Check a statement's attributed source

Check whether supplied source text attributes a statement to the claimed speaker or source.

You need to check who said a statement in a transcript or source excerpt, separately from whether it is true.

Explore this recipe interactively · Source and implementation guide

Use attribution-match 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 { attributionMatch } from 'jev-recipes/attribution-match';

const result = await attributionMatch({
  "statement": "The deadline is Friday.",
  "attributedTo": "Mira",
  "source": "Oren: The deadline is Friday.\nMira: Thank you for the update."
});
console.log(result);

Input contract

Field	Type	Needed
statement	string	Required
attributedTo	string	Required
source	string	Required
minConfidence	number	Optional

Full input and result schemas

{
  "input": {
    "$schema": "https://json-schema.org/draft/2020-12/schema",
    "type": "object",
    "properties": {
      "statement": {
        "type": "string",
        "description": "One statement or faithful paraphrase whose attributed speaker or source should be checked."
      },
      "attributedTo": {
        "type": "string",
        "description": "The speaker or source claimed to have made or explicitly endorsed the statement."
      },
      "source": {
        "type": "string",
        "description": "The supplied transcript or excerpt, including speaker labels or attribution context needed to check the claim."
      },
      "minConfidence": {
        "type": "number",
        "minimum": 0,
        "maximum": 1
      }
    },
    "required": [
      "statement",
      "attributedTo",
      "source"
    ]
  },
  "result": {
    "$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": [
          "matched",
          "mismatched",
          "not_attributed",
          "unclear"
        ]
      },
      "confidence": {
        "type": "number",
        "minimum": 0,
        "maximum": 1
      },
      "probabilities": {
        "type": "object",
        "propertyNames": {
          "type": "string",
          "enum": [
            "matched",
            "mismatched",
            "not_attributed",
            "unclear"
          ]
        },
        "additionalProperties": {
          "type": "number",
          "minimum": 0,
          "maximum": 1
        },
        "required": [
          "matched",
          "mismatched",
          "not_attributed",
          "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 attribution-match.

{
  "model": "demo-fixture",
  "usage": {
    "input_tokens": 0,
    "output_tokens": 0
  },
  "status": "ready",
  "verdict": "mismatched",
  "confidence": 0.97,
  "probabilities": {
    "matched": 0.01,
    "mismatched": 0.97,
    "not_attributed": 0.01,
    "unclear": 0.01
  }
}

Evaluation evidence

Fixture only

No verified live accuracy measurement is available. Evaluate representative cases before using this decision in your workflow.

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

Limitations

Checks only the supplied attribution record. It does not authenticate a source or establish original authorship, truth, or copyright ownership.
Missing statements and unresolved speaker identities require review; absence from an excerpt is not proof of false attribution.
Nested quotation and endorsement are distinct. Include enough surrounding text to establish who is speaking.

Related recipes

causal-attribution: Use causal-attribution to classify the kind of cause in an explanation, rather than who made it.
citation-match: Use citation-match to check which passages support a claim, rather than who made it.
claim-stance: Use claim-stance to label a response's own position toward a claim, rather than verifying a named attribution.
reference-resolve: Use reference-resolve to select the referent of an ambiguous expression from supplied candidates.