---
title: DVIDIA Sensor Corroboration v0.4
sdk: static
---

# Independent camera and actuator corroboration v0.4

A local synthetic PiPER benchmark tests separately processed RGB and drive checks against primary completion readings. It prevents 69 of 112 injected false completions; 43 known orientation, identity, elevation, sampling and replay cases remain. The conservative gate also abstains on 73 of 80 baseline true completions. This is not physical sensor independence, a robot controller or hardware qualification.

Open the [interactive evidence viewer](index.html), inspect [the paired results](benchmark.json), [case catalog](cases.json), [frozen protocol](protocol.json), and [integrity seal](integrity.json).

Reproduce using the pinned source commit in `protocol.json` and the commands in the [repository guide](https://github.com/Dvidia-Inference/dvidia-training/blob/main/docs/sensor-goal-verifier.md).