Chronometer, COSC, and what accurate actually means
A certification measures a movement in a laboratory before it is cased. What happens on your wrist is a different question with a different answer.

At a glance
- COSC chronometer
- Minus 4 to plus 6 seconds a day, movement only
- Quartz, typical
- Plus or minus 15 seconds a month
- What regulation costs
- 60 to 150, and it is often all a watch needs
- The number that matters
- Consistency, not the average
A mechanical watch that gains five seconds a day is a good watch. It is also gaining two and a half minutes a month, which is more than a supermarket quartz watch drifts in a year. Both statements are true and the gap between them is the whole subject.
What the certifications measure
COSC, the Swiss official chronometer testing institute, tests bare movements for fifteen days in five positions at three temperatures. The criteria include an average daily rate between minus four and plus six seconds, a mean variation in rates under two seconds, and a maximum variation under five. Roughly six per cent of Swiss watch production is certified.
The critical detail is bare movement. The movement is then cased, which changes its behaviour, and the watch is shipped, worn, magnetised by a laptop speaker and knocked against a door handle. A chronometer certificate is a statement about a component in a laboratory, not a warranty about the object on your wrist.
Several manufacturers now test the cased watch, with tighter tolerances, which is a more meaningful claim precisely because it is measured on the finished article.
What actually causes drift
- Position. A balance wheel behaves differently lying flat and standing on edge, because gravity acts on it differently. A watch left dial up overnight will rate differently from one left crown down, and this is the easiest thing in the world to exploit.
- Temperature. Modern hairspring alloys are largely compensated, but extremes still move the rate.
- Magnetism. The single most common cause of a watch suddenly running minutes fast. A magnetised hairspring has its coils sticking together, shortening the effective length. Demagnetising takes thirty seconds and costs almost nothing.
- Power reserve state. Many movements rate differently at full wind and at the end of the reserve, which is one argument for wearing a watch daily or winding it fully.
- Wear on the pivots, which is the one that means a service.
If a mechanical watch suddenly gains minutes a day, it is magnetised until proven otherwise. It is not broken and it is not a service.
Regulating a watch to your life
A watchmaker can adjust the rate, and the useful instruction is not make it accurate but make it accurate for how I wear it. Someone who wears a watch twelve hours a day and leaves it crown up overnight has a predictable duty cycle, and a regulation that accounts for both positions will produce a watch that is far closer to correct across a week than one regulated to a laboratory average.
This costs between 60 and 150 and takes a few days, and it is the cheapest meaningful improvement available on any mechanical watch. A great many watches that owners believe need servicing simply need regulating.
The position trick, which is free
Find out how your watch rates in each resting position by leaving it overnight in one and checking against a reference in the morning. Most watches run faster in some positions and slower in others. If yours gains three seconds a day on the wrist and loses two overnight dial down, then storing it dial down every night removes most of the error, permanently, for nothing.
And the quartz comparison, honestly
A standard quartz watch holds about fifteen seconds a month, a thermocompensated one about ten seconds a year, and anything connected to a radio or satellite signal is correct to the second forever. No mechanical watch approaches any of them and none ever will, because the physics is not close.
Which means accuracy is not the reason to own a mechanical watch. The reason is that it is a machine, made by people, that runs on a spring and can be repaired indefinitely. Pretending otherwise sets up a comparison the object cannot win and does not need to enter.
Spotted an error? Tell the desk