What a 99% modelling quality figure actually proves

[Amanhasnoname]

That 99 percent modelling quality number in the report header gets quoted like a certificate. It is not one. In most platforms it only measures how much of the test period was covered by tick-level data rather than interpolated minute bars. It says nothing about whether those ticks resemble what a broker really streamed.

Both of these can be true at once: 99 percent quality and a completely unrealistic result. If the ticks were generated from minute bars and paired with a fixed spread, the figure is high because the generator filled every gap, not because the real market was there.

What the number does tell you is that intrabar ordering is at least being simulated. What it cannot tell you is whether spread widened at news, whether prices were bid-only, or whether the source was clean.

Treat it as an entry ticket, not as evidence. Better habits:

- check whether the ticks are real or generated
- check whether spread is variable or fixed
- compare the trade count with a short live sample

A scalper whose results collapse when the spread input moves by 0.3 pip had an edge made of settings.

Which checks do you run on a test before the quality percentage even gets looked at?