Quality Preset (quality)

The quality preset evaluates general documentation, guides, READMEs, and technical articles for clarity, completeness, and practical utility.

---

Dimensions Evaluated

Dimension	Description	Threshold	Type	Role
clarity	How clear, well-structured, and easy to follow is this document?	\(\ge 0.60\)	Score	Warning (advisory)
tone	What is the primary stylistic tone of this document? (professional, informal, inconsistent)	N/A	Choice	Informational

[!NOTE] In the built-in quality preset, thresholds_as_warnings: true is configured. Clarity threshold regressions emit non-blocking warnings rather than causing immediate gate failures.

---

Usage Example

typesafe-eval docs/*.md --preset quality

Sample JSON response:

[
  {
    "schema_version": "1.0",
    "filepath": "docs/architecture.md",
    "filename": "architecture.md",
    "preset_name": "quality",
    "passed_thresholds": true,
    "composite_score": 0.85,
    "scores": {
      "clarity": {
        "score": 0.85,
        "max_score": 1.0,
        "normalized_score": 0.85,
        "confidence": 0.92,
        "probabilities": {
          "1": 0.02,
          "2": 0.13,
          "3": 0.85
        },
        "near_threshold": false
      }
    },
    "choices": {
      "tone": {
        "choice": "professional",
        "confidence": 0.95,
        "probabilities": {
          "professional": 0.95,
          "informal": 0.04,
          "inconsistent": 0.01
        }
      }
    },
    "violations": [],
    "warnings": []
  }
]

---

Near-Threshold Observations

When scores fall within \(\pm 0.10\) of the decision boundary (\(0.40 \le p \le 0.60\)), typesafe-eval annotates them with near_threshold: true. Autonomous agents and human reviewers should treat near-threshold scores as soft observations rather than hard failures, considering run-to-run variance (see Noise Calibration).