Rule v4.5, 4 Oct 2026 (Bangkok). Answers and tests behind it.
1. Runs and corrections (cycles/runs.py, runs_composite.txt, runs_per.txt): zigzag on daily closes of on-chain baskets
   (run = rise of 30%+, correction = fall of 20%+), 5 definitions (3 composite thresholds, memes only, apps only, each basket alone).
   Per basket median run length: DeFi summer 2020 158 days, 2021 bull 41, 2023-24 bull 28, Nov 2024 to Jan 2025 14, 2025-26 chop 26.
   Correction to the low: median 15 to 35 days (-33% to -46%); back to a new high: median 31 to 54 days in bulls, many never.
2. New coin mania (bull/chainwide.py, chainwide_out.txt): the earlier Solana launchpad finding was wrong (it only covered Dec 2024).
   Full 2024: high launchpad share was followed by BETTER Solana memes. The all-chain new coin share (CoinMarketCap, listed < 30 days)
   was followed by weaker alts market wide in 2025-26 (SOL coin, majors, memes, apps), mixed in 2023-24. Meme mania gate kept.
3. 50 breakout ideas (r50.py, r50_break_*.txt): none beat the 30 day high close (20 in hot markets) in both eras.
4. 50 bottom ideas x 3 modes, then 2 refinement rounds (r50.py, r50b.py, r50c.py): kept BUY (BOTTOM): after a 40%+ fall from the
   90 day high, a higher low, then a close above the last swing high (6 day swings), BTC above its 50 day average.
   2019-22 +3370% vs +2168%; 2023-26 +1901% vs +1914%; worst drop -24% both eras. Bottom buys alone are far weaker than breakouts.
5. Exits and extended, 10 + 4 iterations (exits10.py): the live rule stayed best overall; every rule gives back about 17 to 22% from
   the best close on winners. Live rule per basket type (percls_out.txt): memes, apps/AI and majors all profitable in both eras.