Recipe catalog / appeal-grounds-kind

Classify the ground an appeal asserts

What ground does appeal primarily assert: a factual error, a procedural error, new evidence, hardship, a misapplied rule, or something else?

You need to sort incoming appeals of a benefits, permit, enforcement, or academic decision by the kind of argument they make, so each reaches the right reviewer or template before anyone reads the file.

Explore this recipe interactively · Source and implementation guide

Use appeal-grounds-kind 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 { appealGroundsKind } from 'jev-recipes/appeal-grounds-kind';

const result = await appealGroundsKind({
  "appeal": "I am appealing the denial dated September 3. The letter says my household income is $4,200 per month, which is over the limit. That number includes my adult son's wages, but he moved out in March and no longer lives with me; I told the office this at my interview. My actual household income is my own $2,600 per month. Please correct the income figure and reconsider my application.",
  "minConfidence": 0.8
});
console.log(result);

Input contract

{
  "input": {
    "$schema": "https://json-schema.org/draft/2020-12/schema",
    "type": "object",
    "properties": {
      "appeal": {
        "type": "string"
      },
      "minConfidence": {
        "type": "number",
        "minimum": 0,
        "maximum": 1
      }
    },
    "required": [
      "appeal"
    ]
  },
  "result": {
    "$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"
        ]
      },
      "verdict": {
        "type": "string",
        "enum": [
          "factual_error",
          "procedural_error",
          "new_evidence",
          "hardship",
          "misapplied_rule",
          "other",
          "unclear"
        ]
      },
      "confidence": {
        "type": "number",
        "minimum": 0,
        "maximum": 1
      },
      "probabilities": {
        "type": "object",
        "propertyNames": {
          "type": "string",
          "enum": [
            "factual_error",
            "procedural_error",
            "new_evidence",
            "hardship",
            "misapplied_rule",
            "other",
            "unclear"
          ]
        },
        "additionalProperties": {
          "type": "number",
          "minimum": 0,
          "maximum": 1
        },
        "required": [
          "factual_error",
          "procedural_error",
          "new_evidence",
          "hardship",
          "misapplied_rule",
          "other",
          "unclear"
        ]
      }
    },
    "required": [
      "model",
      "usage",
      "status",
      "verdict",
      "confidence",
      "probabilities"
    ],
    "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 appeal-grounds-kind.

{
  "model": "demo-fixture",
  "usage": {
    "input_tokens": 0,
    "output_tokens": 0
  },
  "status": "ready",
  "verdict": "factual_error",
  "confidence": 0.86,
  "probabilities": {
    "factual_error": 0.86,
    "procedural_error": 0.01,
    "new_evidence": 0.06,
    "hardship": 0.02,
    "misapplied_rule": 0.03,
    "other": 0.01,
    "unclear": 0.01
  }
}

Evaluation evidence

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

Limitations

• Classifies the argument the appellant makes, not whether it is correct or whether the appeal should succeed. Merits review belongs to the adjudicator.
• An appeal that asserts several grounds is classified by the one it presses most; split multi-ground appeals in code when each needs its own review.
• Does not check timeliness, standing, or whether the ground is one the process allows.

Related recipes

• correction-target: Use correction-target to identify what part of a prior output a correction points at, rather than what kind of ground an appeal asserts.
• feedback-kind: Use feedback-kind to classify general feedback on a product or service, rather than a formal appeal of a decision.