Separate facts from conclusions
Perception records observable events such as a person entering a configured zone. Correlation and risk remain separate layers, and neither declares guilt or intent.
Responsible Visual Intelligence
Exceptions, not accusations.
AI observes. SightGuard correlates. Rules identify exceptions. Authorized people investigate and make the final decision.
Keep humans accountable
Investigators review evidence, operational context, and policy before recording an outcome. Automated scores remain signals, not determinations.
Explain and preserve
Detector confidence, model version, correlation rule, risk score, evidence, reviewer, notes, and timestamps are intended to remain traceable.
Responsible product principles
Move less video. Keep the necessary context.
The target SightGuard Edge architecture processes continuous video locally. Cloud services receive metadata, selected evidence, and configured uploads instead of requiring continuous raw CCTV transfer.
Use neutral operational language
SightGuard uses terms such as exception, anomaly, requires review, and potential issue. Human outcomes may include confirmed loss, policy violation, legitimate activity, false positive, or needs investigation.
Measure before claiming
Computer-vision performance depends on camera placement, lighting, occlusion, scene configuration, and representative evaluation data. SightGuard will document measured results and limitations before presenting a detector as reliable.
Minimize access and retention
Product controls are intended to support least-privilege access, configured retention, evidence minimization, review history, and auditable changes. Detailed controls and contractual commitments will be documented as the platform matures.
Development status: These principles guide the product during development. They do not claim that every described control is available today.
Evaluate responsibly
Test the principles against a real operational workflow.
Early Access and Operational Assessments help define useful events, limitations, evidence requirements, and responsible deployment controls.