radar-explorelisted

# Explore — find what we SHOULD be tracking, not only confirm what we do The sweeps (`radar-lab-sweep`, `radar-pulse`, `radar-repo-watch`) CHECK known sources for new items on the axes we already track. This skill does the opposite job: it finds work we do NOT yet track but that is important — the candidate next axis, the off-axis result the field is about to care about. It is list-driven exactly like the lab sweep: the venues live in `SOURCES.md` and you iterate ALL of them. You do not "explore something" by vibe — you execute a list of WHERE new work surfaces. WHAT to look for is never enumerated here (that is the trap that makes the search narrow); WHERE to look is the contract. ## Method (significance over topic-match) 1. ITERATE the venues in `SOURCES.md` → "Discovery / exploration venues" (see the list contract there: cross-category attention venues every run; one rotating arXiv category as supplement). For each venue, read its TOP / most-attention items — HF daily papers by upvotes, an arXiv category's newest listing, a trending feed's lead items — REGARDLESS of whether they match a tracked axis. 2. ADVANCE THE WINDOW: read items NEWER than the last time you covered THAT venue (its previous date in `logs/source_rotation.md`); record the range. A venue that returns "same batch already mined" did not advance — move the window forward or take the next venue in rotation. Never re-mine the same listing. 3. Assess each top item on its OWN merits, NOT