Why we publish our data quality

Because trust in data is earned by showing how it is kept right, not by saying it is. Firearm specifications are copied from site to site, errors travel with them, and a figure repeated in ten places is not ten confirmations. Publishing the task board and the work that closed it lets you check any figure you rely on: what raised a question about it, which page answered it, and in whose words. A catalogue that hid its open questions would have the same questions; you would just not know which figures they touched.

No. The checks are tuned to flag anything that deserves a second look, across 8,651 firearms, 944 manufacturers and 563 cartridges, so they flag generously on purpose. A task is a question, not a verdict: it can close with the record confirmed exactly as it stands. The size of the list reflects how closely the catalogue is watched. See the overview for where each check stands today.

Every figure that changes carries its evidence: the page it came from, the words on that page, and who accepted it. You can open any task and read the finding, the source, the reasoning and the result, and the Methods pages show how each check calculates, with the figures it was fitted to. Nothing is changed on the strength of a model's say so, and nothing is hidden when a record turns out to be right.

Our checks do, automatically. They read every record against its own sources, the physics of what it describes and the records around it, and open a task the moment a record fails one, with nobody choosing which records to look at. The only tasks a person raises are reports from readers like you.

A page someone can open and read the figure on, quoted word for word. Sources are ranked: a maker's own page above a standards body, above a reference work, above a retailer, and a site that copies others ranks below all of them. A search result or an encyclopaedia summary is a lead, never a citation. The sources and quotes page sets out the ranking and why.

Yes. Makers publish their specifications as PDFs more often than as web pages, so the research reads them as documents and rebuilds their tables from where each word sits on the page, so a figure keeps the label it belongs to. A scanned PDF with no text is reported as unreadable, never guessed from. Reading a PDF explains how.

Photographs carry the source and licence they came with, and a check flags any that do not. The line drawings are our own: produced for GunSpec with the help of AI, from reference photographs of each firearm, and owned by us, so you can build on them without inheriting somebody else's rights. The media guide covers what each record carries.

This. The catalogue is not a finished file we sell copies of; it is kept right continuously, and that takes research time, the compute that runs the checks and the research agent, the sources it reads, and the drawings we produce. Your subscription is what pays for that work, and publishing it is how you can see what you are paying for: a catalogue that is checked, corrected on evidence, and honest about what it does not yet know.

Because firearm data is a problem nobody has solved. There are thousands of makers, specifications published in every format and unit, and the same errors copied across hundreds of sites. No team reads all of that by hand. AI lets the research read at that scale: it finds the maker's page, reads the spec sheet and quotes the figure. It does not decide anything on its own: every quote is checked in code against the page it came from, and a person accepts the result. We spend heavily on this because consistency across the whole catalogue is the thing that is hard, and the thing you are buying.

Thousands of hours of searching makers' sites, manuals, archives and standards for the right figure, the right source and the right image, before any automation, and alongside it now. The checks and Redback did not replace that work; they let it cover the whole catalogue on every run instead of one record at a time.

Report it, citing the page that shows the right figure. Every report is read against the record and your source, and a correction that holds up reaches everyone who uses the data. Data reports explains how, and what happens to a report once it is filed.

Yes. Every task has a page of its own on the task board, with what was found, the sources, the reasoning and when each step happened. Closed work is on completed work, newest first.