# analyst3 idle run 14, 10 Oct 2026 (scripts d2-d4, e1-e5 in work/analyst3)

Only 6 lenses run (long sims); the rest were repeats, skipped on purpose.

1. **Launcher / create-tx `to_` (new field join, createtx.db):** v3 band: 0x7399 launcher 15% vs others 54% to 1M, but 0x7399 band pools are born at 500k (senders 1, 29 swaps) = frozen copies; with seed<7.8k both ~30-56%. Launcher = seed proxy. DISCARD.
2. **Create-tx calldata length and ETH value (metadata richness):** within low-seed band hit5 29/18/38/38% across top lengths, value halves 32% vs 29%. Flat. DISCARD.
3. **Repeat-buyer share in first 5 min (first5.db), low-seed v3 band n=155:** rep>=.2 38% vs 24% hit5 but sign flips when split by activity. DISCARD.
4. **Unwatched venues, fresh census (radar.jsonl, 215 crossings of 1M, 7-10 Oct):** v4 nohook native/weth/usdg 38 crossings, 1 non-jump (depth 2B = junk); the 5 "real" v4:stock:hook are the known token/token artifacts (BASCAT, BINU, BASKAT, BLACKROCK, AINU). Fifth confirmation that the missed nohook runners are jumps/mirages. DISCARD.
5. **Gas lens (analogy: sniper priority fees):** Robinhood Chain is Arbitrum Orbit, sequencer first-come-first-served, priority fee carries no urgency. No data request. DISCARD analytically.
6. **LP range at creation (Mint log from create-tx receipt, 92 low-seed band pools):** 92k-tick range hit5 40% vs 1.09M-tick 25%, p~0.15, two kit presets only. PARKED.
7. **NEW, A3-DIP (exit/entry timing, closing-line analogy: buy the post-signal dip, not the signal):** v3 band (cap 300k-1M at +5 min, all pools, no other flag), baseline entry = first print >= +305 s, tp5 sl.5 4 h, 7% slip + 1% fee, n=350, mean 1.07x. Resting buy at -X% from that price, valid 30 min after entry signal, fill on first print <= level (price floor 0.4x base), same tp/sl from the fill price:
   - -10%: filled 255, mean 1.28 vs same pools at base 1.08 (diff +0.20, bootstrap 5-95% +0.12..+0.28), hit5 41 vs 30
   - -20%: filled 239 (68%), mean 1.34 vs 1.03 (diff +0.32, +0.22..+0.43), hit5 43 vs 25, without top 3 winners 1.30 vs 0.98
   - -30%: filled 205 (59%), mean 1.37 vs 0.90 (+0.47), hit5 41 vs 17
   - halves (v3hist/v3live time order) -20%: 1.50/1.20 vs 1.18/0.88; thirds 1.51/1.41/1.11 vs 1.19/1.00/0.89. Unfilled pools earn base 1.18-1.33 (we miss 11-19 five-x winners of 350 that never dip); counting unfilled as cash 1.0 the per-signal mean at -20% is 1.23 vs 1.07.
   - Causal (no future print used; verified by rerun without a glitch filter). Median stays 0.465 (hit rate 18% vs 10%). Mean is not a runner catcher: hunt = 5x flip from a 300k-1M entry (peak 1.5-5M), same as band tests. Depth: band entry sizes as in other band tests; dip prints are sells, so the fill price is real.
   - Risks: in-sample (dip levels 3 tried, all agree), jump trap not applicable (entry on a real print down), slip on dip print assumed 7%. READY as forward arm -> testmgr.

## Exact forward-arm rule A3-DIP (for testmgr)
Pool in v3 mc5 band (cap 300k-1M at +300 s). P0 = first print >= +305 s. Arm A: resting buy at 0.8*P0; arm B: 0.9*P0; arm C: baseline buy at P0. Fill = first print <= level within 30 min after +305 s and >= 0.4*P0; no fill = no trade. After fill: tp5 sl.5 4 h, 7% slip + 1% fee, pull=0. Kill: arm A mean < 1.1 over 30 forward fills or not above C. Hunt: 5x flip from 300k-1M, not a first-minute catcher.