# A1 iterate run 17 (10 Oct 04:00 BKK): ten new lenses on the mc5 cell
Data: /root/fwa1/win/feats.csv (cap300 $250k-1M, n=304 with end4x), Transfer logs of 254 cell tokens via free RPC (ae1.py), creation-tx forensics for 216 cell + 320 outside-cell pools (ae4.py, ae4b.py), r6weth.db, prices.db, GeckoTerminal token info (partial). Outcome x = end4x clipped at 10, halves by t0 median. Scripts ae1-ae5.py in /opt/crew/work/analyst1.

| # | Lens | Hypothesis | Result | Verdict |
|---|---|---|---|---|
| 1 | Holder graph (Transfer logs, new data) | pool share of supply / top-1 holder / deployer share / burn / pre-pool transfers at +300 s predict x | Pool still holds >50% of supply: x 0.99 (n=52, halves 1.05/0.91) vs 2.07, holds in every buyers300 tertile. BUT these are the jump trap: median buyers60 1, $1-13 bought, cap $500k+, x = 1.00 flat (no volume). They lose nothing, we cannot trade them. Other metrics (top1, top5, deployer share, burn, pre-pool transfers) non-monotone. | DISCARD (jump trap, zero PnL) |
| 2 | Creation-tx forensics (to, selector, gas price, nonce) | launcher/gas/nonce of the token-deploy tx separate winners | Direct deploy by EOA (init code in input) n=168: 2.01. Factory-call created n=48: x 1.00, 92% flat (same phantom-cap coins, again no PnL). Inside direct deploys: deployer nonce 0 (first tx ever) n=138 2.18 (halves 2.07/2.28) vs nonce>0 n=30 0.93 (1.13/0.80), also for non-repeat creators (0.84, n=17). Outside the cell (cap <250k or >1M, n=288 direct) nonce0 0.77 vs 0.78: no effect. Gas price top third 2.61 vs 1.96 inside, 0.90 vs 0.71 outside: weak. | WEAK / INFO. Only 19 non-flat nonce>0 trades. Log it in the feed (spec 2), re-test at n>=150 |
| 3 | Kit lifecycle (does a kit show decay before it dies?) | reach-300k/1M share or buyers60 fall in the last days | Kit A reach-1M 36% -> 88% in its last 3 days, kit B ~34% flat while launches/day fell 41 -> 10. Death = volume stop, no quality warning. | DISCARD |
| 4 | Kyle lambda (price impact per $ of buy flow) | thin-depth coins (high lambda) are fleet-pumped | Quintiles x 1.97/1.27/1.33/2.30 (U-shape), halves flip. | DISCARD |
| 5 | Funding-root repeat (operator level, not deployer) | coins from a root that launched before differ | root first 1.82 (n=117), 1-3 prior 1.36 (n=69), 4+ 1.82 (n=118); halves flip. | DISCARD |
| 6 | Orthography / brand impersonation (new meta view) | symbol length, case, digits, non-ASCII, stock tickers, "Robinhood" | Symbol length 5-6: 2.25 vs 1.50 in cell, but 0.68/1.32/1.47 vs rest 0.79/1.29/1.44 in the three bands outside the cell. Stock-ticker/robin names 1.40 and 1.73 vs 1.75. | DISCARD (no replication, 8 tests) |
| 7 | Same-token dual pools = pure arbitrage (not winner picking) | multi-pool tokens trade at dislocated prices | r6weth: 681 of 19,012 tokens have 2+ pools (3.6%); of 24k cross-pool comparisons, same-quote median gap 50% (real for the v4 fee-3000 pairs: ratios 0.13-3.5 between a 290-swap and a 16k-swap pool), but well-arbed fee-100/10000 v3 pairs sit within 0.5%. Reserves of the cheap leg unknown, FCFS sequencer race. | PARKED: small universe, needs reserve depth per leg. Not tested for money |
| 8 | Market regime, contemporaneous (hedge) | outcome moves with ETH/BTC during the 4 h hold so a short hedge could cut variance (Ben: 15 losses while BTC dumped) | Spearman(x, ETH 4 h return) 0.01, BTC -0.03 (n=304); ETH<-1%: 2.10 (n=22), >+1%: 1.01 (n=23); daily means rho 0.03 (19 days). No beta. | DISCARD: nothing to hedge, BTC dump is not a driver |
| 9 | Third-party metadata (GeckoTerminal token info: twitter, website, image, gt_score, dev holding %, holders, launchpad %) | socials at birth = real team | Free API (~10/min, 429 afterwards) has fields for robinhood; only 49 cell tokens fetched in the time: twitter present 6/49, x 0.90 vs 1.59, gt_score rho 0.10, dev holding rho -0.33. Too small, and socials added days later leak look-ahead. | DATA REQUEST: forward snapshot at pool age 60 s (spec 1) |
| 10 | Payoff landscape over entry cap | a second hump exists beyond 250k-1M | cap300 bands (all pools, n=2,579 with outcome): <30k 1.60 (n=1168, 58% flat, fleet family), 60-250k 0.71-0.94 (n=650), 250-350k 1.32, 350-500k 2.06 (halves 3.32/1.65), 500k-1M 1.6-1.8, 1-5M 1.0-1.6 (flat/no volume), 5M+ 1.62. Only known humps. | INFO, no new cell |

## Pipeline output
- Ready to test: none (nothing positive after fees on out-of-sample data with depth).
- Data requests: ppfeeds (1) GeckoTerminal forward snapshot, (2) deployer-nonce and creation-selector columns in the kit/mc5 feed.
- Lessons: (a) three "strong" results (holder share, factory-created, symbol length) were the jump trap or a multiple-test artifact; (b) the cell outcome has no ETH/BTC beta; (c) free Transfer logs give holder structure for any token at 8 calls/s, dataeng does not need to build it.