Scoring Profiles

A scoring profile is a tuned scoring configuration — model weights, decision thresholds, and custom scoring logic — authored and tested by the Deepfaic team for your use case, and delivered to your edge deployment as a single signed file (a profile bundle, e.g. clip-jury_v3.dfprofile.json).

Profiles are strictly opt-in per job: importing a bundle changes nothing by itself. Jobs score with your CONFIG.yaml defaults unless a request explicitly selects a profile with profile_name (see Create Job).

Fully offline

Bundles are files — they reach your host however you move files today. The edge runtime never contacts Deepfaic cloud. Every bundle is cryptographically signed by Deepfaic; a file that was modified in transit, or that was not produced by Deepfaic, is rejected at import.

---

How it works

1. The Deepfaic team exports a profile bundle for you and sends you the file.
2. You install it — either via the Import endpoint (no restart), or by placing the file in <runtime-root>/data/profiles/ and restarting the stack (imported on boot).
3. API clients select it per job: "profile_name": "clip-jury" (latest installed version) or "clip-jury@3" (pinned version).
4. The profile is frozen onto the job at creation — its results never change afterwards, even if the profile is later removed or a newer version is installed.

---

Import a profile bundle

Installs one signed bundle. Re-importing a bundle that is already installed is a harmless no-op.

POST /v1/profiles/import

The request body is the raw bundle file, exactly as received.

curl -X POST http://<edge-host>:9123/v1/profiles/import \
  -H "Authorization: Bearer df_..." \
  -H "Content-Type: application/json" \
  --data-binary @clip-jury_v3.dfprofile.json

Response

{
  "status": true,
  "code": 200,
  "message": "Profile clip-jury@3 imported",
  "data": {
    "name": "clip-jury",
    "version": 3,
    "reference": "clip-jury@3",
    "scripts": ["volatility-guard@2"],
    "already_installed": false
  }
}

Field	Type	Description
name / version	string / number(int)	The installed profile identity.
reference	string	The name@version selector to use at job creation.
scripts	string[]	Custom scoring logic the bundle carried, if any.
already_installed	boolean	true when this exact bundle content was already installed.

Errors

Code	Condition
422	The file is not a valid bundle, the signature does not verify (modified or not produced by Deepfaic), or the same name@version is already installed with different content.

---

List installed profiles

GET /v1/profiles

data is an array of every installed profile, newest first.

Field	Type	Description
name / version / reference	string / number(int) / string	The profile identity.
status	string	active (selectable by new jobs) or removed.
imported_at	timestamp	When the bundle was installed.
scripts	string[]	Custom scoring logic used by the profile, if any.

curl http://<edge-host>:9123/v1/profiles \
  -H "Authorization: Bearer df_..."

---

Remove a profile

Removal is soft: new jobs can no longer select the profile, but every job that already used it keeps its results permanently. Re-importing the same bundle later reactivates it.

DELETE /v1/profiles/{name}/{version}

curl -X DELETE http://<edge-host>:9123/v1/profiles/clip-jury/3 \
  -H "Authorization: Bearer df_..."

Errors

Code	Condition
404	No profile with that name@version is installed.

---

Using a profile

Pass profile_name when creating a job — see Create Job:

{
  "input_path": "/var/lib/deepfaic/uploads/sample.mp4",
  "profile_name": "clip-jury"
}

The job's responses then report scores, thresholds, and the final verdict from that profile, and prediction_source identifies it (e.g. "clip-jury@3"). Jobs created without profile_name are completely unaffected and keep scoring with the CONFIG.yaml defaults (prediction_source: "default").