// Testing
How we test our own system.
Five stages before a program may touch money — every one is allowed to stop the process. Scroll.
01 · Red first
A test only counts once it has been red with the fault present.
First the test, then the repair. Leaving the field on the task list empty means nothing was proven, only claimed.
02 · A frozen comparison run
Two runs with the same input must be bit-identical.
A different input has to produce a different result. If the comparison file is missing, the test fails; it is not skipped.
03 · Searching against chance
Search on the earlier part, confirm on the later one.
The boundary zone (embargo) is dropped, costs and slippage sit inside the figure. The same search runs over shuffled series with no relationship left and over one with an effect built in.
04 · Shadow operation
A candidate first runs along without money.
Under the same locks that decide about the signal: regime filter, repetition stop, only one open position.
05 · Monitored continuous operation
A process being listed proves nothing.
Each cycle writes its own heartbeat. A watchdog ends a stalled run and restarts it only on the next pass.
Stress tests
What we break on purpose.
Six failures, caused deliberately in advance — and what has to come out of them.
Limits
What these tests do not do.
None of them creates an edge in the market. They prevent self-deception. No real broker is attached to our system yet.
At the end: the JKHD mark — Quant Systems, Just Know How to Dominate.