Methodology

What a decoder rule must contain

A serial format only ships once it carries the brand and family it applies to, a name, a pattern, the earliest and latest dates it was used, the logic for extracting a date and where possible a factory and country, a confidence level, its known exceptions, the explanation shown to you, at least one source reference, the date a human last reviewed it, and both positive and negative test examples.

The tests are not optional. Our test suite iterates every rule in the catalogue and fails the build if any of that is missing, which means an undocumented format cannot reach production even by accident. There are currently 20 rules, and every one of them is covered.

How sources are weighed

Manufacturer documentation is preferred. Where a manufacturer has not published the information — which is common, and near-universal for factory codes — we use credible specialist or community references and label them as such in every result. You should weigh a community-maintained factory-code table differently from a manufacturer’s own dating page, so we never present them as equivalent.

What we will not do

Six things are out of bounds regardless of how much they would improve coverage:

• Invent a serial-number format.
• Invent a factory code.
• Invent a manufacturing date, including by extrapolating past the end of a published table.
• Infer an exact model from a serial.
• Conclude that an instrument is authentic.
• Estimate a value.

Where a rule cannot produce a precise answer, it returns a range or several candidates. That is not the tool hedging — it is the tool reporting what the evidence actually supports.

How your input is handled

Before matching, we trim the serial, convert letters to uppercase, remove spaces and separators, and strip a pasted label such as “S/N:”. Your original input is preserved and shown back to you alongside a note of anything that changed.

We do not silently correct characters. If you type a letter O where the instrument has a zero, we will tell you the serial contains a letter O rather than quietly substituting a digit and returning a confident result for a number you did not enter.

How matching works

Every applicable rule is tried, not merely the first one that matches. Extracted day, month and year values are validated against the calendar, which rejects impossible dates outright — day 366 in a common year, month 13, a production day of 000. All surviving candidates are scored and returned in order.

Some patterns are too loose to run safely without a manufacturer. A bare five-digit number is a plausible Gibson Custom reissue serial, but searching every brand for five-digit numbers would return noise, so those rules run only once you have named the brand.

How confidence is scored

Each candidate starts from a base determined by how well documented its rule is, then gains or loses points for each observable signal. Every factor that applied is recorded on the result, so you can see exactly why something scored as it did.

The total is mapped onto four categories: high, moderate, low, and more information required. You will never see a percentage. A number like 92.37% implies a precision that a pattern match against a published table does not have, and presenting one would be a false claim dressed up as arithmetic.

Follow-up questions

Questions appear only when more than one reading survives and answering would actually separate them. Asking “what country is it marked?” when every candidate says Mexico wastes your time, so we check first.

Answers can raise confidence, lower it, or remove a candidate entirely. Telling us an instrument is not a Custom Shop model does not merely make the Custom Shop reading less likely — it rules the format out.

Why you sometimes get several answers

Because several manufacturers reused their numbering. Gibson Custom Shop serials record one digit of the year, so the same pattern recurs every decade. Fender ran the alphabet twice on its Japanese instruments. Martin publishes year-end totals, so a serial near a boundary genuinely could belong to either year.

In each of those cases there is more than one correct reading, and choosing one for you would be inventing certainty. We show them all, ordered by confidence, with what would settle it.

Review cycle

Each rule carries the date a human last checked its sources and confirmed the information it depends on is still there. Those dates are shown on every result and on each manufacturer page, so you can see how current the reasoning behind an answer is rather than assuming it.