Efficiency per bit, not per subscriber

Most efficiency reporting stops at the wrong number. Average bitrate per subscriber looks stable while the actual cost of delivery swings by a factor of three, because the average blends prime-time behaviour with everything else.

What a usable definition looks like

Efficiency is delivered value divided by resource consumed. Both halves need to be measured on the same interval, for the same population, with the same exclusions.

• Delivered value: minutes viewed at a quality threshold the viewer actually experienced, not the top rung of the encoding ladder.
• Resource consumed: bytes over the wire plus the energy and egress attributed to carrying them.
• Interval: hourly, not daily. Daily averages flatten exactly the peaks where the decisions get made.

Three traps

1. Survivorship in the ladder. Sessions that rebuffer once are not a rounding error; they are the population where the ladder failed.
2. Egress that is not counted. Inter-cache fills, retries and pre-warms all cost money and rarely appear in viewer-facing metrics.
3. Comparing across traffic mixes. A 30% improvement on a 4K-heavy catalogue is not comparable to a 30% improvement on mobile-only traffic.

A minimum viable report

Publish median and 95th-percentile delivered quality, bytes per viewing minute, origin egress per viewing minute, and rebuffer ratio — by device class and by hour. Four numbers and two dimensions are enough to make an honest claim.

Any claim without the traffic mix attached should be treated as marketing, including our own.