The AI revolution in video streaming is built on a lie—or at the very least, a dangerous assumption that the data going into these models is trustworthy. This applies to ad insertion, personalization, churn prediction, QoE monitoring, security, you name it, because ultimately these systems are only as smart as the logs they are being trained on. And judging by the murmurings heard by Faultline across the industry, those logs are a mess. First, there is a lack of standardization across players and SDKs, which all have different ways of logging the same event types—pushing out inconsistent data. This could be a playback error in an analytics engine, for instance, which is labeled and timestamped differently on a Samsung Tizen…