# Forensic watermarking

Forensic watermarking embeds an invisible identifier inside a video stream or file. The identifier is unique to a session, account, device or token. When a captured sample of unauthorized distribution is analyzed, the embedded identifier can be recovered to attribute the source.

## How A/B watermarking works
- Two perceptually identical variants of each segment are produced (A and B).
- Per session, an edge mixer or server-side workflow selects a unique sequence of A and B segments.
- The sequence is the identifier. From a captured sample, the sequence is recovered and mapped back to a single session.

## Properties
- Invisible: no impact on QoE or QoS in production conditions.
- Robust: designed to survive re-encoding, cropping, blurring, downscaling, screen capture and camera capture in real-world ranges.
- Pseudonymous: identifiers do not contain personal data on the wire.
- Independent of DRM: works alongside multi-DRM environments.

## What it solves
- Source attribution of unauthorized restreams and leaks.
- Evidence preparation for enforcement workflows.
- Deterrence at the subscriber and distributor level.
- Operational shortening of the unauthorized distribution window.

## Limitations
- Forensic watermarking does not block access. DRM does.
- No system is unconditionally unremovable. Forensic watermarking shortens the window of unauthorized distribution and produces attribution evidence.
- Detection latency depends on sample availability and workflow setup.