How to read a backtest.

Every trading scam fails at least one of these ten checks. Most fail at three or four before you even look at the numbers.

CHECK 1

Selection bias

Was the config chosen after seeing the results?

A "PF 2.21" is often the best of a hundred configs, picked after seeing the out-of-sample — lookahead bias. The honest walk-forward figure was 1.65. Ask how many configs were tried, and whether the winner was chosen on in-sample or out-of-sample.

CHECK 2

Fills

Does the strategy assume a fill at the exact trigger price?

Breakout strategies with gap-throughs get filled worse than the backtest assumes. One strategy's first pass was PF 3.64 with 100% winning months; honest fills cut it to 1.49. Sharpe above 3 — or 100% winning months — means audit the fills, not applause.

CHECK 3

Survivorship

Is this the survivor of many attempts?

A seller runs ten accounts and shows you the three that won. The "verified" track record is the top of a distribution you can't see the rest of. Ask to see the losers. If they can't show them, that's the answer.

CHECK 4

Lookahead

Does the model see the future?

"Self-trains on each new candlestick" is retraining on the data it's predicting. Same with filtering today's trades on today's close. If the strategy uses a signal that didn't exist at decision time, it's cheating.

CHECK 5

Sample size

How many trades?

PF 5.91 sounds extraordinary until you learn it's 15 trades over 2.5 years. A profit factor on a tiny sample is noise wearing a number. Under ~50 trades it's a hypothesis; under 15 it's a rumor.

CHECK 6

Costs

Are slippage and commission in?

Pre-cost and post-cost are different claims. One revalidation dropped from PF 1.62 to 1.47 adding real costs — "real edge" became "flat." If costs aren't stated, assume they're missing.

CHECK 7

The negative claim

Is the headline about what's absent?

"No martingale, no grid" is not a profit claim. It's a statement about what the strategy doesn't do — and it's only a selling point because everyone learned grids blow up. When the headline is about what's absent, the profit claim is absent too.

CHECK 8

Out-of-sample

Where's the untuned window?

In-sample results prove the strategy fits the past. Out-of-sample — a window the parameters never saw — is the only test that means anything. A strategy that wins in-sample and dies out-of-sample is the rule, not the exception.

CHECK 9

Demo vs live

Is the "track record" a simulation?

A "performance monitor" on a demo server is a simulation — restartable, seedable, no real fills. Live, third-party-verified, auditable capital is a different category of evidence.

CHECK 10

The disclaimer

What does the fine print actually say?

Every scam carries its own verdict, unread: "This EA cannot profit on live accounts." "Past performance is not indicative of future results." In a subscription product that's not legalese — it's the business model stated plainly. Read it before the sales page.

The meta-rule: every check is one question — was the result generated the way it'll be traded, on data it wasn't tuned on, with the losers shown, after costs? If any answer is no, the backtest is marketing.

A strategy, an EA, or a claim — tested and scored, every week. No selling, ever.