Transcoding Standards in 2026: Measured, Not Assumed
The Measurement Desk records how modern video compression actually behaves once a source is captured, quantised, predicted and reordered into a compliant stream. We do not repeat marketing claims about codec generations. Every figure on this portal comes from a bench run that another engineer can repeat with the same source clip, the same rate control mode and the same scoring routine.
Read the baseline summary Standards baseline
What the lab holds constant
Compression results drift for boring reasons: a different source grain, a changed colour primaries tag, a decoder that discards metadata, or a rate controller that quietly shifts its quantiser step between runs. To compare encoders honestly we fix the input. Each clip is stored lossless at a known matrix, with a documented frame count and a recorded checksum. Reference decodes are kept alongside the source so that any distortion score can be recomputed years later without the original master.
The lab records four families of output: the compressed elementary stream, the reconstructed frames, a per-frame distortion series, and a machine-readable run note that names every option that touched the encoder. That last artefact matters most. A result without its option list is folklore, not evidence.
How the stages connect
The portal follows a five-stage pipeline. Prep covers scope, this landing page and the benchmark protocol. Standards covers container formats, constant-quality rate control and accelerated encode engines. Encode opens up H.264, HEVC and AV1 internals, ladder design and the two anchor setting tables. Verify explains distortion metrics and the neutral encoder comparison. Reference holds the glossary, questions, about page and the legal notices.
What the desk measures, and what it refuses to measure
We measure bitrate at a fixed distortion target, the shape of a distortion curve across a quality range, decode time on a documented host class, and the gap between a claimed conformance profile and an observed one. We do not measure subjective preference and we do not publish a single opaque quality number detached from its source. A score is always tied to a clip, a crop and a metric definition.
Every page carries the same header bar so a reader can see the stage, the section and the revision date at a glance. Figures are drawn in the page itself as vector shapes, which keeps the document readable offline and keeps the file count small. Photographs, where used, come from open image banks and are credited on the about page.
Where to begin
A reader arriving for the first time should open the 2026 baseline, then the benchmark protocol that defines how a run is recorded. Engineers chasing a specific tier should jump to the 1080p or 4K setting tables. Anyone weighing encoders should read the comparison matrix before tuning anything.