Check if a document is signed

The decision, and the limit

People searching for this usually want a yes or no before a contract, application, or loan packet moves forward. The honest output of this API is narrower than that question:

• Zero detections: no visual mark was returned.
• One or more detections: a box and a confidence score for each mark.
• Not included: the signer's name, a match against a signature on file, or proof that a certificate is valid.

That is still enough to stop obviously unsigned scans. It is not enough to skip review on a document where the wrong person might have signed, or where a mark might be a stamp or a logo. Keep a reviewer for anything you would not want a model to settle alone.

A small status function

The request is the image API. Branch on HTTP status first, then on result.count.

async function checkSigned(imageUrl, apiKey) {
  const response = await fetch("https://signature-detector.com/api/detect", {
    method: "POST",
    headers: {
      "Content-Type": "application/json",
      "x-api-key": apiKey,
    },
    body: JSON.stringify({ imageUrl }),
  });

  const data = await response.json();
  if (!response.ok) {
    throw new Error(data.error || `Detection failed (${response.status})`);
  }

  // 0.5 is an example cutoff for a review queue, not an API parameter.
  // The server already omitted anything below 0.1.
  const confident = data.result.detections.filter(
    (detection) => detection.confidence >= 0.5
  );

  if (data.result.count === 0) {
    return { status: "no-mark-detected", review: true, data };
  }
  if (confident.length === 0) {
    return { status: "low-confidence", review: true, data };
  }
  return { status: "mark-detected", review: false, data };
}

Store detectionId with the application or contract id. When someone asks why a file was flagged, you can show the boxes instead of a bare boolean. Cropping those boxes for the reviewer is covered in extracting signature crops.

PDFs and multi-page files

A single page image uses imageUrl or imageBase64. A PDF uses pdfUrl or pdfBase64. pages, count, and creditsUsed are top-level fields, and signed is true only when count is greater than zero. A file is a poor "fully signed" candidate when any required page has count: 0. Which pages are required is your checklist, not a field in the response. See PDF signature detection and automating signed-contract checks.

False confidence

The sample score in the API docs is 0.192, just above the fixed cutoff of 0.1. A mark that low is a weak signal. Sending it straight to "signed" will pass noise. The sample function above files those under low-confidence so a person sees them. On an image you can also send threshold from 0 to 1 and the server will omit lower scores. The default is 0.1. PDF pages always use 0.1. A value outside 0 to 1 is HTTP 400.