# h3n operator count (#709), testmgr2, 10 Oct
Source: /root/tm/replay/h3n_nat2/trades.csv arm A, pool -> created_tx (ppdata r6weth/r6nat/r6tape meta) -> tx.from (dataeng creators_tx.db). All trades resolved, none missing.
- Arm A all: 233 trades, **227 distinct tx.from**. Max 3 pools per operator (2 operators with 3, 2 with 2; those repeats are ~0x loss pools, mean 0.00-0.03x).
- Arm A reserve>=0.1 (the 4.99x subset): 168 trades, **168 distinct operators**. Mean of per-operator means 4.99x, median 4.94x.
- So the 80/168 winners are NOT one operator or a few; independent samples. Same-template launches come from many different deployers (the hook is a shared launchpad template, each hook-unique by the rule).
- Caveat: tx.from is the tx sender; a funder-level cluster (same funding wallet) is not checked. Hook is unique per pool by design, so a single scripted deployer would show as repeated tx.from; it does not.
Still open for the auditor: sellability at exit (waits on Ben's $5 sell test), hand rebuild of 3 fills, replay vs live overlap. Funder clustering can be done if dataeng has cfund.db coverage (/opt/sw/dataeng/cfund.db) - say so if you need it.