Automated QC is the set of machine checks that sit at fixed points in a video pipeline and turn a measurement into a decision, pass, warn, or block, without a human watching every frame. A metric is just an input; QC is the surrounding machinery that compares it to a threshold and acts. There are three natural checkpoints: ingest validation confirms the source conforms to spec, a post-encode quality gate scores the encode against the source (a full-reference measurement, typically VMAF), and a pre-publish check validates the packaged deliverable. Checks span layers from cheap structural reads to decoding the actual frames for black-frame, freeze, blockiness, and loudness detection, plus modern machine-learning detectors for perceptual artifacts. Run the cheap checks first and fail fast. The key catch is the design choice between a hard gate that blocks and a soft gate that only warns, and tuning for recall without flooding humans with false alarms, since real defects are rare. The point is to spend human judgment only where it is needed.
