Editorial Policy
Moreno Valley View is written by an AI language model, working only from public sources, without a person reviewing each item before it publishes. That's disclosed plainly on /about and /how-we-gather-this. This page is about what stands in for that human review: a set of automated checks every piece of generated text has to pass first.
What every check has in common
None of this is a style pass. Each check either passes or fails; a failure means the text is rewritten once with a specific correction, and if it fails again, the item is either replaced with a plain, unstyled template restating the source facts directly, or — for the highest- stakes content — not published at all. No failed check is ever silently ignored to get something out the door.
The checks themselves
- Every fact must be in the source
- Numbers and proper nouns in the generated text are checked against the actual source document or feed item. A name, figure, or detail that doesn't appear in the source fails this check — regardless of whether it happens to be true.
- No verbatim copying
- Generated text is checked against the source for copied phrases of eight words or more. Everything is a paraphrase, never a lift.
- Claims about a named employer or business are always attributed
- Worker Pulse and New in Town summarize what reviews or outlets say about a real, named company — never as settled fact. A sentence naming a tracked business has to say "reviews mention…" or "according to…"; a bare, unhedged claim fails.
- Closure Watch never predicts a closure
- Before an official closure is announced, the Closure Watch page describes the current situation only — what alert is active, and that no announcement has been made. Predictive language ("likely," "expect," "keep the kids home") fails outright. If a paragraph can't pass this check, nothing publishes for that state at all, rather than risk a wrong guess.
- Home sales coverage is never financial advice
- The monthly home-sales digest reports what was recorded — prices, counts, ZIP-level activity. It never tells a reader it's a good or bad time to buy or sell, and never speculates on future prices beyond a real prior figure already in the data.
- Ongoing local stories never predict an outcome
- A tracked civic project's page can report a real, already-decided outcome (a vote that happened, a permit that was issued) — that's a fact. It can't say a decision is likely, expected, or on track to happen, because that hasn't occurred yet.
- Culture and science columns are third person, no invented experience
- The model has no experiences of its own to draw on, so it's never allowed to write "I think" or invent a personal anecdote. These pieces argue a point about something real and current — a work, a trend, a local angle — grounded in what's actually true about it, not a fabricated first-person voice.
- No arrests, court records, or obituaries
- Covered on /about too — this is enforced as a hard content filter, not just a stated intention.
When a check fails
Every check above runs before publication, not after. A failed check triggers exactly one retry with the specific problem named to the model, then falls back to a safe, minimally-styled restatement of the source facts — or, for Closure Watch specifically, to publishing nothing at all. Readers never see an item that failed its checks presented as if it passed.
Corrections
Automated checks catch what they're built to catch, not everything. If something still gets through wrong, tell us — see the corrections log for what's been fixed and when.