Repository navigation
Description
What is on master
emrg/server/self_media_prompt.md gives one round two instructions about the same absence, and the round reads the first one first.
§1 grades the trend read, and its fallback level reads:
level "fallback" it could not be read, and the product says so
The Fallback bullet under it says to mark the product no real-time data was obtained.
§3, whose text was extended in the change that merged as 613dedd2, forbids exactly that: a field whose source cannot be named "does not appear in the draft at all, and not as "unknown", "not read" or "no data" either: a placeholder spends the reader's attention on the round's own gap. Where the gap still matters to the host, §5's readings carry it."
The Forbidden list restates the marker at the product too — "§1's fallback level, marked no real-time data was obtained, is the honest answer and a usable one" — so the instruction survives a reader who reaches the end of the file.
Why it matters
A round that follows §1 writes a placeholder into a published piece that §3 says must not be in it. A round that follows §3 leaves the absence unrecorded where §1 said to record it. Which rule applies is decided by the order the round happens to read, so both outcomes are defensible from the file and neither is a defect a later reviewer can point at.
The readings table already holds the right home: trend_data carries live with what was read and when, or no real-time data was obtained with why.
One smaller claim in the same table is false on its own terms: the unread row says "this is the one place an absence is recorded", while the trend_data row above it records exactly such an absence and the publish row records not attempted.
What the fix must satisfy
- An absence is recorded in §5's readings; the piece carries no marker for one.
- §1's fallback level, read before §3, names §5 as that home rather than the product.
- No section contradicts another about where an absence goes, the Forbidden list included.
- A guard reads the rendered prompt per section, in both
extra_promptstates, so that "which section says this" is a reading rather than a raw-text count.
Measured on master
Rendering the template through the daemon's own path (jinja2.Environment(undefined=jinja2.Undefined) against the evolution_context mapping) and splitting the result by heading, no real-time data was obtained appears in three sections: 1. Read the trends, 5. Archive, and Forbidden. §1 mentions §5 nowhere.
Activity
argszero commentedon Nov 15, 2026
Handled by #2907