Reading the response

Everything below is diagnostics. The decision was already made from status and spoofing; this page is for understanding what happened, populating your logs, and filing tickets with context.

Response fields

Field	Type	What it’s for
`spoofing`	boolean	The verdict. `false` with `200` is the only approval.
`id`	string	Unique transaction identifier. Always log it.
`code`	number	Mirrors the HTTP status code.
`message`	string	Human-readable result. Useful in logs, not for deciding.
`info`	object \| string \| null	Your reference and, when present, the diagnostics. See below.
`date` / `time`	string	When the service ran the analysis.
`values`	object	Internal model scores (`FAKE`, `REAL`). Diagnostics.
`confidence`	number	Confidence of whichever component answered. See the scale warning.
`imageQuality`	object	Optional map of quality indicators. See below.
`benchmark`	number	Milliseconds the analysis took.
`errorCode` / `errorMessage`	number / string	Internal detail. Diagnostics.

Do not build business logic on errorCode, values, confidence, or info. They are not part of the contract and their shape changes between service versions.

`info`: three shapes, one reader

info is polymorphic. It can arrive as an object, as a string, or as null — and null shows up in two different places: the top-level info and the nested info.info.

{
  "info": {
    "info": "onboarding-12345",
    "mode": "active",
    "service": "as-face",
    "user": "userName",
    "deepfake-filter": { "provider": "A", "detected": false, "confidence": 99 }
  }
}

{
  "info": "onboarding-12345"
}

{
  "info": null
}

function reference(body) {
  const i = body?.info;
  return i == null             ? null
       : typeof i === "string" ? i
       : i.info ?? null;
}

An object in info does not mean the person was blocked. In our test set, all 47 approvals carried info as an object, with deepfake-filter and detected: false — the filter ran and found nothing. The only field that separates approve from do-not-approve is spoofing.

`imageQuality` is optional, and so are its keys

imageQuality is a map of optional keys: it can be missing entirely, and when present it may carry some keys and not others. Across 116 test responses, 102 carried face, 10 carried brightSpots, and the 4 422 No Face Detected responses carried no imageQuality at all — consistent enough: no face, nothing to measure.

"imageQuality": {
  "blurry": false,
  "underExposure": false,
  "overExposure": false,
  "borderDetected": false,
  "face": {
    "is_centered_x": true,
    "is_centered_y": true,
    "is_close_to_borders": false,
    "is_too_small": false,
    "is_too_large": false
  }
}

Read every level defensively. body.imageQuality?.face?.is_too_small is correct; body.imageQuality.face.is_too_small throws the day a 422 arrives.

Which component answered

When info is an object, its keys reveal what the service found. This table is for reading a case, not for deciding on one:

Situation	Signals in `info`	Typical `imageQuality`
Clean analysis	`deepfake-filter` with `detected: false`, plus `mode`	`face`
Watermark or overlaid text	`text` and `matches[]`	`brightSpots`
Chromatic tampering (“blue people”)	`blue-filter`	`brightSpots`
Deepfake detected — provider A	`deepfake-filter` with `detected: true`	`brightSpots`
Deepfake or AI-generated content — provider B	`deepfake-filter` with `detected: true`	`brightSpots`
Photocopy	Text in `info`, or `null`	`face`
Screenshot	Text in `info`, or `null`	`face`
Another screen’s border in frame	`borderDetected: true`	`face`

Never write logic against the value of info.service. Internal service names change. If you need to know whether diagnostics were attached, check for presence, not content.

Confidence values are not comparable

This is the most frequent misreading. Several numbers in the response are called “confidence”, they use different scales, and two of them point in opposite directions:

Value	Scale	What it measures	On detection
`deepfake-filter.confidence` (provider A)	0–100	Probability the image is authentic	Drops to `0`
`deepfake-filter.confidence` (provider B)	0–1	Probability the image is fake	Rises to `0.99`
`blue-filter.confidence`	0–100	Intensity of the chromatic tampering	—
`matches[].score`	0–100	Similarity of the found text to the reference list	—

Never compare, average, or apply a shared threshold to these values. A confidence: 0 can mean “absolute certainty of fraud” (provider A) and so can confidence: 0.99 (provider B). Any rule of the form “if confidence is above X” produces inverted decisions depending on which component answered.

Errors

Status codes, how to tell your problems from the service’s, and which are safe to retry.

Start here

ID Analysis

Face Analysis

Facial Antispoofing

Connection Fingerprint

Response fields

`info`: three shapes, one reader

`imageQuality` is optional, and so are its keys

Which component answered

Confidence values are not comparable

Errors

​Response fields

​info: three shapes, one reader

​imageQuality is optional, and so are its keys

​Which component answered

​Confidence values are not comparable

Errors

Response fields

`info`: three shapes, one reader

`imageQuality` is optional, and so are its keys

Which component answered

Confidence values are not comparable