Methodology
A good-looking backtest is cheap.
Try enough ideas on enough history and some of them will look excellent by pure chance. That is not a flaw in the search — it is arithmetic. The only thing that separates a real edge from a lucky one is what happens after the pretty curve appears, and that is the part this page is about.
What we hold to
Six rules a candidate has to survive.
None of these make a strategy work. They make it harder for a strategy that does not work to look like one that does.
01
Judged on data it was never tuned on
A strategy is fitted on one stretch of history and then scored on a later stretch it has never seen. The score that counts is always the one from the unseen part.
02
Discounted for luck — and for how many variants were tried
A result is not taken at face value. It is reduced to account for the noise in the sample, and reduced again to account for how many alternative settings were tested before that one came out on top.
03
The more we search, the higher the bar
Searching harder produces better-looking winners by construction. So the acceptance threshold moves with the size of the search: a candidate found after thousands of attempts has to clear a taller fence than one found after ten.
04
Out-of-sample proven, not curve-fitted
The test is repeated by rolling forward through history, so a candidate has to hold up in several different market periods rather than in one flattering window. Settings that only work at the edge of the allowed range are treated as a warning, not a find.
05
Worst drawdown shown, never hidden
Every record states the deepest loss it lived through, how long it stayed underwater, and how often it traded. What was not measured is left blank rather than filled in with something flattering.
06
Most candidates are rejected — and that is the point
The normal outcome of the process is rejection. A run that produces nothing publishable is not a failed run; it is the system doing the job you are paying it to do.
The trust ladder
Nothing is promoted by opinion.
Every record carries the rung it reached and the reason it has not gone further. Rungs are earned by evidence — there is no manual override to the top.
- 00
Incubating
A candidate that cleared everything except one criterion. It is kept visible, and kept separate: interesting enough to watch, not proven enough to publish.
- 01
Validated
Cleared the research review: reproducible, evidence-backed, deployable. Still private to its owner — clearing research is not the same as earning a public listing.
- 02
Publish approved
Cleared the publication review and is listed on the public shelf, where anyone can read the record behind it. This is the public floor.
- 03
Highlighted
The top public tier. It is granted by the evidence itself when the record is created — nobody, including us, can hand a strategy up to this rung.
After the verdict
A record is not finished when it is published.
Evidence keeps arriving after a decision is made, and a decision that never looks at it again quietly becomes a claim about the past.
One slice of history stays sealed
A period is set aside before the search begins and is not looked at while the rules are being chosen. When it is finally opened, what it says is written onto the record — it is a later reading, not a second chance to qualify.
A verdict carries its date, and is stamped again
Every decision records when it was taken and which thresholds were in force that day. As a record accumulates more out-of-sample evidence the decision is re-taken, and a record whose evidence is too short to decide on is held as exactly that — undecided — instead of being passed or failed.
Numbers carry the version that produced them
Every measurement is stamped with the version of the engine that produced it. When we find and fix an error in that engine, older numbers are marked as awaiting re-measurement rather than quietly compared with new ones.
What none of this promises
Backtests are not forecasts. They describe how a set of rules would have behaved on history that already happened; they do not tell you what the next month holds.
Past performance is not indicative of future results. Every strategy on this platform can lose money, including ones that cleared every gate described above. Markets change, and a rule set that survived the past can still fail the future.
Everything starts on paper. Simulated trading is the default, going live is a separate and deliberate decision, and your exchange keys stay yours. Nothing here is investment advice, and we do not manage money on your behalf.