ai-security-research-radar

Regenerate README.md, the repo's landing page and only output surface: badges, a 3-4 bullet "since last scan" digest with inline links, a clickable trend table, the study shelf, and an output map. Use after every change to TRENDS.md, in the same commit. README.md is fully derived and never edited by hand: if it disagrees with TRENDS.md, TRENDS.md wins — regenerate.

Compute the radar's self-calibration: weekly funnel metrics (queue dynamics, exploration compliance, off-axis rate), a monthly hit/miss retrospective against what actually became big, and up to 3 curator proposals. Use during every weekly recalibration, after the source-strategy review; results go into the append-only `logs/calibration.md` and into the weekly report.

Sweep the primary security feeds — vendor/lab red-team & security blogs, research venues, and advisory feeds — on EVERY daily run (not rotated) so no attack-technique, red-team tool release, or disclosure is missed. Prefer RSS/Atom over HTML to survive JavaScript-rendered indexes. Use at the start of every daily scan, before the rotating exploration slot.