# analyst3 idle run 6, 10 Oct 2026 (scripts work/analyst3 r1-r10.py)

Hunt for all tests below: mc5-style v3 band entries (+306 s, cap 200k-1.5M at +5 min, n=282-294, v3hist+v3live) or v3 population (2709 tokens, >=20 swaps, 5-print median peak). Cost 7% round trip, mark-to-print.

1. REGIME / crowding (concurrency of band pools in prior 30/60 min, hour of day). Concurrency causal: 0 prior 1.37, 1 prior 1.31, 2+ 1.03 (n=28) = noise. UTC 00-05 (07-12 BKK): 2.09 n=50 (halves 1.39 / 2.69), matches A1 #150 hour effect. MATURE as sizing hint only; no pause rule.
2. POOL DEPTH / FLOW at +5 min (pool ETH after, net ETH inflow, buy share): net inflow, buy share, depth-to-cap flat. pool_eth_after is NULL for 76% of swaps (n=68 only; >=16 ETH 1.70 vs 0.7-0.96, tiny n). DISCARD.
3. EXITS on band entries (the one that matters): TP/SL grid, 4 h window. tp5 sl.5 1.28; tp10 sl.5 1.54; tp10 sl.3 1.67 (halves 1.78/1.57); tp20 sl.3 2.15 (2.29/2.01, only 16 hits); tp40 sl.2 2.38 (hit 0.7%). Ladders (half at 2-3x + trail) 1.07-1.27 = worse than single TP. Tight stop sl.7 1.06 vs sl.3 1.36 (stop is cutting winners that dip first). 1 h cap 1.10, 4 h 1.28, 24 h 1.30 (tp5): time stop under 2 h kills it. Jump trap: 0 of 30 tp10 hits came from a print under 5x; 0 not sustained 5 prints later. Depth to sell into at 10x NOT proven (pool_eth missing). READY as forward paper arm A3-WIDE (tp10 sl.3 4 h) + A3-WIDE20 (tp20), both on existing mc5 band entries, with real sell quote at exit.
4. META by name-word heat (new lens, ppdata names.db): token whose name/symbol shares a word (len>=3) with a token that crossed 1M in the previous 48 h (strictly before birth): P(1M) 20.3% vs 7.7% (n=344), halves 18.6/5.9 and 23.1/9.3. But entry at +60 s on same tokens (n=1708): hot 1.81 vs cold 2.21 tp5. Priced in; hot words: super, cat, wif, commander, robinhood, hood, protocol. DISCARD as entry filter; INFO that metas exist (clones of a running name run 2.6x more often, so the opportunity is being earlier than +60 s, A1 hunt).
5. v3 fee10000 (19 runners, 8 h to 1M, unwatched): first non-jump 300k touch with >=25 prior swaps: only 19 pools, tp5 4 h 0.94, tp10 24 h 0.69. fee3000 n=2. Fee100 same rule 0.88. DISCARD.
6. Token metadata supply/decimals (names.db): 98% are 1e9 x 18 dec. Non-round supply 31% hit 1M vs 8.8% but it is look-ahead (post-launch burns of winners: 72% below 1e9) plus real Robinhood stock tokens (SPY 21M, GOOGL, cbBTC, USDT). DISCARD (leak). Data ask: totalSupply at pool-creation block (ppdata, below).
7. PHANTOM RUNNERS by name: 78 tokens named "... • Robinhood Token" (tokenised stocks) + cbBTC/USDT lookalikes show as 1M-20M "runners" (SPY, AMC, GOOGL on radar today). Suggest runner-check/radar skip names containing "Robinhood Token" (see ppfeeds ask). INFO.
8. DATA ASK (wallet freshness): see a3-req-walletage.md.