@greymoth/rules-core

A deterministic rule engine. No LLM, no network, no randomness. Define rules as plain data, run any subject through them, get back findings plus a 0–100 score where every point lost is attributable to a named rule.

Part of greymoth-devkit.

npm i @greymoth/rules-core

import { evaluate, noisyOr, failures, rule } from "@greymoth/rules-core";

const rules = [
  rule({ id: "license", title: "LICENSE present", severity: "error", weight: 3, test: (r) => r.hasLicense }),
  rule({ id: "tests",   title: "Has a test suite", severity: "error", weight: 3, test: (r) => r.hasTests }),
];

const r = evaluate(rules, { hasLicense: true, hasTests: false });
r.score;       // 50
failures(r);   // [{ ruleId: "tests", lost: 3, ... }]

API

• evaluate(rules, subject) → Report — linear budget scoring. A failing rule subtracts up to its weight; rules may return { passed, penaltyRatio } for partial credit. score = round((1 − lostWeight / maxWeight) × 100).
• noisyOr(rules, subject) → RiskReport — independent signals accumulate but saturate: risk = 1 − ∏(1 − weight_i) over failed rules (weights in [0,1]).
• failures(report) → Finding[] — failing findings, worst-first.
• rule(def) — identity helper for authoring a typed rule.

Deterministic and idempotent: identical inputs always produce identical output.

MIT © greymoth