A profit factor of 1.3 is what a real edge looks like
In part one we showed how fill artifacts inflate backtests, and that after re-running everything on real order mechanics, two edges survived out of everything we traded, tested or mined. Their profit factors: 1.30 and ~1.1. This post is about why those unimpressive numbers are the impressive ones — and what a thin edge actually demands from you on a prop account.
The numbers that survived 18 years
| Edge | PF (real fills) | Years positive | Sample |
|---|---|---|---|
| Overnight session-conditioned long, index futures | 1.30 | 18 / 19 | N = 3,367 |
| EU-session flow read → day direction | 1.10–1.16 | 8 / 8 (2019–26) | multi-year, both indices |
| Both combined, 2016–2026 portfolio test | 1.14 | 10 / 11 | worst year: −$879 |
Compare that with the candidate our data-mining sweep loved the most: +$166/trade, t = 4.9 in training (2008–17). Out of sample it collapsed to +$44, t = 1.4 — statistical noise. A four-handle t-stat, gone. That is what magnitude without year-count consistency is usually worth.
The pattern we now trust: real edges show up as small numbers that refuse to die — positive in 18 of 19 years across regimes (2008 crisis, ZIRP, 2020, 2022 hikes), on thousands of trades. Fake edges show up as big numbers that live in one regime, one symbol, or one fill assumption.
Why almost nothing real is above ~1.5 intraday
On liquid index futures, every systematic dollar is being arbitraged by people with better infrastructure than you and us. What's left for bar-level strategies after spread, slippage and commissions is structural scraps: behavioral session effects, flow persistence, overnight risk premium. Scraps are real — our overnight edge cleared a 3× friction hurdle on 3,367 trades — but scraps are thin. When a backtest hands you PF 3 on a liquid future at bar resolution, the polite explanation is a fill artifact. We wrote part one about learning that with our own money.
What a thin edge demands on a prop account
Here is the part most algo traders meet only after they've paid for several evaluations. A PF 1.3 system is a coin with a small bias. Its monthly outcome is dominated by variance, not by the edge. On a personal account that's a patience problem. On a prop account it's a rules problem, because the rules are written against exactly this shape of system:
Daily loss limits cap how much variance you may express in a day — two correlated stop-outs can spend 90% of a $2,000 trailing budget in one session if you sized both legs like the backtest told you to. Trailing drawdown turns every losing streak — which a PF 1.3 system will produce, repeatedly, while perfectly healthy — into a potential account death. Consistency rules punish the very hot streaks that make thin edges pay: one +$1,300 day against a $3,000 target with a 40% consistency cap silently raises your real target to $3,250.
None of that is strategy. All of it is arithmetic — per firm, per account, per day, interacting with your stop sizes and your legs' correlation. Get it wrong and a genuinely profitable system fails the eval; get it right and even a thin edge can be walked through the rule maze with its expectancy intact.
The uncomfortable conclusion
After two months of gate-testing, the hierarchy of what matters surprised us:
1. Fill realism decides whether your edge exists.
2. Rule math decides whether your edge survives a prop account.
3. The strategy — the part everyone obsesses over — decides the rest.
We build for layers 1 and 2: broker-truth accounting, per-firm rule engines, guard layers, and pass-lock — flatten and halt the moment target, consistency and minimum days are all satisfied, so a pass can't be given back. Our own bots trade behind that layer in public, thin edges and all, losses included.