Case studies
Every project can be shown from specification to test report.
The DLL could not run on the new server, and there was no documentation of its internal logic.
Publishable results: The expert now runs on the new server with no external file, and the function parity results are documented.
On volatile days the client kept re-entering and blew past the daily loss limit they had set for themselves.
Publishable results: Replaying the twenty worst days of the test set, no day's loss crossed the limit, and the account survived its first full …
Three independent experts kept entering, and on news-heavy days combined exposure reached twice the agreed ceiling.
Publishable results: Daily reporting never showed combined exposure above the ceiling, and the firm built its first internal risk note on those numbers.
In volatile index stretches the expert opened many positions at once and intraday equity fell with no limit at all.
Publishable results: Replaying two crisis days, the guard fired before equity reached the ceiling and the system returned to normal the following day.
The expert sent orders with no slippage ceiling, and in busy hours they filled several pips worse.
Publishable results: The logged gap between requested and filled price came in under half a pip after the change.
The team kept one copied build per account: every fix had to be applied four times and the copies slowly drifted apart.
Publishable results: Applying a fix went from four repetitions to one change, and for the first time the team can see every account's version …
The client opened a separate window for every analytical question, and in busy periods the terminal slowed down and decisions got harder.
Publishable results: Chart load time in testing fell from seconds to under a second, and the client now makes every decision from one screen.
To pick trading hours the client reviewed history by hand, and the conclusion differed each time.
Publishable results: The client changed their trading hours based on the output, and trades taken in their weakest session fell to under a third.
The team optimized on one period and then reported that same period — and the result never repeated on fresh data.
Publishable results: Two of the team's three strategies showed a clear drop in evaluation windows and were revised before any live account.
The gold backtest had been built on incomplete data: hourly gaps and duplicate candles flattered the statistics.
Publishable results: After cleanup the gold backtest came out more conservative, and the client shortened the window to three years to keep the data …
The team had tested each index alone; combining them could either worsen drawdown or make it look artificially small.
Publishable results: The result pointed to a lighter weighting for better risk-adjusted returns, and the team paper-tested it before touching an account.
Optimisation produced a single good point with no way to tell a real peak from a random bump in the data.
Publishable results: The report showed one parameter sat on an edge; shifting it slightly made the result more stable without giving up return.
