It won under half its trades, and still finished the week up.
Here's one Priorsum paper account across a clean week: six market sessions, June 25 → July 2, 2026, that the engine ran start-to-finish with no outages. Every trade is on the record. Nothing is cherry-picked, the losing days are left in, and the numbers below come straight out of the trade journal.
Account value, at the close of every session.
Real end-of-day account value, straight from the journal. It's a paper account, so this is simulated money, but it's filled against live market data and settled like the real thing. One down day (June 26) is right there in the line; we didn't smooth it out.
Being right less than half the time is fine if you're wrong cheaply.
The system lost 45 of its 107 trades. It finished up anyway, because the average win was a little bigger than the average loss and the winners were allowed to run to target while losers were cut fast. That's the whole game.
The trade ledger
A hair under half its trades won. On its own that number means nothing. A coin does better. What matters is what happened on the wins versus the losses.
Wrong cheaply, right fully
Average win vs average loss:
That 1.15× payoff ratio, times a 47.7% hit rate, is a positive expectancy: the system makes a little on average every time it pulls the trigger. Over 107 trades it added up to +$281.96.
Where the money came from, and where the losses were stopped.
Every closed trade, grouped by how it exited. Take-profit and exit signals did the earning; the stop-loss did its one job: killing losers before they got expensive. It never won a trade, and it was never supposed to.
| Exit reason | Trades | Win rate | Net P&L | Its job |
|---|---|---|---|---|
| take_profit | 17 | 88.2% | +$495.95 | Let winners hit target |
| exit_signal | 68 | 52.9% | +$486.00 | Read the reversal, took the gain |
| stop_loss | 18 | 0.0% | −$652.30 | Cut losers fast, the safety net |
| manual | 4 | 0.0% | −$47.70 | Hand-closed by the operator |
The best strategy, and the worst one, in the same week.
Ten strategies traded this window. Some earned, some bled. We're showing you both ends, because a track record that only shows its winners isn't a track record.
31 trades · 74.2% win rate. Buying oversold dips in ranging names and taking the bounce; the engine leaned into what was working and it paid.
26 trades · 30.8% win rate. Breakouts kept failing in a choppy tape. This is the drag the learning loops exist to notice, and pick less often next time.
Why only these six days, and what we left out
- We only counted sessions the engine ran clean. This is a system under active development. Around this window there were days it was taken down mid-session by an out-of-memory crash, a data-feed stall, and a fill-accounting bug being fixed. Attributing that noise to the strategy would be dishonest in the other direction, so those days are excluded, and named here rather than hidden.
- The losing days that ran clean stayed in. June 26 lost $71. It's in the curve, in the win rate, and in the net number.
- One paper account, one window. This is a single account's result over one clean week, not an annualized return, not a projection, and not a promise about the next week.
16,748 decisions. It said “no” to most of them.
Across the window the engine evaluated the tape thousands of times and mostly chose to do nothing: holding, or vetoing a setup that didn't clear the bar. The 107 trades above are what was left after all that patience.
Every trade came with a plain-English reason.
These aren't marketing copy. They're the actual explanation strings the system wrote to its journal as it closed these three trades. Verbatim.
And it kept learning the whole time: 90 fast-loop weight adjustments fired across the window, nudging the signals behind winning trades up and the ones behind losers down, live, while it traded.
This is last week. Come watch the live one.
Sign in to follow the same system, decision by decision, in real time, the good days and the messy ones, same as this.