← ClaudeAtlas

growth-looplisted

The learning loop for the clip page: log every post, read metrics at 12 h / 24 h / 72 h, run ONE one-variable test per day, no verdict under 24 h, daily and weekly reviews by pillar, slot and platform, write lessons as rules in learning.md, and judge posts on non-follower reach. Use for daily reports, weekly reviews, picking today's test, and proposing rubric changes.
vaibhav-barman/live-clipper · ★ 0 · AI & Automation · score 72
Install: claude install-skill vaibhav-barman/live-clipper
# GROWTH LOOP — make every post teach the next one Adapted with credit from Parker's (@DotComParker) faceless clipping-page setup prompt. Kept: the scoreboard columns, 12/24/72 h reads, one test a day, the daily and weekly reviews, learning.md. Changed: reads are automated (`fetch_stats` files each read in its age bucket) and the scoreboard is exported from the ledger (`export_scoreboard`). ## The rules 1. **Never guess a number.** If a metric isn't available it stays blank. A blank is honest; a plausible number poisons every later decision. 2. **Reads at 12 h, 24 h, 72 h.** `fetch_stats` keeps the first read past each mark. Call it a few times a day (`/clip-report` does). 3. **One test per day, one variable.** Hook style, clip length, posting time, caption length, layout (split vs gameplay-only), tease vs no tease. Name it (`length-15-vs-30`) and tag the day's clips with `update_clip(..., {"data": {"test_name": ...}})`. 4. **No verdict under 24 h.** Prefer the 72 h read if the post is still moving. 5. **Judge on reach to non-followers**, not follower count. A new page grows from people who don't follow it yet. Early engagement on a re-purposed account will be low — that's expected. 6. **Daily review (evening):** best and worst post, one lesson written as a RULE, tomorrow's test. 7. **Weekly review:** top 3 and bottom 3, median views by pillar, slot and platform. Double down on what's above median; cut or rework what's been below median for a full week; p