Media
An image is evidence about itself and nothing more. The metadata tells you how it was encoded; it does not tell you who made it, whether a model was involved or what it depicts in the world.
The work is the evidence that it held.
Most pipelines strip it: a chat app re-encodes, a CMS runs an optimiser, a screenshot pipeline rebuilds the pixels. What arrives at your editor is usually a bare JPEG with no camera, no software tag and no Content Credentials, which is why absence of provenance proves nothing.
What it does tell you is that the image cannot be authenticated from the file alone, and that any claim about its origin needs a source outside the file.
The media checks look at what the bytes actually contain.
No offline check can tell you whether a picture was generated, whether it depicts a real event, or whether using it creates a rights problem. Detectors produce probabilities, and probabilities are not sign-off.
That is the point where a media specialist reads the asset in context: the campaign, the market, the claims on the page and the licence behind it. They name what the file can support and what it cannot.
Stripped provenance
Assets whose EXIF or Content Credentials were destroyed before delivery.
Specification drift
Wrong dimensions, colour space or format for the placement it was ordered for.
Quality loss
Visible compression and upscaling that will fail in print or on a large screen.
Residual strings
Prompts, generator tags and file paths left inside the container.
Caption mismatch
Alt text and captions describing something the pixels do not show.
AI media quality
The automated pass is free, needs no account and uploads nothing. Escalate to a named reviewer only where being wrong is expensive.
