Field-test deployment of AI monitoring devices across four ECEC centres in NSW, running on NVIDIA Jetson Orin Nano Super edge hardware.
docs/methodology.md— site selection (2 × 2 urban/regional × large/small), governance dataset sources, study design rationale.docs/jetson-processing.md— hardware and software stack on the Orin Nano Super, privacy boundary, failure modes considered.docs/data-acquisition-protocol.md— what is captured, what is persisted, consent and notification, retention, access control, incident workflow.docs/timestamp-attestation-procedure.md— how each Monte Carlo run's start/end is signed on-device, chained, and uplinked. Forward-looking procedure; no live attestation records are stored in this repo.jetson-logging/— device-side logging, privacy filter (the chokepoint), tegrastats collector, retention sweeper, and self-tests that gate the inference pipeline at boot.
Raw video and raw audio are processed on-device and never persisted or
uplinked. Only event records and aggregates pass the privacy filter in
jetson-logging/filters/privacy.py.
The supervisor refuses to start inference if the filter's self-tests
fail at boot.