# FW Analyst 2, round 4: 10 iterations, first principles, winners vs rest (8 Oct 2026, evening Bangkok)

## The finding that matters most
Every way of entering Robinhood memecoin launches OUTSIDE v3 loses money for outside buyers, and the money flows to creators and launch operators:
- v4 launches (2.3M positions): outsiders get back 0.61-0.74x at every entry time from 10 s to 15 min.
- Pons bonding curves (2.3M positions): outsiders' average position 0.60-0.94x at every entry time; creators 1.8-3.6x median.
- GMGN hot list (first 2 min after a coin appears): 0.67x. GMGN's own "smart" wallets on v4: 0.76x; their followers 0.64-0.70x.
The one place outside money is verifiably paid is v3 WETH pools with the live mc5 rule (live fills = model, live 20x and 30x).

## Iteration results
| # | angle (first principles / gap) | result |
|---|---|---|
| 1 | early crowd mix (outsiders vs busy wallets known at T) | no bin above 1.0x in both halves; KILLED |
| 2 | winners vs rest on every launch-time field: LP range, lock, creator history, X handle/followers/reuse, site, text, solid, deepread, CASIG, lane, fee, hook family, factory | winners (outsiders net positive) are 1-2% of coins; best groups 0.8-0.9x; legitimacy fields only exist from 4 Oct (thin); KILLED for now, retest when day files pass 7 Oct |
| 3 | take-profit proven fillable by outside sells | outside sells after entry are rare; no cell >= 1.0x; KILLED. Exposed a method gap: "outside = in < 20 pools" also excludes skilled real traders; the right yardstick is GMGN-routed swaps (needs v4 swap sender, PP request R4) |
| 4 | tokenized stocks (NVDA, TSLA, AAPL, MSFT, AMZN, GOOGL, SPY, GLD, COIN, GME, MU, COST, IBM, HIMS, KLAC, SPCX) trade 24/7 on chain while the stock market closes | structurally rug-free and positive-sum if the chain price drifts from the real price; NOT TESTED: needs main-pool swap history + reference stock prices (PP request R3). OPEN, research |
| 5 | GMGN hot list first appearance | buyers in the first 2 min: 0.67x (33 coins overlap); KILLED (small n) |
| 6 | GMGN smart / KOL wallets, and following them | smart 0.76x, KOL 0.67x, followers 0.64-0.70x; KILLED |
| 7 | size the proven edge by pool depth | needs pool WETH at entry in the per-trade file; handed to the mc5 owners as a sizing study |
| 8 | Pons bonding curves (no LP to pull) | outsiders lose at every entry time; creators win; 1.1% graduate; KILLED |
| 9 | the live mc5 rule on every venue, May-Oct (1-min bars, 6/7 of pools) | v3: May-Aug 0.88x, Sep 1-21 2.12x, Sep 22+ 3.15x (no top 3 2.38x). v2 and v4 dead now. THE LIVE EDGE IS A REGIME THAT STARTED IN SEPTEMBER |
| 10 | regime gate on mc5 (trade only while the last 20 closed mc5 trades average >= 1.0x) | May-Aug -$3,257 -> -$135 at $100; Sep-Oct keeps $23,600 of $28,213 (84%); PROPOSED for the live bot |

## Ranked by expected profit, after the real-money check
1. Live mc5 on v3, scaled up (proven with real fills). Biggest lever = size.
2. A2-MC5GATE: regime circuit breaker on the live mc5 (protects #1 if September's regime ends).
3. A1-ADD (FW A1): second buy on mc5 coins at $700k; same v3 coins, so likely real; needs v3 tx.from for its outside money test (PP request R1).
4. A2-STOCK: tokenized-stock price gaps vs the real market. Unknown size, structurally sound; needs data (R3).
5. Live test 2 (AI picker): paper with the outside money brief.