Skip to main content
VectorLint reviews each document against the rules that apply to it and assigns the document a score on a ten-point scale. A score of 10 means the document fully meets the rule’s criteria. A score below 10 means VectorLint found issues, and it reports a violation at the rule’s configured severity.

​
Why it exists

The score is a shared, measurable quality signal. It makes quality comparable across documents and trackable over time, and it doubles as a CI gate: a score below 10 reports a violation, and a violation at error severity exits non-zero and fails the run. A numeric signal is easier to act on than a binary pass/fail.

​
How it works

VectorLint scores a document with one of two methods, chosen by what the rule asks for.
When a rule defines…VectorLint scores the document with…What the score reflects
Discrete issues to find (a banned phrase, a missing element, a structural error)Density scoringHow often issues appear relative to document length
Quality criteria to rate on a spectrum (clarity, tone, completeness)Rubric scoringA weighted average of LLM ratings against a rubric
Both methods report a score where 10 means the document fully meets the rule’s criteria and lower scores reflect more issues. Density scoring rewards documents where issues are rare relative to length; rubric scoring rewards documents that rate highly across the rubric’s criteria. Density scoring is tunable per rule through strictness, a multiplier on error density. Rubric scoring is tuned through criterion weights in the rule’s frontmatter. The formulas and worked examples live on the density scoring and rubric scoring pages.
  • How it works: the full review pipeline, from file discovery through scoring and reporting
  • Strictness: control how strongly density scoring penalizes errors