Deeptrack Logo
Deeptrack
Gotham back office
Client view
OverviewScan logAPI usageCredits & billingClient accounts
Internal
FP / FN queueModel feedbackScan forensicsDataset pipeline
System
API keysWebhooksAudit logAlertsSettings
AD
Admin
Admin
Settings
Platform configuration — thresholds, notifications, model version, admin users
Scan verdict thresholds
Applies to all new scans immediately
Deepfake verdict minimum
Score ≥ 75% → verdict: deepfake
75%
Review zone minimum
Score ≥ 40% and < 75% → verdict: review
40%
Review zone maximum
Above this = deepfake. Below = authentic
74%
Current logic: score ≥ 75% → deepfake  | 40–74% → review  |  <40% → authentic
Per-media type threshold adjustments
Applied on top of global thresholds
Audio threshold adjustment
Audio has higher FP rate — raise the bar by this many points
5%
Video threshold adjustment
0 = use global threshold with no adjustment
0%
Image0 pts — Global only
Document0 pts — Global only
Model version
⚠ Switching model affects live traffic
v4.1.2active
Released 2026-03-28 · Accuracy 94.1%
SELECTED
v4.0.8previous
Released 2026-02-14 · Accuracy 93.3%
v3.9.1archived
Released 2026-01-10 · Accuracy 92.5%
Model rollback to v4.0.8 takes ~30s. All in-flight scans complete on current version.
Notification preferences
Email alerts
Alert emails to all admins
Slack alerts
Via wh_004 internal Slack webhook
Credit warnings
When a client hits 80% usage
FP/FN escalation alerts
When auto-escalation triggers
Weekly summary report
Emailed every Monday 08:00 EAT
Model drift alerts
When confidence drops > 3pts vs 30d avg
brian@deeptrack.io
tauil@deeptrack.io
Admin users
BK
Brian K'Oyundiadmin
brian@deeptrack.io · Last login: 2 min ago
TM
Tauil M.reviewer
tauil@deeptrack.io · Last login: 1h ago
WN
Wanjiku N.reviewer⚠ No MFA
wanjiku@deeptrack.io · Last login: 2d ago
⚠ Danger zone
Flush Redis scan cache
Clears all cached scan results. In-flight API calls may fail.
Rotate platform API salt
Invalidates ALL existing API keys. All clients must re-issue keys.
Export full dataset
Downloads all labelled scan data as encrypted ZIP for ML team.