Published
My verdict function is three lines of code. A pair/family row only gets stamped LIVE_READY if it clears my minimum sample size and profit factor >= 1.2 and win rate >= 45% and net return > 0. Miss any one of those and it's NO_LIVE, full stop, no partial credit for the other two. I wrote that rule to keep weak setups out of live rotation. I didn't expect it to reject one of the best-sampled rows in the whole report.
The row that failed on one metric out of three
MEGAUSDT Β· range_bounce ran 301 real trades in my latest pair-hotness pass -- more than every other row on the page except one. Profit factor: 1.268, clears the 1.2 floor. Net return: +35.13%, comfortably positive, the best net result in the entire report. Win rate: 43.5%. The floor is 45%. That's it. That's the whole rejection -- a single metric missing a hard cutoff by 1.5 percentage points, on a sample ten times larger than most of the rows that passed.
| Pair / family | Trades | PF | Win rate | Net % | Verdict |
|---|---|---|---|---|---|
| MEGAUSDT Β· range_bounce | 301 | 1.268 | 43.5% | +35.13% | NO_LIVE |
| BILLUSDT Β· volume_reversal | 61 | 1.456 | 45.9% | +12.19% | LIVE_READY |
| 1000000MOGUSDT Β· volume_reversal | 136 | 1.526 | 46.3% | +23.44% | LIVE_READY |
| LLYUSDT Β· range_bounce | 65 | 2.969 | 69.2% | +31.90% | LIVE_READY |
Line that table up and the thing that stands out isn't the winners -- it's how close MEGAUSDT sits to all three of them on everything except the one metric that matters for the gate. Its profit factor (1.268) beats BILLUSDT's floor-clearing minimum by less than BILLUSDT's own margin above 1.2. Its net return (+35.13%) is nearly triple BILLUSDT's and more than the other two combined. Its sample (301 trades) is 2.2x to 4.9x larger than any of the three that passed. The only place it loses is win rate, and it loses by a margin smaller than the gap between BILLUSDT's 45.9% and the 45% floor it just barely cleared.
Why a win-rate floor is a worse filter than it looks
A fixed win-rate cutoff treats 43.5% and 45.9% as categorically different outcomes -- one gets a strategy live, the other gets it buried -- when the actual gap between them is inside normal sampling noise for anything under a few hundred trades. Flip four trades from loss to win out of MEGAUSDT's 301 and its win rate crosses 45% on its own, no change to the underlying edge. My rule doesn't know that. It just reads one column and stamps a label. Profit factor and net return both already encode win rate implicitly -- a strategy can't post +35% net on 301 trades with a bad risk/reward unless something in the win/loss mix is working. Gating on win rate again, as a separate hard floor, rejects rows that the other two metrics already vouched for.
What I'm changing because of this row
I'm not dropping the win-rate floor -- a strategy that wins 20% of the time and only survives on a few fat trades is exactly the kind of thing I want flagged, and MEGAUSDT's own 48.1% skip rate and a 108-trade losing streak inside that same run are reasons for caution on their own. But a row that clears PF and net by a wide margin and misses win rate by under 2 points shouldn't get the identical NO_LIVE stamp as a row that fails on all three. I'm adding a borderline tier between the two: if win rate sits within 3 points of the floor and the other two metrics clear it comfortably, it gets a second look instead of an automatic reject. Readers running their own backtests can sanity-check this same trap with the win rate calculator and the breakeven win-rate calculator -- plug in a strategy's actual risk/reward and see how many percentage points of win rate separate "funded" from "filed away," because on this row it was 1.5.