Questions

Short answers to the questions the desk meets most often. Where a longer note exists, the answer links to it.

Which codec should carry my top rung?

At 1440p and above, the newest codec usually wins on size at a fixed distortion. Keep the floor in the widest-read codec so weaker devices are not stranded, and use the newest codec only at the ceiling where the strongest devices arrive. See ladder design.

Why does my size change when I change nothing?

A driver revision, a thermal limit or a frame-threaded decode can move a result by a few percent. Record the host class and the engine beside every figure so a small drift is not read as a codec change. See the protocol.

What does the rate factor actually control?

It scales the quantiser, so it moves size and distortion together. A low value keeps detail and lowers the size saving; a high value does the reverse. The scale differs between codecs, so compare curves, not factors. See constant quality.

Is ten-bit worth it at a fixed size?

At high definition, yes. The size penalty is small and the banding it removes on gradients is large. The desk keeps ten-bit reconstruction for every 4K rung and most 1080p rungs.

Do I need a per-title ladder?

Only where a title is watched enough to pay for the extra measure, store and check. A fixed ladder with a few content classes covers most catalogues at far lower operating effort.

What is a run note?

The record of one encode, listing every option that touched the output. Without it a result cannot be repeated and is treated as folklore.

Why do two metrics disagree?

Each weights error differently. A pixel metric catches shifts; a perceptual metric catches visible structure. Report both, with the crop and the viewing assumption.

How does the desk score an accelerated engine?

On the same clip and the same codec path as a full encode, with the engine and driver recorded beside the figure. See accelerators.