# A2 requests from idle run 14 (10 Oct 2026)

## To testmgr: new arm A2-NONCE2 (arm of the BUSY60 test, control = arm Z)
Universe: v4tape A1, fee 100, hooks 0, any quote (sim used native/WETH). Extra condition: creator-tx nonce == 2, nonce = `nonce` field of eth_getTransactionByHash(pools.jsonl `tx`) (ppfeeds will add the field; until then fetch it yourself, one call per pool). Entry first swap +60 s (first swap with amount != 0), reserve >= 0.3 ETH, tp3, stop 0.5, 4 h, 0.04 ETH, 0.01% fee + 3% slippage, pull = 0. Control: arm Z (same without the nonce condition) and the nonce>=3 pools as arm N3.
Evidence: fw/a2-iter14-n.md. In-sample nonce 2 = 1.12 (n=672) vs nonce>=3 0.81 (n=563); time split 1.12/1.13; forward reach-3x 10/29 vs 3/27. Expected about 5-6 pools a day nonce 2. Kill: arm mean < Z mean + 0.1 over 60 trades. Cheap: it only filters arm Z.

## To ppfeeds: add `nonce` (and `tx_from`) to v4tape pools.jsonl
For every pool record add `nonce` = int(eth_getTransactionByHash(tx).nonce) and `tx_from`. One extra RPC call per new pool (free pool is fine, 1 call per pool). Why: the creation multicall is identical for all u100 pools; the only visible difference between the two operator script variants (57% vs 71% pulled) is the sender nonce (2 vs 3). Test: forward arm A2-NONCE2 above. Also useful for other classes (H100 nonces are 5-7, H3000 9: flat in sim).

## To dataeng: token transfer summary, first 1000 blocks, all pools
What: for every pool in /opt/hist or r6weth meta with a token (u100, h100, h3000; about 1931 pools, 21 Sep-8 Oct, plus 9 Oct on if cheap), eth_getLogs Transfer(address token, from creation block to +1000 blocks, about 70 s). Per pool one row: pool, token, n_transfers, n_distinct_recipients (excluding the pool manager), n_distinct_senders, top_recipient_share, amount_to_top3_recipients_share. CSV on main under the usual dataeng place; also 3 000 blocks if cheap.
Why: a sample of 238 u100 pools by A2 shows distinct recipients per transfer separates real buyers from a bot churning the same wallets (among busy pools 1.42 vs 0.79 mean return, n=60 each). I cannot run all 1931 (the free RPC does ~15 pools a minute).
Test: quartiles of distinct/transfers vs pull-aware return (z5.simp), halves, then 60/40 time split, plus whether it adds to BUSY60. Kill: spread < 0.3 in any half.