CLI Companion Baselines

`ed companion baselines`

Your own delivery baselines: the median and spread of stored signals, bucketed by recording context and language.

Usage:

ed companion baselines [--json] [--endpoint <url>]

Absolute acoustic values are close to meaningless. A hundred and forty words a minute is fast for one person and slow for another. But this is a system holding thousands of hours of exactly one speaker, so the comparison is against your own history rather than a population, and that turns weak features into strong ones. "Speech rate 2.1 below your own baseline, sustained for four minutes" is auditable and personal in a way no general model could produce.

Buckets matter as much as the numbers. A different microphone or room shifts energy and pitch estimates more than mood does, and your baseline in English is not your baseline in Hindi, so signals are compared only within their own bucket.

Until about twenty hours of audio have accumulated, deviations are suppressed entirely and this command says so. Showing deviations against four recordings would be noise wearing a number.

After that global gate, a kind and context bucket still needs at least 20 samples before it receives deviation scores. The scorer currently recognizes wpm, pause_s, f0_median, f0_range, rms and filler_rate; the command also lists other stored signal kinds. audioSeconds sums every episode with a duration, including timed video episodes.

--json shape: {audioSeconds, coldStart, baselines: [{kind, contextBucket, median, iqr, samples}]}.

These numbers are plumbing, not a dashboard. They exist so the system notices where to look and can justify what it says, not so you can watch a daily score.