scheduled-publish-preflightlisted

# Scheduled publish preflight Scheduled runs fail in a specific, boring way: they inherit stale assumptions from the previous run and then build an entire run on top of them. Four consecutive ThumbGate runs (2026-08-05 → 2026-08-10) burned their budget this way. Every failure below actually happened; none of them were caught by the run that made them. Run these four checks first. They cost about a minute together and they protect everything downstream. ## 1. Establish the date from the clock, never from the repo ```bash date -u +%Y-%m-%d ``` Cross-check it against the date in your system context. If they disagree, stop and say so — do not pick one silently. **Why this is first:** on 2026-08-10 a run read the branch `content/2026-08-06-am` and the last log row (`2026-08-05`), inferred "today is 2026-08-06," and was wrong by four days. That single bad inference produced a wrong campaign id, a wrong branch name, a wrong PR title, wrong dates in every log row — and, worst, it made the run read the *wrong* morning run in check 4, so it duplicated that day's persona and meme. A whole run had to be closed and redone. Repo artifacts tell you what happened last time, never what time it is now. The most recent branch or log row is a lagging indicator, and on a run that fires late or after a gap it is guaranteed to mislead. ## 2. Detect where you are actually executing ```bash uname -s -m; hostname echo "$CLAUDE_CODE_REMOTE / $CLAUDE_CODE_REMOTE_ENVIRONMENT_TYPE" ls ~/.config/g