New experiment

Declare a hypothesis, a predicate that classifies your posts, and a primary metric. The system snapshots a baseline when you start it.

Title Hypothesis Category TIMING — which hours you post in LENGTH — character count range MEDIA — text vs image vs carousel HOOK — opening phrase pattern TOPIC — posts tagged to a specific topic CADENCE — posts per day (window-level) ENGAGEMENT — your own reply behavior (window-level) CUSTOM — manually tag posts Predicate (JSON) TIMING: {"hours": [19, 20, 21]}
LENGTH: {"min_len": 80, "max_len": 200}
MEDIA: {"media_types": ["IMAGE", "CAROUSEL_ALBUM"]}
HOOK: {"prefixes": ["you vs me", "hot take:"]}
TOPIC: {"topic_label": "AI models & Anthropic commentary"} Primary metric Reach rate — Median views per post ÷ follower count. The fraction of followers the algo is showing each post to. Reply rate per view — Total replies ÷ total views. The strongest conversation signal the algorithm scores you on. Reply : like ratio — Replies ÷ likes. 18%+ is strong. The algo rewards conversation over passive approval. Zero-reply posts — Share of posts that got zero replies. High = the algo is training a low-quality prior on your account. Top-decile reach multiple — Views at the 90th percentile ÷ median views. How far your breakouts travel above your baseline. Follower velocity — Follower gain per day, 7-day rolling average. Net growth from the account snapshots. Variant window (days) Target delta (optional, e.g. 0.25 for +25%) Start immediately (snapshots the baseline now) Create experiment