# FW Analyst 2: new edge hunt, 8 Oct 2026 (Bangkok)

## 1. NEW EDGE: the 10-second flip on plain v4 launches (buy at +10 s, sell 60 s later)
- Universe: every native-ETH Uniswap v4 Robinhood Chain launch with NO hook, 5 Sep to 7 Oct (day files /opt/hist/rh), 25,274 pools.
- Rule: buy $100 at 10 s after the pool's first swap; sell 60 s later. Skip the 0.01% "pond" pools and the 0.25% tier. Best tiers: 1% and 0.05% fee.
- Priced from REAL trades, no curve model: our buy = average price real buyers paid in the 4 s after entry, plus our own impact; our sell = average price real sellers got +/-10 s around the exit, minus our impact; GMGN 1% each way; a liquidity pull before exit = 0; **no real sell near the exit = 0 (counted as a total loss)**.

| set (conservative) | trades | mean x | median | up | total loss |
|---|---|---|---|---|---|
| 1% tier, Sep 5-19 | 1,079 | 1.24x | 1.09x | 54% | 41% |
| 1% tier, Sep 20-Oct 7 | 7,709 | **1.44x** | 1.51x | 84% | 14% |
| same at $500 a trade | 7,709 | 1.31x | 1.40x | 84% | 14% |
| standard launchpad tokens only (factory 0x58daec3116aae6d93017baaea7749052e8a04fa7), 1% + 0.05%, Sep 20-Oct 7 | 5,709 | **1.50x** | 1.53x | 90% | 10% |
| hold 30 s instead (1% tier, Sep 20-Oct 7) | 7,954 | 1.23x | 1.24x | 92% | 5% |

- Per day (1% + 0.05%, $100): every day from 21 Sep to 7 Oct positive; 27 Sep to 7 Oct between 1.37x and 1.60x on 550 to 725 launches a day. **16 to 20 Sep lost** (0.46x to 0.94x, 41% to 73% of pools had no exit): the regime can flip, so a daily loss stop is required.
- Timing cliff: entry at 5 to 15 s works; entry at 20 to 30 s and holding 1 to 3 min loses (this is why the 30 s random baseline on the honest board loses).
- Ground truth from wallets: outside wallets that bought 5 to 10 s after launch and sold 45 to 120 s later: median 1.57x, 94% up (n 301), and only 9% of those early buyers never sold. Sell taxes: 5 of 13,284 sells show an extra token transfer, so none to speak of.
- Model-free cross-check (1 Oct): real sells at 70 to 90 s paid a median 1.81x the price real buys paid at 15 to 20 s, in 1,015 pools.
- Why it can exist: launch crews run buy bots that paint the chart for the first minutes, then pull (plain v4 median about 5 min). A flip of under 2 min exits before the pull.

### Not proven by data (must be tested live, small)
- Whether crews react to OUR buy (dump or pull on outside size). An older pond test with a 1.3x to 1.5x target lost 22% a coin live (brx/edges-v2.md).
- Latency: we must land the buy by about 14 s after the first swap. GMGN may not route a 10-second-old v4 pool; a self-signed direct swap (SW engine wallet) is safer.
- A sell can fail on a pool that gets pulled inside our 60 s.

### Proposed test (Ben's process)
1. Paper twin on the live chain feed (rhfast) for 1 day: same rule, real-trade pricing, log latency.
2. Then $10 live, self-signed, one position at a time, stop the day after 5 total losses in a row or minus $50.

## 2. KILLED: late entry and multi-day holds (all venues, May to Oct)
- Built 1-minute bars for all 11,703 pools that reached about $100k (every swap, /opt/sw/e1). Grid: cap $300k to $5M, age 0 to 72 h, rising/falling, hold 1 h to 7 days, with and without a 0.5x stop, depth-based fills, rugs = 0.
- Holds of 72 h and 7 days lose EVERY week (0.02x to 0.6x). 24 h holds about 0.5x to 1.0x since August. Buy at $500k to $1M and hold 4 h made money only in the early-July bubble; since August about 1.0x without the best coin. Picked on May to Sep 14, nothing passes Sep 15 to Oct 6.
- New coins crossing $300k a week fell from about 950 (early Sep) to about 450.

## 3. KILLED earlier today: mc5 on v4 (mean 0.17x, 79% pulled). See a2-edges-8oct.md.

Code and rows: research-box /root/fwa2/edge/ (flip.py, agg_flip2.py, agg_flip3.py, wallets3.py, spread.py, trap.py, late.py, build_bars.py, bars_*.db).