What the FYP ranker reads
TikTok's per-video ranking model takes raw like count as one input but weights two derived signals more heavily: like-velocity (likes per minute, normalized to current distribution size) and like-source-quality (the ranker grade of the accounts liking the video). A video distributed to 1,000 viewers that gets 50 likes in the first 10 minutes shows velocity-strong (5%/min). A video distributed to 1,000 viewers that gets 500 likes in the first 30 seconds shows velocity-anomalous — the ranker reads that as bot-pattern and quarantines further distribution while the spam-heuristic samples the source accounts.
