Recipe catalog / policy-compliance

Check an expense against a written policy

Does expense, as described, comply with the written policy?

You need a yes/no check on whether a described expense follows the rules in an expense policy before it is approved, flagged, or sent back to the submitter.

Explore this recipe interactively · Source and implementation guide

Use policy-compliance in TypeScript

Install with npm install jev-recipes. Requires Node.js 22.9 or newer and ES modules. Set TYPESAFE_API_KEY in your server environment for live calls, which send input to TypeSafe and use API quota. See the installation guide.

import { policyCompliance } from 'jev-recipes/policy-compliance';

const result = await policyCompliance({
  "expense": "Team dinner on Thursday for 4 engineers at a steakhouse after the release. Included two bottles of wine. Total $412, paid on my personal card, receipt attached. No client was present.",
  "policy": "Meals: Team meals are reimbursable when approved in advance by the manager and when they mark a project milestone. Alcohol is reimbursable only at client-facing events. Meals over $75 per person require a manager's written approval before submission. Receipts are required for any expense over $25.",
  "minConfidence": 0.8
});
console.log(result);

Input contract

{
  "input": {
    "$schema": "https://json-schema.org/draft/2020-12/schema",
    "type": "object",
    "properties": {
      "expense": {
        "type": "string"
      },
      "policy": {
        "type": "string"
      },
      "minConfidence": {
        "type": "number",
        "minimum": 0,
        "maximum": 1
      }
    },
    "required": [
      "expense",
      "policy"
    ]
  },
  "result": {
    "$schema": "https://json-schema.org/draft/2020-12/schema",
    "type": "object",
    "properties": {
      "model": {
        "type": "string"
      },
      "usage": {
        "type": "object",
        "properties": {
          "input_tokens": {
            "type": "integer",
            "minimum": 0,
            "maximum": 9007199254740991
          },
          "output_tokens": {
            "type": "integer",
            "minimum": 0,
            "maximum": 9007199254740991
          }
        },
        "required": [
          "input_tokens",
          "output_tokens"
        ],
        "additionalProperties": false
      },
      "status": {
        "type": "string",
        "enum": [
          "ready",
          "review"
        ]
      },
      "probability": {
        "type": "number",
        "minimum": 0,
        "maximum": 1
      },
      "confidence": {
        "type": "number",
        "minimum": 0,
        "maximum": 1
      },
      "verdict": {
        "type": "string",
        "enum": [
          "compliant",
          "noncompliant"
        ]
      }
    },
    "required": [
      "model",
      "usage",
      "status",
      "probability",
      "confidence",
      "verdict"
    ],
    "additionalProperties": false
  }
}

Saved example result

This hand-authored response demonstrates the contract. It is not a model accuracy measurement. Run it without an API key: npx jev-recipes demo policy-compliance.

{
  "model": "demo-fixture",
  "usage": {
    "input_tokens": 0,
    "output_tokens": 0
  },
  "status": "ready",
  "probability": 0.09,
  "confidence": 0.91,
  "verdict": "noncompliant"
}

Evaluation evidence

Use the evaluation guide to measure this decision on your own labeled cases.

Limitations

• Judges the described facts against the written rules. Amounts, per-diem limits, dates, and receipt thresholds must be compared in code; the recipe treats stated numbers as facts to match against the text, not as arithmetic to perform.
• A rule the policy does not mention is treated as permitting the expense. The recipe does not know the organization's unwritten norms or approval history.
• Compliance with the wording is not an approval decision or an enforcement outcome.

Related recipes

• policy-severity: Use policy-severity to grade how serious a known violation is, rather than to decide whether the expense violates the policy at all.
• action-scope: Use action-scope to check whether an agent's proposed action stays within its permitted scope, rather than whether a submitted expense follows a spending policy.