you've run the optimizer for hours and every candidate trips one wire. the profit factor looks good but the out-of-sample columns are noisy. you fix the OOS and the profit factor collapses. or the strategy health card flags concentration risk or performance decay. nothing passes everything at once.
first thing to know: you're not doing it wrong. that's the tools doing their job, filtering out setups that would've lost you money live. here's how to push the search further, which constraints you can bend, and how to tell when the honest answer is "there's no tradeable edge here right now."
you can't max every metric at once
a strong profit factor, clean out-of-sample results, a healthy trade count, and low concentration risk rarely all show up in the same setup. they pull against each other.
tighten your filters to lift the profit factor and you cut your trade count — a smaller sample, which makes the OOS columns noisier. loosen the filters for more trades and the edge dilutes, so profit factor drops. chase the highest backtest number and you drift toward a setup fit to noise, which is exactly what the out-of-sample check and the health card are built to catch.
so "every candidate fails one thing" isn't a wall you built. it's the normal shape of the search. you're not looking for a perfect setup. you're looking for the best honest tradeoff.
which constraint you can bend, and which you can't
some of these are hard lines. some are just preferences. know the difference before you compromise on anything.
never trade a setup that fails these:
out-of-sample that falls apart — an OOS win Δ down 10%+ or an OOS PF Δ down 0.5+ means the edge didn't survive data the optimizer never saw. that's overfit, full stop.
a concentration-risk flag on the health card — most of the profit is coming from a handful of trades. one different week and the edge is gone.
a lower number here can still be a setup you'd trade:
a moderate profit factor — a 1.4 that holds out-of-sample beats a 2.5 that doesn't. steady beats spectacular.
fewer trades, within reason — as long as you clear roughly 30-50 trades in the sample, a lower frequency is fine. below that the stats get shaky.
the rule of thumb: protect out-of-sample stability and low concentration risk no matter what. let profit factor and trade count settle wherever an honest setup lands.
widen the search before you call it
before you decide there's no edge, make sure you actually looked. most "I can't find anything" runs were only searching one narrow corner.
run more than one instrument — MES, MNQ, MYM, and M2K all behave differently. an algo that's flat on one can have a clean edge on another.
test more than one timeframe — the same ORB on a 5m opening range vs a 15m opening range are two different strategies.
try a different algo — ORB, IB, and engulfing each want different conditions. if breakouts are dead right now, an IB retracement might still have an edge.
lengthen the lookback — more history puts the setup through more market conditions and steadies the out-of-sample columns.
move the session and weekday filters — cutting a chop-heavy day or a slow session is often the difference between noise and an edge.
run standard first and wide to see what kinds of setups even rank, then switch to holdout on tighter ranges around the winners. the full workflow is in understanding holdout, in-sample, and out-of-sample in the algo optimizer.
when the honest answer is "no edge right now"
sometimes you widen everything and still nothing holds up. that's not a failure of the tool or of you. it's information.
markets go through stretches where a given pattern on a given instrument just doesn't pay. forcing a setup to exist during one of those stretches is how you end up trading the most overfit configuration in the whole table.
a negative result is a real result. "this algo isn't tradeable on this instrument this month" saves you the money you'd have lost proving it live. park it, check back when conditions change, and put your capital where an edge actually shows up.
this is the honest part of how the platform works: the algos are strategy templates you configure, not money printers. some configurations won't clear the bar, and the tools telling you so is them working as designed. more on that in what to expect from algos.
why we don't just hand you settings
the most common ask when the search gets frustrating: "just give me updated setups to start from." here's why edgeful doesn't publish one set of numbers to copy.
a setup that's right for one person's account, risk tolerance, and instrument is wrong for the next. hand the same numbers to everyone and you've created the most crowded, most overfit config on the platform, and the moment the market shifts, everyone's holding the same broken setup.
what does help you start faster:
the playbooks — the free ORB playbook walks through the exact process of finding settings for the 5-minute ORB on ES: every setting and why it's there. it teaches the method, not a magic number.
refine-algo calls — André runs live sessions walking through the optimizer on real setups. ask in an in-app chat or check Discord for the next one.
Discord — members share what's working for them as starting points you can test yourself.
start from those, then make the setup yours in the optimizer. the settings that matter are the ones you validated and can reproduce on the exact setup you'd actually trade.
when to reach out
open an in-app chat or book a refine-algo call if:
you've widened instrument, timeframe, algo, and lookback and still can't get a setup past the health card
the OOS columns or health-card flags are behaving in a way you can't interpret
you want a second set of eyes on a candidate before you push it live
bring the algo, instrument, timeframe, and a screenshot of your results so it's a fast conversation.