functype-eval

Fitness scoring for functional TypeScript codebases using the functype ecosystem.

functype-eval measures how well a TypeScript codebase adheres to functype / functional-programming idioms and produces a 0–100 fitness score with a per-dimension breakdown. It runs eslint-plugin-functype programmatically (no shelling out), plus type-coverage and a non-null-assertion scan, and aggregates the results.

Install

pnpm add -D functype-eval
# or run directly
npx functype-eval score ./src

Usage

functype-eval score <target>            # score a directory, print breakdown + 0–100
functype-eval score ./src --json        # machine-readable output (for CI)
functype-eval score ./src --threshold 80   # exit 1 if score < 80 (CI gate)
functype-eval score ./src --project tsconfig.json   # explicit tsconfig for type-coverage

Exit codes: 0 scored ok (and at/above --threshold if given), 1 score below --threshold, 2 no TypeScript sources found under the target (nothing to score — guards against a misleading 100/100 on an empty or non-TS directory).

Example

functype-eval score ./src

  Dimension       Weight   Score
  ─────────────────────────────────
  immutability     0.15     0.94
  option           0.15     0.88
  either           0.15     0.91
  composition      0.10     0.97
  collections      0.10     1.00
  do-notation      0.05     1.00
  safety           0.10     1.00
  loops            0.05     0.96
  type-coverage    0.10     0.99
  non-null         0.05     1.00
  ─────────────────────────────────
  Fitness score:  94 / 100

How the score works

The composite is a weighted sum of ten dimensions (see CLAUDE.md for the full table and weights). The eight ESLint dimensions and the non-null dimension are normalized by violation density per 1000 lines of code:

score = 1 / (1 + (violations / KLOC) * k)

type-coverage contributes its native percentage. If the target has no resolvable tsconfig.json, the type-coverage dimension is skipped and the remaining weights are renormalized — you still get a score. Weights and per-dimension sensitivity are overridable via an optional functype-eval.config.json.

The score is deterministic: same input → same output. No network, no LLM calls.

Roadmap

• Phase 1 (current): the score CLI.
• Phase 2: functype-eval bench — run programming tasks through LLMs and score the output with this same engine.
• Phase 3: an MCP server exposing score-codebase / suggest-functype-fix.

License

MIT © Jordan Burke