Integrity Monitor
Configurable Weights & Thresholds
How integrity deductions are calculated and configured.
All integrity settings are stored in the rubrics table (type: integrity) and editable via /admin/integrity.
Deduction Weights
Each flag type has a configurable deduction weight (how much it reduces the integrity score per occurrence):
| Flag | Default Weight | Configurable? |
|---|---|---|
multiple_faces | Configurable | Yes — slider |
extended_gaze_away | Configurable | Yes — slider |
tab_hidden | Configurable | Yes — slider |
window_blur | Configurable | Yes — slider |
clipboard_paste | Configurable | Yes — slider |
window_resize | Configurable | Yes — slider |
multiple_displays | Configurable | Yes — slider |
Detection Thresholds
For camera-based flags:
| Threshold | Default | Description |
|---|---|---|
| Gaze angle | Configurable | How far the candidate must look away to trigger |
| Gaze duration | Configurable | How long they must look away to trigger |
| Human review threshold | Configurable | Total deduction score that triggers a "human review" flag |
Admin UI
The admin page (/admin/integrity) shows slider controls for every weight and threshold, with a "Reset to Defaults" button.
Your Review
Key questions:
- Are the default deduction weights proportional to severity? (Is
multiple_facesworse thantab_hidden?) - Should weights scale with frequency (3rd tab switch worse than 1st)?
- What integrity deduction score should trigger "human review"?
- Are any flag types unfair? (e.g.,
window_resizemay trigger on legitimate window management)