# Analyst1 idle run, 10 Oct 2026 (iter 26): where are the runners really?
Scripts r1-r9.py in /opt/crew/work/analyst1. Data: e1x-runners-v2 (27k pools, 21 Sep-9 Oct), r6 tapes, live radar.jsonl (7-10 Oct).

## Headline
1. **The "missed $20M runners" in unwatched pools are mostly not runs.** In the live radar (3 days), of the pools that crossed $1M: v4 USDG 28 of 30 were born above $500k (cap_first >= half the level, depth 0 or absurd), v4 native 24 of 24, v4 WETH nohook 5 of 5. Mouse (24.8M at first print, depth $24k, 1 sender, fee 1%), BONER (41.6M first print, depth $62 billion), SPY, STANDARD, CASHCAT, AMC: all crossed 300k..20M at age 0. Climbers: USDG 2, native 0, WETH nohook 0 at $1M; v3 36, hooked WETH 1, v2 2. The runner-check / runner-map score should drop born-above pools (cap_first >= 0.5 x level) or flag them. Otherwise "missed runners" and the focus on nohook pools are inflated.
2. **Run 23 USDG-nohook 300k-touch "trail 2.93" was inflated by a glitch cap print** (usdg_caps has 1.8e302 and 3000M caps; one print became a 1e4x exit). After masking caps > 60M: n=59, tp/trail variants 1.26-1.48, best half-at-3x + trail .5 after 3x: 1.64 (CI 1.31-1.97, halves 1.78/1.50, ex top 3 1.42). Honest verdict: thin, about 1.3-1.6 gross, no handover.
3. **H3000 hooked pools (fee 3000): birth hour 08-20 BKK real-runner rate 38.6% vs 25.5% at night** (swaps>=10, n=395; day>night on 17 of 19 days; both halves). Other fee classes flat. Possible size/pause regime for the h3k test: ask testmgr to split h3k trades by entry hour.

## The 10 iterations
| # | lens | test | verdict |
|-|-|-|-|
| 1 | launch preparation: first-swap latency after pool creation (new) | P(real 1M) by latency, USDG/WETH nohook; sim of 300k entry by latency | population USDG lat<=100 blocks 4-7% vs 1.6-2%, but conditional on 300k touch flat (70-87%), sim n=59 flips (lat>100 better). DISCARD |
| 2 | hot hand: runners cluster in time (new) | hourly real-runner counts, prior 3h vs next 3h, by type | rho 0.21 all, v3 -0.06, USDG 0.11, hook 0.04; daily 8-65 steady. DISCARD |
| 3 | exits: partial TP / trail on USDG 300k entries | n=59, glitch-masked | 1.3-1.6, correction of run 23. MATURE-thin, OOS needs #476 |
| 4 | token lineage: later pool of a token (relaunch/derivative) | real rate by pool rank, previous pool peak, gap | later nh_USDG pool of a token whose earlier pool was a real runner 6.1% vs 1.5% (n=413) = same token already visible in a watched pool (a3 #144). INFO |
| 5 | market regime: hour/weekday of birth | real rate by 4h block and weekday, by type | USDG and v3 flat; H3000 day 38.6% vs night 25.5% (see headline). MATURE, ask testmgr |
| 6 | order-flow bots: gas price / priority fee of first 10 buys (new, field only exists in fresh hooked tape) | spearman with price extreme, A1 n=137, H3000 n=28 | A1 |rho|<=0.14; H3000 -0.15..-0.41 on n=28 (more bot fee = fewer swaps). DISCARD (tape lacks nohook pools) |
| 7 | IPO oversubscription: ETH volume into pool before the 300k touch, NHW n=388 | quartiles of volume vs sim result | 1.46/1.42/1.34/1.59, flat. DISCARD |
| 8 | OOS check of nohook with the live radar (non-thin, age<1h, 300k cross) | P(1M) by venue | v3 0.50 (115 non-thin), native 0.25 (28), USDG 0.75 (8), WETH nohook 2 of 2 but both born above. n too small. MATURE only for v3; waits for radar history |
| 9 | score-keeping: born-above vs climbed (inversion) | radar cross events | see headline 1. INFO to manager/research |
| 10 | data request | radar history 30 days with cap_first at true birth | see below |

## Request (ppfeeds, extends #475)
- What: radar-style cross events (schema of /opt/sw/radar/radar.jsonl: pool, venue, token, th, cap, cap_first at the pool's FIRST swap, age_s, depth_usd, thin, n_senders, fee, hooked, quote) backfilled 13 Sep-7 Oct, same file format, /opt/sw/radar/radar_hist.jsonl.
- Test: per venue, number of climbing runners (cap_first < 0.5 x level) at 1M/5M and P(1M|300k, non-thin, age<1h). Pass: a non-v3 venue with >=30 climbers and P(1M) >= 0.5 non-thin; then replay entry at 300k cross. Kill: <15 climbers on every non-v3 venue (then unwatched-pool focus moves to v3 sub-types and hooked pools).