Repository navigation
Description
The self-media template (emrg/server/self_media_prompt.md) tells a round how to read trend data (§1), choose one topic (§2), draft (§3), publish once (§4) and archive readings (§5). It says nothing about where the round's material comes from when a piece needs more than a trend figure, and nothing about how many kinds of source it should try before concluding there is none.
What was observed
An installation's self-media task carries a task-specific block that states, in its own words, that each round should try several different kinds of source — repository authors and their commits, people cited in a project's docs, records already on this machine, network or browser sessions, public rankings — and should record which kinds it tried, which did not work, and why.
That is a general method, not a fact about one installation, and the template has no counterpart. Its §1 covers trend data only, with a live / fallback level and a trend_data reading; its §5 unread field records fields the draft omitted for want of a source. Neither says the round should vary the kind of source it looks in, and neither asks for the failed kinds to be written down.
The same block also states the intended effect: a round that tried one source and found nothing is indistinguishable, in the archive, from a round that tried five and found nothing — so the next round repeats the same dead end.
Why it matters
The template already prices the honest-absence rule for trends very carefully ("a round that reports a trend it never read is worse than a round with no live data"). The same reasoning applies one level up: a round that reports "no material" without saying which sources it tried teaches the next round nothing, and the archive cannot distinguish a barren subject from a lazy search. Source variety is also what makes the difference between a piece with specifics and a piece of generalities, which §3 asks for without saying where the specifics come from.
What the fix must satisfy
- The template states, in the section that owns "what was read this round", that material is gathered from several distinct kinds of source rather than one, and names them as examples rather than as a closed list.
- The round records which kinds it tried, which answered, and which did not and why — in the readings block that already carries absences, so an absence stays out of the piece.
- A round that tried one kind and found nothing says so; it does not present that as "no material exists".
- No new step is added: the rule lives inside the existing round, and the round still ends at a published, archived piece.
- The template's existing rules are untouched — the fact-discipline rule in §3 already states that a third party's properties come from their own statement, and this change does not restate or weaken it.
Acceptance
Reading the template, a round can answer "which kinds of source should I try, and what do I write down about the ones that failed" without consulting any installation's configuration. The prose guards over this template still pass.
Activity
pm25coder commentedon Nov 17, 2026
Handled by #2942