How a figure is held to its evidence

The principle

A figure is a sentence on a page. Not a number a model remembers, not a number two sites agree on, not a number that looks right: the words a page uses to state it, found again on that page by code when the figure is written. Everything below follows from that, and the rules are the same wherever a figure is found, by Redback, by a person, or by a research tool run by the team.

The rules run cheapest first, and each refusal names the rule that refused, so a reviewer reads why and a later run can be measured against it.

flowchart TD
  figure["A figure, its quote<br/>and its page"] --> source{"Can the page be cited?"}
  source -->|"retailer, forum,<br/>aggregator, wiki"| no1["Refused"]:::fail
  source -->|"yes"| quote{"Is the quote on the page,<br/>with the number in it?"}
  quote -->|"no"| no2["Refused"]:::fail
  quote -->|"yes"| who{"Is the page about<br/>this exact gun?"}
  who -->|"another model"| no3["Refused"]:::fail
  who -->|"a variant, a chambering,<br/>an unclassified site"| doubt["Doubt: a person decides"]:::warn
  who -->|"yes"| fits{"Does it fit the record?"}
  fits -->|"no"| no4["Refused"]:::fail
  fits -->|"yes"| ok["Written as the page states it"]:::ok

The source can be cited

A retailer, a forum, an aggregator and a wiki are leads, never citations. Their figures are copied from somewhere, often from the wrong variant, and a wiki is followed to the source it cites instead. A site's front page proves the company exists, not a figure.

A site the source hierarchy has never classed is neither refused nor trusted: its figure is a doubt, for a person to confirm. A measured case: an airsoft news site's weight for one variant of a rifle passed as another variant's weight until this rule.

The quote is the page's

The quote must be on the page, compared after the differences that carry no meaning are removed (whitespace, curly quotes, soft hyphens, markdown a model added), and nothing fuzzier. It must be at least twelve characters, or a word and a number of at least eight (a spec table's Barrel: 5" is evidence; a bare 16 is not).

The number stated must be in the quote. A figure written as a word (a five-inch barrel) counts. When a researcher converted a unit (770 g off a page that says 27.2 oz), the page's own quantity that agrees with the claim is written instead, so the value always comes from the page's words. Pounds and ounces together are one quantity. A weight read beside the words with a magazine is refused for an empty weight, reading only up to the next label, so a table's next row does not count against the one before it.

The page is about this gun

A page on the Browning Hi-Power does not state the FN High Power's weight, however exact its quote. The words that name the model (its name and every alternate name, less the maker's words and words like pistol or arms) must appear at the head of the page, beside the quote or in it.

A designation of letters and digits (C14, X9, P229) names the gun on its own; a generation mark does not. A maker's initial before a number counts (G17C is the 17C). A title that names the model's words, with its number named elsewhere on the page, names it.

Some pages name the model and still describe another gun. Those are doubts rather than refusals: a variant word beside the product's name in the title (Compact, Full, Commander) that the record lacks; a chambering in the title the record is not made in (a .380 review for a 9 mm record); a barrel stated beside the quote that contradicts the record's own barrel, which is how a mid-size model's page is told from the full-size one whatever its title says.

The figure fits the record

The figure must be one a firearm can have, within wide bounds per field, and a barrel must be shorter than the gun. These catch a misread unit or a transposed figure, not a subtle error; the subtle ones are caught by identity, and by the family and cohort checks that run after a figure is written.

Doubts, and who settles them

A doubt is not a refusal. The figure is filed with the doubt first and its confidence capped, and a person decides. Where the doubt is about a chambering, a maker's own statement settles it: when a maker states a figure is the same across chamberings ("the same as the 9mm model"), that statement, once a person accepts it, confirms such a page for those fields only, and the maker's words are kept beside the figure as the reason.

Measured on the answer key (see Measuring it), the rules let no wrong figure through in 56 hidden figures researched by a strong model, and caught every wrong one Redback proposed but one, a genuine disagreement between a maker and a reference work.