The Overfit Museum

A small gallery of the classic ways a backtest lies to its owner. Every exhibit here is made up to teach the pattern — no real traders, funds, or track records. The scores are what Wembro would hand each one, and none of them pass.

These are illustrative teaching examples, not case files. If a strategy you actually trade looks like one of these on the wall, that's the point — run yours through the same honest tests before you size up.

The Perfect Curve

GHOST · 18

An equity line so smooth it looked machine-polished — because it was. Every parameter had been tuned until the past had nowhere left to hide.

What happened

A strategy was re-run hundreds of times, nudging the moving-average length, the stop, the entry hour, the day filter — keeping whatever number made the backtest prettier. The final curve fit the last three years almost exactly. Forward, it fell apart inside a month: the rules described that specific history, not the market.

Over-optimization. If you searched a thousand variations and kept the best one, the best one is mostly luck. Wembro penalizes the number of variations you tried (a Deflated Sharpe), so a result that only shines after heavy tuning scores low — exactly as it should.

The Winners' Circle

GHOST · 24

A basket that 'always went up' — assembled, conveniently, only from the names that survived to today.

What happened

The backtest bought a list of 'great companies.' But the list was written now, looking back. Every business that went bankrupt, delisted, or quietly died had already been deleted from history. The strategy looked unstoppable because it was only ever allowed to trade the survivors.

Survivorship bias. A test that can only pick from things that lived will always look brilliant. Honest analysis judges decisions made with information available at the time — not a roster curated by hindsight.

The One That Paid Rent

GHOST · 29

Ninety-nine ordinary trades and one monster. Remove the monster and the whole 'edge' evaporates.

What happened

The track record was, on paper, deeply profitable. But a single trade — caught in a once-a-year gap — accounted for the overwhelming majority of the gains. The other ninety-nine trades, taken together, barely broke even after costs. The 'system' was really one lucky lottery ticket wearing a hundred trades as a disguise.

One-lucky-trade concentration. An edge should be spread across many trades, not balanced on a single outlier. Wembro flags when one trade carries too much of the result, because what you can't repeat, you can't rely on.

The Eleven Coin-Flips

INSUFFICIENT · 0

Eight wins out of eleven. Impressive — until you remember a fair coin does that more often than people think.

What happened

A trader went 8-for-11 and declared the strategy proven. But eleven trades is far too few to separate skill from a hot streak. Run the same coin enough times and plenty of eleven-flip stretches land on eight heads. The 'edge' was real-looking and statistically meaningless at the same time.

Too few trades. Small samples are dramatic and untrustworthy. There's a minimum track-record length before a Sharpe ratio means anything — below it, Wembro refuses to issue a confident verdict instead of flattering a streak.

The Free Lunch

GHOST · 21

Wildly profitable in the simulator, mysteriously flat in real life. The difference was the bill nobody added up.

What happened

The backtest assumed every order filled at the perfect price with no spread, no commission, no slippage. A high-frequency strategy clipped tiny profits hundreds of times a day — profits smaller than the real costs of trading. Once spreads and fees were charged, the same trades quietly bled the account.

Ignoring costs. Frictionless backtests are fairy tales, and the smaller and more frequent your edge, the more the bill matters. An edge that only exists before costs is not an edge — it's a subsidy you're paying to the market.

The Staircase and the Cliff

GHOST · 27

A near-perfect win rate and a curve that climbed like a staircase — right up to the edge of a cliff.

What happened

Every losing position was doubled down on until it turned green, so the strategy 'won' almost every time and the equity line looked serenely smooth. The hidden cost was buried in the rare bad streak: when the market refused to bounce, the doubling compounded into a single, total wipeout that erased years of tidy gains.

Martingale risk. A smooth curve and a high win rate can hide catastrophic tail risk. Wembro simulates risk-of-ruin and worst-case drawdowns precisely so a strategy can't hide a cliff behind a pretty staircase.

Is your edge a real one — or another exhibit?

Paste your trades and get one honest Edge Score. Same brutal math we used on every painting in this room.

Score my strategy — free
Hypothetical, impersonal statistical analysis — not investment advice, no guarantee of future results.