The vocabulary

38 words the corpus argues about, 33 of them used by more than one piece — and 4 places in 1,070 sentences where anybody says what one of them means.

The rule, and what it cost

A first version of the definition rule matched the pattern anywhere in a sentence and found 29. It also read “the decline of local news is a global phenomenon” as that piece defining local news as a phenomenon eroding democracies — which is not what the author wrote, and printing it beside another author’s definition would have misrepresented them both.

So the rule now requires the term to be the subject of its own sentence, and finds 4. 25 apparent definitions were given up to avoid putting words in people’s mouths. A looser rule would have produced a richer page and a worse one, and both numbers are published so you can judge the trade rather than take it.

Where two pieces say what the same word means

Two pieces saying a word means different things is not a contradiction and neither piece is wrong. It is where the industry has not agreed with itself — and it is invisible unless you read all twenty-one at once. Whose byline matters: the same author writing twice is a repetition, not a disagreement, and the gate checks that label against the bylines rather than trusting it.

Every term

A term is counted where it appears as a word and nothing more is claimed. The patterns are in data/claim-rules.json.

In the vocabulary and used by nobody: copyright, safety.