than the rate itself. A watch that spends every night in the same position has that position setting much of its daily average. It is also the one adjustment you can make without anyone opening anything. Leave a watch that gains in whichever position it runs slowest, and the two pull against each other.
How much you wear it. An automatic watch winds from your wrist, so worn daily it sits near a full mainspring. Left in a drawer it runs down, and with less power behind it the balance swings through a smaller arc. The rate shifts when that happens. Which way it shifts is particular to the watch, and the two directions are both common enough that no rule of thumb will tell you which yours does. A rate measured across a weekend the watch spent in a drawer is measuring the drawer as much as the movement.
Temperature, and the life it leads. Modern hairsprings are far less temperature-sensitive than older ones, but a watch worn against skin all day and left on a cold windowsill all night is living through a range. A week of unusually active wear will move the number too.
None of these are problems. They are why the same watch honestly produces different numbers in different weeks, and why one reading is an anecdote rather than a measurement.
The less ordinary reasons
Some causes are worth mentioning to somebody who works on watches.
Magnetism, or a knock. A watch that suddenly gains a great deal, far beyond its usual few seconds and sometimes minutes a day, has usually had one of two things happen to it. From the outside they look the same.
The first is magnetism, picked up from something mundane like a laptop, a speaker, a cabinet catch or a phone case. Magnetised, the coils of the hairspring cling to one another instead of breathing freely. The spring is effectively shorter, the balance swings faster, and the watch races.
The second is mechanical. Shock protection does a great deal, but a hard knock can still leave a coil of that same hairspring caught on the regulating pins, or touching itself. That shortens it just as effectively, and the watch gains just as hard.
Time. The oils inside a mechanical watch are doing real work and do not last forever. As a service comes due some watches drift, some become erratic, and some carry on apparently unbothered. Manufacturers’ own recommended intervals run from about three years to about ten. That spread is honest. There is no single correct interval, only the one that suits a particular watch and the life it leads.
What a number needs before it means anything
A rate is a comparison between two moments, so it inherits whatever was sloppy about either.
Two things separate a number from a guess. The first is time between readings. Twelve hours is about the minimum before a daily rate means much, a day or two is better, and for a quartz watch, whose drift is far smaller, a week or two reads truer still. Read too close together and any small error in either reading gets divided by a very small number of days and comes out enormous.
The second is a reference you can trust. Your phone’s displayed clock is not quite it. Phones set themselves from a variety of sources, and the second on screen is not necessarily the true one. Network time, the same sort of reference computers use to agree with each other, is good to about a hundredth of a second, which against a watch measured in seconds a day is effectively exact.
Ratekeeper does both of these for you, but the arithmetic is not a secret, and there is a calculator here if you would rather not do the division yourself.
What your rate cannot tell you
A daily rate is one measurement of one property. Watches are commonly held against reference bands such as −4/+6 seconds a day, or the tighter, gaining-only 0/+5. Those are the strict end of the scale, and a great many perfectly good watches are specified far wider than either. What a band describes is exactly one thing, which is how much a watch gained or lost per day across the readings you actually took.
A band is a reference line, nothing more. Ratekeeper measures daily rate and only daily rate; how a watch holds up against magnets, water or a knock is outside what it can see.
A watch sitting comfortably inside a band can still be overdue for attention. A watch sitting outside one can be in fine health, regulated to a different taste, or measured across an unrepresentative week.
What a rate is genuinely good for is noticing change. A watch that has run at +3 for a year and is suddenly at +40 has told you something worth acting on. That does not say what is wrong. It says something changed, which is exactly the kind of observation a watchmaker can use.
Where to start
Take two readings a day or two apart and see what falls out. Most people are surprised in one direction or the other, and most watches turn out to be more consistent than their owners expect.
The support pages cover how readings group into runs, and why the current rate and a run’s average answer different questions. The rest of the guides go further into each of the causes above.
Ratekeeper measures all of this from a five-second tap against network time, groups the readings into runs, and breaks the rate down by the position the watch rested in. It is free for your first watch.