DQM
analyze a wallet

post-mortem · 4 venues · 2 years

I lost $1,032,315.
The data says it didn't have to happen.

764 consolidated positions across Jupiter, Hyperliquid, Lighter, and Bybit. A 38.5% win rate. This site exists because I trained a risk model on my own wreckage to find out whether I was a bad trader or a good trader making decisions in a bad state. The answer was the second one, and the difference was a million dollars.

FINDING 0131.6% → 2.1%

I paid a novelty tax on every new coin

My mistake rate the first time I traded an asset was 31.6%. After 20+ trades on the same asset: 2.1%. Fifteen times worse, purely from unfamiliarity. Nobody told me the most expensive thing on a perp DEX is a ticker you’ve never seen before.

FINDING 0225%

Red days made me dumber

When my session was already down more than 5%, a quarter of everything I opened was a mistake. Up big? 3.7%. Same trader, same markets, same week. The only variable was how much I wanted the money back.

FINDING 034x

My longs were four times worse than my shorts

14.9% mistake rate on longs vs 3.7% on shorts. I was a disciplined short seller and a degenerate buyer living in the same body, and I had no idea until the data separated us.

FINDING 0438.5% → 52.1%

The skill was there the whole time

Filter out only the trades flagged by behavioral risk — the unfamiliar assets, the oversized impulse entries, the loss-streak chases — and the same history flips profitable. Same entries, same timing. The edge wasn’t missing. It was buried.

the conclusion

I didn't need a new strategy. I needed a mirror.

Every number above is computed from raw fills — no journaling, no self-reporting, no selective memory. The same analysis runs on any wallet, in the browser, in about 30 seconds. If you trade perps, you owe yourself the look I didn't take until it cost seven figures.

Run yours →

methodology: 7-factor behavioral risk model (asset unfamiliarity, sizing anomaly, drawdown state, loss-streak tilt, session state, overtrading, negative momentum) · partial closes consolidated before scoring (1,745 fills → 764 positions) · mistake = bottom-decile PnL · stats above are from the demo dataset, viewable in full via “see an example analysis”.