CtrlK
BlogDocsLog inGet started
Tessl Logo

jbaruch/face-recognition-calibration

Production-grade dlib face_recognition toolkit: piecewise confidence formula, enrollment quality diagnostics, and producer-side persistence for flicker suppression.

96

2.70x
Quality

93%

Does it follow best practices?

Impact

100%

2.70x

Average score across 6 eval scenarios

SecuritybySnyk

Passed

No known issues

Overview
Quality
Evals
Security
Files

criteria.jsonevals/scenario-2/

{
  "context": "Tests whether the agent applies the correct blur threshold (40, not 80) for pale-skin-safe capture validation, targets the correct intra-class distance range, uses pickle for enrollment persistence, and avoids re-enrolling from raw images in the load path.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "Blur threshold is 40",
      "description": "The Laplacian variance rejection threshold used in frame validation is 40 (not 80, not 100, not 150)",
      "max_score": 16
    },
    {
      "name": "Blur threshold not 80",
      "description": "Does NOT use 80 as the blur threshold (80 is explicitly the wrong value that causes false rejections for pale skin)",
      "max_score": 10
    },
    {
      "name": "Intra-class mean target lower bound",
      "description": "Pre-publish validation checks or design_notes.md states intra-class distance mean target minimum of 0.25 (not lower)",
      "max_score": 12
    },
    {
      "name": "Intra-class mean target upper bound",
      "description": "Pre-publish validation checks or design_notes.md states intra-class distance mean target maximum of 0.40 (not higher)",
      "max_score": 12
    },
    {
      "name": "Intra-class max check",
      "description": "Pre-publish validation checks or design_notes.md flags intra-class max distance > 0.50 as an outlier signal",
      "max_score": 10
    },
    {
      "name": "Pickle for enrollment storage",
      "description": "Enrollment is saved using pickle (not JSON, not numpy .npy, not CSV) — design_notes.md or code uses pickle.dump/pickle.load",
      "max_score": 14
    },
    {
      "name": "No re-enrollment in load path",
      "description": "The production load path loads from the pickle file directly — it does NOT re-compute encodings from JPEG/image files at load time",
      "max_score": 12
    },
    {
      "name": "Pale skin rationale",
      "description": "design_notes.md mentions that pale or fair skin produces lower Laplacian variance on sharp photos, justifying the lower blur threshold",
      "max_score": 14
    }
  ]
}

evals

README.md

tile.json