A losing streak ends, a dead edge does not, and the only way to tell which one you are in is structure: how the current run compares against the worst your backtest ever produced, and which way the last few hundred trades are drifting. Feelings are useless here. Quantprove's Monitor reads that drift across rolling windows of trades and reports it as a Health Score.
Compare the run you are in against the worst run your backtest ever produced. Your record already contains a worst losing streak, a deepest drawdown, and a longest flat stretch. While the live run sits inside those bounds, you are watching weather the system has already survived. Break the bounds, in length or in depth, and you are watching something the record has never seen.
The trap is judging by recency and pain. Five losses feel like a verdict when real money is attached, and most systems produce five in a row routinely. Our demo NQ momentum system carries a 29R max drawdown across its 760 backtest trades, and anyone watching it live without that number in hand would have called it dead somewhere inside that hole.
One honest caveat, because traders get burned by the clean version of this advice: the record is always late. By the time decay is statistically obvious it has already cost you the proof. Structure narrows the window where you are guessing, it does not close it.
A drift, not an event. Winners shrink while losers stay the same size, drawdowns overstay the schedule your backtest set, and the curve spends more and more time flat or underwater. In simpler words... each window of trades comes in a little worse than the one before, and no single week ever announces it.
Markets adapt to the people trading them. A pattern that pays attracts the traders who found it, their orders move the prices that created the opportunity, and the edge compresses until it no longer covers costs. McLean and Pontiff measured this on published anomalies: returns shrink substantially once a strategy becomes public knowledge. Structure does the rest without any crowd, spreads change, volatility rotates, participants leave.
An edge rarely dies in one crash. It usually gets arbitraged into rent money. None of that means your original edge was fake, it means an edge is a claim about a market that keeps moving, which is why a real edge gets rechecked out of sample on a schedule instead of once.
Here is the part that trips people up... one score over the full record can read fine while the strategy is dying, because the strong early run subsidizes the weak recent one, the average flattens the slope, and 400 trades of history will happily hide 150 trades of decline inside an acceptable number. Rolling windows take that hiding place away.
Quantprove's Monitor runs this across a live record of at least 100 trades, scoring windows of 20 trades stepping 8 at a time below 200 trades, widening to windows of 50 stepping 15 past that, then compares the first three windows against the last three. The trend lands in your Health Score. Monitor describes what the record shows; what you do with a fading score stays your call.
A drift you can see at trade 150 is cheaper than one you admit at trade 400.
Check performance against conditions. A trend system in a rangebound market produces losses that look exactly like a dying edge, then recovers fully when the trend returns. In a regime pause the strategy pays when its setup appears and sits flat when it does not. In decay it is weak even on the setups that used to pay, and that conditional read is the same logic behind why live results drift from a backtest.
One bad reading is noise. Several together, holding across windows, is a signature.
| Signal | Normal streak | Decay |
|---|---|---|
| Streak length | Inside the backtest's worst run | New records, repeatedly |
| EV per trade trend | Flat across windows, noisy | Sliding window after window |
| Drawdown duration | Within the backtest's longest | Longer and more frequent |
| Live vs backtest distribution | Stability Score holds at 60 or above | Stability Score erodes |
| Performance in good conditions | Pays when its setup appears | Weak even on preferred setups |
A streak in the left column is also a drawdown question, and depth has its own reading. Before you trust the feeling either way, pull your last 200 trades and run them.