Dating a firearm

Firearm makers are franker about the limits of their own records than any other trade on this site. Browning say pre-war dating is “strictly a guess”. Winchester say there is “no original, single, totally accurate database of serial numbers from 1866 forward”. Colt say their lookup is “approximate, and should not be relied upon”.

That candour is the useful part. This page collects who publishes what, what each of them warns you about it, and — for one maker — why their published figures cannot honestly be machine-read at all.

Browning — a cipher you can compute

Browning will date the gun for you from its serial number, on their own site, for nothing. From 1975 the date is a two-character code; before that it is a table specific to the model.

Neither the cipher nor the tables are reproduced here, and the decoder that used them has been withdrawn from the date codes page too. Browning's terms forbid commercial use of their content without written permission and this site is going to carry advertising, so what was checkable here is now checkable only on their page. Permission is being asked for. What this site found still stands and is worth saying: the cipher was verified identical across 32 of Browning's 53 model pages, and the pre-1975 material ran to 120 year rows across 40 separate model tables — a maker who documents their own history unusually well.

Browning attach the bluntest caveat of any maker here, on the Auto-5:

“Very few serial number records remain.”

“Exact production figures are not available. Year of manufacture on Pre-World War II production is strictly a guess.”

Winchester — published, and not machine-readable

Winchester publish a PDF, “Winchester Firearms Manufacturing Dates by Serial Number and Year, 1866 through the early 1990s”. It covers the Model 1866, 1873, 1876, 1885, 1886, 1887 and more, with the last serial assigned at the end of each calendar year.

Their own description of it is the first warning:

“The following pages were scanned from documents compiled over the years by the customer service department of Winchester Repeating Arms. None of the page scans are of original documents”

“No original, single, totally accurate database of serial numbers from 1866 forward” exists.

“Only serial number ranges with an approximate year are provided… No representation is made that all serial number and year combinations are totally accurate.”

And the scan is visibly damaged. It is a photocopy of a compilation, put through optical character recognition. The word FACTORY appears in three different broken forms on the same document:

This is why there is no Winchester decoder here. A corrupted word is obvious — nobody reads FACOTRY and thinks it is a real word. A corrupted digit is invisible. Three serials in the document are visibly broken:

Those three are caught because they stopped looking like numbers. A serial scanned as 63587 when the paper said 63537 looks perfectly normal and would be wrong forever. On a document where one word is mangled three different ways, transcribing several hundred serial numbers and presenting them as data would be inventing precision the source does not have.

So Winchester is described and linked, not decoded. Their PDF is the source; read it yourself, and treat every figure in it as the approximation Winchester say it is. For pre-1900 guns they point at the Cody Firearms Museum, and for anything unusual at their own customer service.

Colt — a lookup, and serials that collide

Colt run a serial lookup: you enter a number, they return model and year from their database. Nothing is published to compute against, so there is nothing to reimplement.

Their caveats are worth reading before you trust an answer from anyone:

“The information provided here is approximate, and should not be relied upon for legal, compliance, valuation, or other purposes”

“If your firearm does not appear in the listing, please understand that this is not a comprehensive database”

And Colt say their serials can collide. Certain vintage Colts share serial numbers across different models, so one number can return several firearms and the model is what separates them. That is not imprecision — it is the same number issued twice, which is exactly the situation the piano page meets with Kawai's overlapping tables. For a definitive answer Colt sell a Letter of Authenticity from their archive.

Ruger — their own history tool

Sturm, Ruger & Co. run a serial history lookup on their own site. Again a form into their records rather than published figures, and again nothing here to reimplement. Use theirs.

Who else was checked

The pool is 190 firearm makers, listed in firearm-makers.txt. Wikipedia has no category simply called “Firearm manufacturers” — checked — so it comes from their US and defunct-manufacturer categories plus named makers, with websites from Wikidata's official-website property rather than guessed.

And read the swept figure knowing how much was there to read. Of the 65 counted as swept, 25 offered this method fewer than 25 pages to look at, and 20 of those offered nothing from a sitemap at all. A host that answers, allows us, and then has almost nothing to search still counts as swept. It is the number of doors opened, not the number of rooms searched.

Of the 36 not probed: 20 did not answer, 0 disallow us in robots.txt, 5 returned 403 on robots.txt itself so their rules could not be read. Among those refusing: ArmaLite, Kimber, Mossberg, Palmetto State Armory, Rock River Arms and John Rigby & Company. Nothing was taken from any of them.

The sweep found the Winchester page but not the PDF behind it — there are no PDFs in Winchester's sitemap at all. That is the same blind spot that hid Trane's decoder on the HVAC page. Read 65 as swept, not as cleared.

Building the list, and the filter that finally works

A firearms maker list built from categories fills up with things that are not makers: cartridges (.204 Ruger), gun models (Ruger Red Label, Savage Model 24), and buildings (Colt Armory, Winchester Mystery House, two Historic Districts). Sixty-two such entries were in the raw pool.

They are removed by a positive test, not a blacklist. A company is filed under a category containing “companies”, “manufacturers”, “corporations”, “businesses” or “brands”; anything with none of those is dropped and reported. That matters because blacklisting categories is how the HVAC list nearly lost Sanyo and Toshiba to the word film — the eighth time on this site that a pattern matched a word sitting inside another word. A positive test cannot make that mistake, because it looks for the thing it wants rather than guessing at everything it does not.

It caught something immediately: the Wikipedia article Glock is about the pistol — its categories are all calibres — and the company is Glock GmbH. The filter dropped the pistol correctly, and the company went back in under its right title.