Report format

shibbo check --format json prints one report per file (an array when there are several). The Node API returns the same object.

{
  "file": "README.md",
  "kind": "markdown",
  "words": 662,
  "score": 87,
  "band": "clean",
  "penalty": 4,
  "suppressed": 5,
  "findings": [
    {
      "id": "contrastive-negation",
      "name": "\"It's not X, it's Y\" / \"not just X, but Y\"",
      "severity": 3,
      "origin": "registry",
      "line": 12, "col": 5, "offset": 431,
      "match": "isn't a library, it's",
      "fix": "State the real claim. Keep a contrast only when someone holds the other view.",
      "counted": true
    }
  ],
  "judge": [{ "id": "triads", "question": "Are three-item lists used where one or two items would carry the point?" }]
}

Field	Meaning
score	0 to 100; see Scoring
band	clean 80+, light 60+, moderate 40+, heavy 20+, saturated
penalty	Sum of the weights of counted hits
suppressed	Hits inside shibbo-ignore blocks, not scored
findings[].origin	registry for curated rules, lexicon for measured phrases
findings[].counted	false when a density rule stayed under its rate
findings[].model, ratio	For measured phrases: the model that over-uses it, and by how much
judge	Questions to answer by reading, from tells no pattern can settle. For prose, the three from Say it clearly are always among them