Strictness

Strictness is a per-rule setting, configured in a rule’s YAML frontmatter, that controls how strongly density scoring penalizes errors in the documents a check rule reviews.

Why it exists

Not every rule should penalize errors equally. A rule that catches banned phrases in polished marketing copy should be unforgiving; a rule that flags style preferences in early drafts should be more lenient. Strictness scales the penalty per rule without changing the rule’s criteria or the scoring formula.

How it works

Strictness is the multiplier in the density scoring formula:

rawScore = clamp(0, 100,  100 - density * strictness)

A higher strictness means a given error density causes a larger score drop. VectorLint accepts three named levels or any positive number:

Level	Value	Behavior
`lenient`	5	Forgives occasional errors; small penalty per issue
`standard`	10	Balanced - the default when strictness is unset
`strict`	20	Penalizes errors heavily; small issues still cost points

Any positive number is also valid, for fine-grained control between the named levels.

Key properties

Set strictness with the strictness field in the rule’s frontmatter. A check rule using the strict preset:

---
specVersion: 2.0.0
evaluator: base
type: check
id: AIPattern
name: AI Pattern
severity: warning
strictness: strict
---

A numeric value works the same way:

strictness: 15

When strictness is omitted, the rule scores at standard (10).

lenient forgives occasional errors - suited to draft-stage or low-stakes content.
standard is the balanced default for most rules.
strict penalizes errors heavily - suited to published, high-stakes content where even rare issues matter.

Constraints

Strictness applies only to density scoring - rules that count discrete issues. It has no effect on rubric scoring, which uses criterion weights instead.
Strictness is set per rule, in its frontmatter.

Density scoring - the formula strictness multiplies
Rubric scoring - unaffected by strictness
Rule frontmatter - where the strictness field lives
Handling false positives - strictness controls scoring severity; CONFIDENCE_THRESHOLD controls whether a candidate issue is surfaced at all. Different layers, different problems.

Overview

Get started

Configuration

Rules

Scoring

Guides

Reference

Why it exists

How it works

Key properties

Constraints

​Why it exists

​How it works

​Key properties

​Constraints

​Related

Why it exists

How it works

Key properties

Constraints

Related