Names predict content
Claim: a reader can predict what a file, folder or heading holds from its name before opening it, and each thing has one name throughout.
- Grade: supported with caveat for names that predict content; one name per thing argued from theory for people and supported with caveat for language models.
- Evidence: file structure, against (names), LLM readers.
- Check:
name-matches-title(heuristic): the name's words appear in the title or opening;name-content(meaning): name and opening close in meaning;bibliography-keys(exact in one file, heuristic across repositories): one key, one work. - Reader styles: strengthened by expert and LLM reader.
What it asks
- Full words for anything seen from far away: files, folders, headings, public names.
- Short names are fine in a small scope, and so are abbreviations the reader already knows (URL, HTML, README): a name's length follows its scope.
- One name per thing. A synonym used for variety reads as a second thing.
- The field's own term, which is what a reader types into search.
Why
Readers navigate by information scent, the cues a label gives about what lies behind it (Pirolli 1999). Full words gave the best comprehension of code (Lawrie 2006). Long-context retrieval in language models leans on literal word overlap between question and text (see LLM readers), so a thing with two names is found under one of them only. For people, one name per thing is argued from a model of comprehension, in which each sentence links to the last by a shared word (Kintsch 1978); no study tests it for names.
Limits
Full words were often not significantly better than well-known abbreviations (Lawrie 2007), and longer names took longer to process and lowered recall (Binkley 2009): good naming also limits length.