Case studies
Every project can be shown from specification to test report.
The earlier version filled directionless stretches with low-quality trades, and costs ate the result.
Publishable results: Daily trade count halved while entry quality held up in the out-of-sample test.
Yen positions were carried into the weekend, and the Sunday gap undid the week's work.
Publishable results: No position reached a holiday during the test, and the client's weekly report no longer contained gaps.
Copying trades by hand across six accounts produced size errors and delays on busy days.
Publishable results: Size differences between accounts went to zero and the team doubled the number of accounts under management.
The old settings panel had become incompatible with newer terminal builds, and unnamed set files made correct configuration guesswork.
Publishable results: Reconfiguring every expert fell from hours to minutes, and set-file mix-ups disappeared from the reports.
Trade analysis was manual: every trade copied out of the terminal, and a typo in the data would invalidate the results.
Publishable results: Monthly data collection dropped from two days to minutes, with no numeric difference from the terminal report.
The strategy text was full of qualitative phrases — "calm market", "clear move" — and without numbers every implementation behaved differently.
Publishable results: The mapping table showed every rule in the client's text landing on a code condition, and off-rule trades dropped to zero.
The idea was clear on paper, but nobody knew how many signals it produced historically or whether execution cost justified them.
Publishable results: After reading the report the client changed the window parameter, and the final build behaved the same way on fresh data.
The client could not watch charts during working hours, and rushed decisions in their free two hours ruined the month.
Publishable results: In the test most trades closed within the first two hours of the session, and the client reads the end-of-day report on …
The indicator draws its arrow at candle close; manual execution always arrived a candle late and the stop was often forgotten.
Publishable results: The bot took 21 of the 23 arrows in the test window, and the two misses were reported with their cause.
The signal appeared on screen, but by the time the team clicked, the market had moved — spreads at the open made it worse.
Publishable results: Entry delay fell from seconds to under half a second, and the team could measure each threshold separately.
Higher-timeframe data arrives later; without synchronisation the bot decided on an incomplete candle.
Publishable results: Trades taken on an incomplete candle dropped to zero, and the logs show every decision came after the data refreshed.
The MT4 build leaned on limited buffers and legacy functions; a naive port would break colours and alerts.
Publishable results: Numeric output comparison showed no difference, and the indicator draws the same familiar picture on MT5.
