# A1 iterate run 15 (10 Oct): ten new lenses, one new metadata field (token names)

Scripts: /opt/crew/work/analyst1/ac1-ac14.py (research-box). Names: ac11.py reads name/symbol/supply/decimals of 13,082 tokens via Multicall3 eth_call (free rpcpool, ~1 min). Table: http://159.223.88.168/hx-c9ois/fw/a1-names.csv.gz

| # | Lens | Hypothesis | Result | Verdict |
|---|---|---|---|---|
| 1 | Money arriving (bridge ETH in/out per day, chainday.db) | mc5 cell payoff follows bridge inflow of the day before | 19 days, Spearman -0.42..+0.30, signs flip | discard |
| 2 | Pull contagion | a pull in the family is followed by more pulls within 40 s | share preceded 9 % vs 6-8 % random (u100/v3100), h100 3 % vs 1.4 % | discard (no exit signal) |
| 3 | Portfolio of families | daily payoffs correlate, regime switch needed | h100 2.2-3.0, v43000 2.3-3.0, u100 1.0-1.8 daily, correlations -0.3..+0.5 (n=17 days, noise) | info: families are independent bets, size them separately |
| 4 | Ticket size / buyers in 300 s | tiny average ticket = dust pump | feats cell non-kit: 106 of 156 coins have <=3 buyers and pay 0.89; but in the 202-trade replay only 6 trades have <=5 buyers (mean 0.13) because moved_check already removes them | discard (already filtered) |
| 5 | Entry/exit: buy the dip, re-enter after stop | wait for -10/-20/-30 % after entry; re-buy at entry cap after a stop | dip entry: coins that never dip are the runners (4.09 vs 2.22), per-slot 2.41/1.96/1.51 vs 2.42; re-entry n=55 mean 1.07 median 0.47 | discard |
| 6 | Entry delay / confirmation | enter later, after proof | harness delay 306/420/600/900 s: total mean by half 2.85/2.00/2.00, 3.84/1.96/2.53, 2.53/1.72/3.65, 3.88/1.89/3.08, trades 221/216/203/187 (flat). Confirmation (no -10 % for 5/10 min) n=32/19, ex-top3 1.2/0.47 vs base 1.54/0.59 | discard; info: mc5 is not speed sensitive between 5 and 15 min (a 429 retry costs nothing) |
| 7 | Inversion: 60-120 s rugs | creation features predict early pull | 46 % of u100 and 56 % of v3100 pools are pulled in 120 s (hooked 2 %); only reserve at 3rd swap has AUC 0.32-0.36 (already the reserve gate) | discard (confirms gate) |
| 8 | NEW DATA: token names (clean population, not survivors) | names/metadata carry a signal inside a family | Dexscreener drops dead pairs (1/25 of 2-day-old pools left), so names must come from chain: name()/symbol() via Multicall3 for 13,082 tokens. Within families (u100 1342, v3100 1075, h100 411, v43000 109): length, words, digits, caps, non-ascii, supply, owner: all flat (+-0.15) | no rule, but the field now exists |
| 9 | PROJECT TYPE by name | pseudo-professional names ("... Protocol / Labs / Pad / Finance / Coin / Inu / Token") are the template factory | mc5 cell (n=310): coin-word names n=53 pay 1.04 (halves 0.75/1.23) vs 2.30; kit A 13 of 13 stopped at 0.5; finance/stock names (Robinhood tokens, NVDA...) n=32 2.35 but halves 0.73/3.45; animals n=22 3.0 | mature as tag only (n small) |
| 10 | REPEATED NAME (who is on the other side: serial operator vs one-off amateur) | a name already used by another token in the prior 48 h means a repeating operator | tape: u100 +5 s: dup 302 pools 1.41 (halves 1.49/1.31) vs no dup 268 1.19; at +60 s 1.12 vs 0.84; v3100 1.15 vs 1.04. mc5 cell: dup 131 pay 2.97 (halves 3.04/2.92) vs 179 pay 1.44 (1.07/1.89); non-kit 44 vs 112: 2.49 vs 0.90 but median 0.98 both, 3 winners drive it. In both sets dup is always a DIFFERENT token with the same name (same-token relaunch = 0) | ready for paper as tag/arm (in-sample, thresholds: any earlier token, 48 h) |

## Sent
- testmgr: A1-DUPNAME (tag on mc5 trades and arm on u100 +5 s), exact rule in the board post.
- ppfeeds: add token name/symbol (and supply, decimals) at creation to the kit/v4/v2 feeds, so the dup flag can be computed live (spec in the board post).

## Pitfalls
- Dexscreener only keeps living pairs: any name study from it is survivor-biased.
- Multicall3 (0xcA11...CA11) works on the chain: aggregate3 with 40 tokens x 5 calls per request, no batch limit trouble (plain batch of 80 calls fails with 429/500).
- Name dup must be computed from earlier tokens only, strictly by creation time (done).