Coin picks v3 + social X test, 4 Oct 2026

1) Picks v3 (live in picks.py, used by chains.py for BUY alerts and the "Coins to Buy" pop-up)
   Pool: coins of the basket category, 24h volume >= $100k, listed 8+ days, top 50 by CMC rank.
   Drop: coins up 30%+ vs the day before the BUY close, and the single 7 day leader.
   Take: the 5 biggest left (by CMC rank). Equal amounts. Sell all when the basket sells.
   Why: on 85 buys on 11 never-used baskets (this session) and 210 buys (other session) no pick rule beat random.
   Equal spread gave the smallest losses. +30% day coins beat the basket 19% (median -22%), worse in all 4 periods.
   Biggest is a mild tilt only (other session 42% vs 36% random; no edge in this session's data).

2) Social X chatter as a pick factor (xq.py/xcollect.py fetch from twitterapi.io, key read from a 600 file, never printed;
   xanal.py; output in xanal_out.txt). 832 coin picks inside basket BUYs, Solana and Base, 2023 to 2026.
   Dead/quiet coins (0 to 2 posts in 7 days) beat the basket 49% vs 36% for alive coins, both halves (46/29, 51/40).
   Chatter fifths make a U: quietest 45%, loudest 42%, middle 36%, 4th fifth 27%. The "happy middle" was worst (35% vs 39%).
   Surging chatter (over 3x vs 2 weeks before) was worst in 2025-26 (29%, median -10%) but mixed in 2023-24.
   Likes per post add nothing clear.
   Verdict: not shipped as a rule (quiet coins are often illiquid and the U shape is not a clean signal). Shown as a reminder only.