A missed meme-coin rally is a question, not a backtest
A token surges after an agent passes it over. The chart is vivid; the question is harder: did the system miss a repeatable opportunity, or are we studying the one winner that survived a field of failures? Today's research question is how an agent should learn from a missed meme-coin rally without letting hindsight rewrite its rules.
Method
We would start at the decision timestamp, not at the later price peak. For every discovered candidate, preserve the data then available: market age, liquidity, recent price and volume, safety evidence, intended position size, quoted buy route, quoted full sell route, and the reason for accepting or rejecting it. Freeze a baseline rule and propose one challenger change, such as a shorter observation window. Run both on the same subsequent candidate stream. Record every candidate, including those that never rallied.
The primary outcome is a completed, cost-adjusted round trip at the proposed size. A paper signal is labeled paper; a shadow observation is labeled shadow; a funded outcome requires actual buy and sell receipts. A peak price after rejection is an opportunity description, not a realized return. We would also count false entries, unsellable positions, missed candidates, and the delay between discovery and an executable quote.
Evidence
This design responds to two established problems. Bailey and colleagues' research on backtest overfitting explains why selecting a rule after seeing many historical outcomes can make chance look like skill. Novy-Marx and Velikov's study of trading costs shows that costs can materially reduce apparent strategy performance. The SEC staff's 2025 statement on meme coins describes values driven largely by demand and speculation, a reason to treat a striking chart with care. These sources motivate the test; they are not evidence that our own agent has passed it.
Limitations
Historical quotes may not represent a fill, especially when liquidity shifts during a rapid move. Candidate collection may miss tokens entirely. Contract checks, taxes, slippage, and transaction failures can make an apparently attractive entry impossible to exit. A funded sample may be too small to distinguish a genuine improvement from luck. We have no verified comparative result for the proposed challenger to report here.
Next experiment
Choose one narrowly defined change, register its entry and exit rules before the next observation window, and keep the baseline running alongside it. Audit the complete candidate log and every closed outcome after costs. If the challenger catches more rallies but also buys more collapses, publish both sides. The point of the journal is to make that trade-off visible to readers, including when the answer is uncertain.
