Define before use

Claim: every term and symbol is defined at or before its first use, premises come in the order they are needed, and each thing keeps one name.

• Grade: supported with caveat for a definition before its use. Premises in the order they are needed: measured for language models, argued from theory for people. One name per thing: argued from theory for people, supported with caveat for language models.
• Evidence: psycholinguistics (strand 4), learning science (strand 3), LLM readers.
• Check: defined-before-use (heuristic): a bold term with a definition cue, used earlier in the same file.
• Reader styles: strengthened by novice, visual-relational and LLM reader; relaxed by expert.

What it asks

• The term in bold where it is defined, with its definition on the same line; afterwards, the same word every time.
• A symbol is named in words in the sentence before the formula that uses it.
• A term defined in another file is linked at its first use here.

Why

• A reader holds a few propositions at a time and links each new sentence to them by a shared argument. When nothing in the buffer matches, they search memory, which is slow (Kintsch & van Dijk 1978).
• Revising a text where a comprehension model found such gaps improved learning, most for readers short of skill or knowledge (Britton & Gülgöz 1991; Gülgöz et al. 1999, secondary).
• Learning the names and parts before the lesson helped transfer in all three experiments and retention in two (Mayer 2002).
• Language models reason better when premises come in the order they are used (see LLM readers).
• One name per thing keeps the shared argument that model links by: a second name leaves the reader to infer that the two match (Kintsch 1978, a model, untested for names). Retrieval in language models leans on literal overlap, which a synonym breaks (Modarressi 2025).

Limits

Some high-knowledge readers learn more from a text that leaves links to infer (McNamara et al. 1996b). The skilled ones do not (O'Reilly & McNamara 2007).

A term the reader already knows needs no definition. Across 60 studies, assistance helped novices and hurt advanced learners (Tetzlaff 2025, a meta-analysis, measured; Kalyuga 2003 is an earlier review of the effect). The studies are mostly in science and mathematics, the evidence is less clear for humanities and language (10 of the 60 studies), and the authors advise giving assistance when in doubt.

Closer to definitions, explanatory notes built into a text helped readers with low language competence and slowed more knowledgeable ones (one study, reviewed in Kalyuga 2003). So with an unknown reader, define.

Practice agrees: Google's style guide exempts well-known terms such as API (Google 2026b, convention). Which terms are known depends on the reader, so the reader style decides.

A tutorial that withholds explanation, as the Diátaxis documentation framework advises, links the explanation instead (Procida 2026, convention).