A 74 percent win rate is not a result
Before Rōvn, I built a futures trading system. Selected configurations of it backtested at a 74 percent win rate with two-to-one reward to risk across roughly fifteen years of data. That sentence is true, and on its own it tells you almost nothing.
I know this because I spent years learning what the number hides.
A win rate without a sample size is a coin you flipped a few times. Without commissions and slippage, it's a chart, not an account. Without maximum drawdown, it leaves out the month that would have ended you. Without a clean split between the data you built the rules on and the data you tested them against, it's a description of the past dressed up as a forecast. And if you can't point to the line between what the research did and what a live account did with money, you haven't finished the work yet.
So when I say 74 percent, I say it with all of that attached, or I don't say it.
What the number cost
The hard part of BOVYN was never writing an indicator. The hard part was translation. I could look at a chart and know a setup was there. The software couldn't know anything I hadn't defined. Every condition had to be written down. Every invalidation had to be explicit. Every entry, every filter, every stop and target had to become something a program could evaluate on a bar close with no one watching.
Then I ran those rules against about fifteen years of futures history. Not one good year. All of it, regime by regime, including the stretches where the idea I loved most simply didn't work.
Most of what I tried died in that test. That's what the test is for.
What survived became seventeen engines that vote, a consensus threshold that decides whether a signal ships at all, a risk layer that sizes positions from account risk and stop distance instead of from how sure I feel, and a guardian that pauses after consecutive losses and kills the session at a limit. I deployed it on a server, connected it to funded accounts, and then more than two hundred people paid for it every month.
The 74 is the residue of that process. It's not the process.
Why I still bring it up
Because the same failure shows up everywhere, and I'm now working in an industry that runs on it.
Healthcare flattens a person into a single word. "Credentialed." A license says one thing. A résumé says another. A primary source says a third thing, on a specific date, for a specific purpose, with specific limits. The industry collapses all of that into one status, and then, because nobody trusts the collapse, every new employer rebuilds it from scratch. A nurse in my family lived that. She was qualified and she was stuck.
A win rate without its methodology and a credential without its provenance are the same mistake. Both take something that only means something in context and present it bare.
The fix in trading was to refuse to publish the number without the sample, the costs, the drawdown, and the out-of-sample line. The fix in healthcare is to refuse to flatten. Evidence should carry its source, its date, its purpose, and its limits with it wherever it goes. A model's output is not a fact. An upload is not a verification. One organization's decision is not another organization's decision.
That is how Rōvn is built: evidence travels with the person and with consent, and decisions stay with the people who are authorized to make them. When we describe what the product does, we say what exists, what's planned, and what's proven with a date. I learned the price of confusing those three in markets, where the penalty arrives the same afternoon. In healthcare the penalty is slower and it lands on someone else.
So, 74 percent. Here is what it needs. Ask me for the rest.