How we research

There is a specific failure mode in watch writing. A figure appears somewhere, gets quoted, gets quoted again, and after ten years it is "well known" without anyone ever having gone back to the document. Water resistance ratings, jewel counts and chronometer tolerances are all riddled with it.

Our method is designed to break that chain.

Who writes and who checks. Pages are drafted by the editorial desk and then checked, as a separate pass by a different function, by Standards Review, whose only job is to take each figure back to the document it came from and delete it if it will not go. A page that has been through that check says so in its byline. Nothing here is published because a draft sounded convincing.

Nobody pays to be in it. No brand, dealer or repairer pays for coverage here, nothing on this site is written in exchange for payment, and no page exists because a company wanted it to. No conclusion has been softened for a commercial relationship, because there are none to protect. Members of the desk take no gifts, loans or review units from brands, dealers or repairers, and where a writer owns a watch that is relevant to what a page concludes, the page says so. The full position is on the ownership and funding page.

The source hierarchy

We rank sources, and the ranking decides what gets published. It is a working hierarchy rather than a labeling scheme: you will not find little tier numbers scattered through the guides. What you will find is the source itself named in the text where it matters ("ISO 22810 requires", "the caliber's specification sheet gives") and the full list at the foot of every guide.

Anything below tier 6 is not a source. Retail copy, marketplace listings, forum posts and AI output are starting points for a question, never answers to it.

Where AI fits in this

It should be said in the open, on the page about method, rather than only in a policy document. We use large language models to assemble research, to produce first drafts and to edit. They are fast at gathering candidate material and confident at getting details wrong, which is why the whole method above exists.

So a model sits at the start of the process and never at the end. Nothing a model asserts reaches a published page until it has been traced to a tier 1 to tier 6 document, and a citation we cannot confirm exists is deleted rather than printed. A person decides that a page is fit to publish. The full rules are section 8 of our editorial standards, and if you ever catch a figure here that cannot be traced, that is a failure of this method and we want to hear about it.

What we do when sources disagree

Often enough, they do. Our rules, in order:

1. Prefer the more specific document. A caliber's own specification sheet beats a general brand page.
2. Prefer the more recent document, but check what changed. ISO 22810 superseded ISO 2281, and a claim sourced to the older standard has to be re-checked rather than assumed to carry over.
3. Separate the standard from the practice. What a standard requires and what the market actually does are two different facts, and we report both when they differ.
4. If it cannot be resolved, say so. We would rather write "sources differ on this, and here is the range" than manufacture a consensus.

What we test and what we do not

We are honest about this because it is the line most sites blur.

We do not run a laboratory. We do not pressure-test watches, we do not run watches on a certified timing bench for this site, and we do not conduct comparative product testing. When we describe what a timing machine shows or what a pressure test involves, we are describing a documented procedure, not reporting our own results. No guide here claims a measurement we made ourselves. If one ever does, it will name the instrument, the method and the conditions in the same breath, because a test you cannot audit is a claim wearing a lab coat.

What we do do is procedural verification: reading the standard, tracing each figure to a document, checking that a cited source says what it is claimed to say, and reproducing arithmetic (the tachymeter formula, the moonphase drift calculation, the 7-day rate averaging method) so a reader can follow it.

If that ever changes, we will say so on the page and describe the equipment and method, because a test you cannot audit is a claim, not a test.

How we handle numbers

Every numeric claim on this site goes through the same four questions:

1. Where is it from? A named document, edition and clause where possible.
2. What exactly does it apply to? COSC tolerances apply to an uncased movement of a specified size, not to your cased watch. That distinction changes the meaning of the number entirely.
3. Is it a requirement, a typical value or a manufacturer claim? These are three different kinds of fact and we label them differently.
4. What is the honest precision? We do not write "270 to 310 degrees" as though it were a legal limit. It is a typical healthy range, and we say so.

Units get the same treatment. We give the figure in the unit the source publishes, which in horology is almost always metric, and add the conversion a reader may be working in alongside it, rather than making you convert or quietly rounding a standard's figure into something it does not say. Prices are ranges rather than single figures, in US dollars and quoted before sales tax, which varies by state, with the caveat that any list price moves with market and discounting, and that outside the US local retail also moves with tax, import duty and distribution.

If a number fails any of those, it is cut. A page with a gap in it is better than a page with a confident invention in it.

How we handle history

Historical claims in horology are unusually contaminated by brand marketing, because "first" is a valuable word. We apply three tests: is there a document (a patent, a dated advertisement, a dated register entry)? Is the claim specific (first self-winding wristwatch is a different claim from first self-winding watch)? And does the claimant have a commercial interest in the claim?

Where those tests do not produce a clean answer, the page says the record is contested and gives the competing accounts, with the evidence for each.

How we handle safety

Some subjects here can cost a reader money or damage their health if handled badly: radium dials, quickset date mechanisms in the danger zone, water resistance, opening a case. On those, we state the risk before the instruction, we say plainly when a job belongs with a professional, and we never write a procedure so breezily that a reader would attempt something they should not.

Language and framing

• We define a term the first time we use it.
• We prefer the plain word to the trade word, and give the trade word alongside.
• We avoid superlatives about products entirely.
• We do not write in a way designed to make you feel behind, uninformed or excluded. That tone is a sales technique.

Reader input

Readers catch things. Watchmakers, in particular, catch things quickly. Every page invites a correction, every correction is read, and the substantive ones are logged publicly on the corrections page with a date and a description. That is not a courtesy, it is part of the method: a reference that no one can challenge is just an assertion with better typography.

Errors that survive all of this get corrected the same way: a dated public entry, a note on the page itself, and no quiet edit pretending the sentence was always right. How quickly a page comes back around for a fresh read is set out in the review cycles in our editorial standards, which run from every twelve months for the main guides to every twenty-four for glossary entries, and any page is re-read out of cycle when a standard it cites is revised.

Related reading: our editorial standards, the desks who write and check, and our ownership and funding statement.