Results so far (September 2026)
We report negative results as carefully as positive ones. All numbers below are paper results (simulated fills on real chain data), unless stated otherwise. "Multiple" = money out / money in for one round trip, after all costs; 1.00 is break-even.
1. Strategy Lab — signals from a public Telegram bot (7–10 Sep)
- 1,086 signals, 300+ strategy configurations with realistic costs.
- No configuration was profitable after costs. Mature tokens showed +6–9% gross, which commissions, gas and slippage consumed.
2. Tracked-wallet signals on Robinhood Chain (11–16 Sep)
Signal: the k-th distinct wallet from a private list of 1,132 tracked wallets buys a token.
- A grid of 739 strategies × k = 2…6 (3,695 variants) ran on paper in real time.
- A selection protocol was written down before the out-of-sample recomputation
(both periods above 1.00, bootstrap lower bound above 1.00, no bad days, parameter plateau,
≥90% executability). Recomputation used simulated real transactions (
eth_callat the block after the signal and N seconds later), not heuristics. - 15 Sep: no configuration passed → NO-GO. Launchpad fee measured at 4.7% per round trip.
- 16 Sep: gas per round trip had been an unmeasured assumption (0.0004 ETH). Measured value: 0.0000104 ETH (median of 250 buys + 250 sells, cross-checked against sender balance changes on 12 transactions: 0.0% difference). This was a 40× overestimate.
After the correction:
| Population | Multiple (lower bound) | n |
|---|---|---|
| Early tokens (curve raised < 1 ETH), k = 2, hold 120 s | 1.051 (1.017) | 913 |
| Same, placebo entry 1–10 min later | 0.918 | 654 |
| All tokens, hold 120 s | 0.978 (0.944) | 817 |
- One configuration passes the original protocol: k = 2, 120 s hold, early tokens only, 0.01 ETH size — period means 1.059 and 1.029, lower bound 1.021, 3 of 4 days positive, worst day 0.99, best trade contributes 1 %.
- It fails one condition of the stricter addendum: margin over placebo ≥ 0.10 in each period (+0.082 in the early period, +0.153 in the recent one).
- The whole population still does not pass; the median trade is 0.93 (41% of trades profitable) — profit comes from the right tail. The gas correction applies equally to signal and placebo, so the finding "the signal carries information" does not depend on it.
Verdict: a candidate, not a proven strategy. Next step is a small, capped live test to check that real fills match the paper model.