Most volume spikes on M1 bars are feed noise, so test before trusting them

[Amanhasnoname]

A single one-minute bar with triple the usual tick volume looks meaningful. Run the numbers on a few weeks of data, though, and a large share of those spikes line up with nothing: no range expansion, no follow-through, no news. They are often caused by reconnects, a feed hiccup, a rollover period or a platform clock aligning stored ticks into one bar.

Because the unit is the one-minute bar, a few seconds of burst can double a bar's count. A feed that delivers late and then catches up will also distort which minute a tick lands in, so the spike sits in the wrong bar.

A better filter is to demand confirmation from range. A spike counts only if the bar's range is above its own recent average as well, and ideally if volume stays elevated for a second bar. That cuts the number of signals sharply but the ones remaining are less likely to be artefacts.

Also exclude the few minutes around rollover and the first minutes of the Sunday open from any statistics, because those are known to produce odd counts.

Which filter did you settle on to separate genuine volume spikes from garbage, and how many signals survived?