a long red stretch doesn't mean your algo is broken. most of the time it means the algo is doing exactly what you tested it to do. you're just watching the part of the curve nobody screenshots.
here's how to tell the difference between normal variance and an actual problem, and what to do about each.
a losing streak isn't a broken algo
every strategy that wins over time still loses in clusters. that's not a flaw — it's what a win rate under 100% looks like up close.
if your backtest showed, say, a 60% win rate, then 4 or 5 losers in a row is going to happen, repeatedly, over a large enough sample. the backtest already lived through streaks like that. the equity curve you liked was the sum of them.
the mistake isn't the drawdown. the mistake is changing the algo in the middle of one, turning a normal losing streak into a permanent decision.
you already know your algo's normal range
you don't need a gut feel for what's "too far." the optimizer already told you, and it's worth writing down before you go live.
when you validated the setup, two numbers on the strategy health card describe your normal downside:
max drawdown: the deepest peak-to-trough dip over the tested period
consecutive losses: the longest losing streak in the sample
if your live results are inside those numbers, you're not in trouble. you're in the part of the distribution you already agreed to trade. that's why applying algo settings changes puts it this way: only deploy a setup whose "max drawdown and consecutive losses are sizes you're comfortable holding through." the point of picking a setup you're comfortable with is so you don't panic when it happens.
why live will never match the backtest exactly
even a healthy algo drifts a few percent from its backtest live, and knowing why keeps you from "fixing" something that isn't broken.
the usual reasons, covered in full in why your backtest, optimizer, and live results don't match:
send-to-fill latency: your broker fills on the next tick after the webhook, around 300ms on average
stop-loss slippage: a stop sends a market order, and on losers that slippage always hurts, so live exits usually look slightly worse than the backtest
commissions: the optimizer reports gross P&L; your broker's per-round-trip fees come off the top live
contract rollovers and decimal precision: small differences that add up over a lot of trades
treat the backtest headline as the chart's version of the trade — not a forecast. the gap to live is usually a few percent, not a rewrite.
when to leave it alone
leave the algo running, untouched, when all of this is true:
the setup passed all 7 checks on the strategy health card before you deployed it
your live drawdown and losing streak are still inside the max drawdown and consecutive-losses numbers from the backtest
nothing structural has changed: same contract, same session, same timeframe
in that situation the data is behaving. changing settings here is just reacting to noise, and it throws away the tested edge you started with.
when to actually change something
edgeful doesn't publish a magic number. there's no "re-optimize every 2 weeks" or "stop after 6 losses" rule, because the right answer depends on your setup, not a calendar. instead, change something when the setup leaves the envelope you tested:
live drawdown pushes past the max drawdown your backtest ever showed
your losing streak runs longer than the worst consecutive-loss stretch in the sample
something real changed underneath the algo: a contract rollover, a session time, or a market regime that no longer looks like your test window
when that happens, the move isn't a tweak — it's a fresh optimization. re-run it, re-validate against the health card and the out-of-sample columns, and treat the result like a new setup you have to earn confidence in again. do edgeful algos have verified or published performance? says it plainly: past results don't guarantee future ones. a setup that worked over the last year can still struggle in a new market regime.
what changing settings doesn't fix
a setup that failed the strategy health card on the backtest isn't going to work live, and no amount of trading it small changes that. there's no version of "I'll just size down" that fixes an overfit setup.
so if you're tempted to change settings, first ask which problem you're solving. normal variance inside your tested range isn't a problem to solve — it's the cost of the edge. overfitting is a problem, and it gets solved before you deploy, not during a drawdown.
the honest part
getting an algo to perform takes customization, testing, and time. the tools hand you the raw material. turning it into something you'd trade with confidence is work you put in, and it's worth doing before you go live.
that's also why edgeful doesn't hand you a single "verified return." the algos are strategy templates you configure — so the performance depends on your settings, not ours. the numbers that matter are the ones you produced and can reproduce on the exact setup you'd actually trade.
the algo analyzer is built for exactly this: Monte Carlo simulations, a strategic health score, and a prop-firm pass-rate projection so you can see the range of outcomes a setup can produce before a live drawdown surprises you.