The hours I turn the tool off

Most writing about scheduled events is about how to trade them. This is about why I stop — and what a tool honestly can and cannot show while the book is thin.

The schedule that matters on CL

Crude has a weekly rhythm that is not a matter of opinion:

EIA petroleum status report — Wednesdays, 10:30 ET.
API inventory estimate — Tuesdays, after the close.
US session — activity concentrates roughly 09:00–14:30 ET.
The thin window — overnight hours where the book is materially lighter.

The conventional advice around the first of these, which you will find at most futures brokers, is to be flat going into the release and to wait afterwards before re-engaging. I follow it, and the tool follows it with me.

Why I stop rather than adapt

The tempting alternative is to keep everything running and let the rules handle it — widen a threshold, lengthen a hold, and carry on.

I do not, for a reason specific to what this tool measures. Order-flow reading is comparative: a print is large relative to what has been printing, absorption is one side holding against the flow arriving. Those comparisons assume the recent past is a reasonable reference for the present.

A scheduled release breaks that assumption on purpose. The reference window immediately before the number is nothing like the seconds after it. Everything gets flagged, because everything is unusual relative to a baseline that no longer applies.

So the honest response is not a wider threshold. It is to say the measurement is not meaningful right now, and to stop reporting it.

What actually breaks in the reading

Adaptive thresholds lag the regime change. A threshold tuned on the last N minutes is describing a market that has just ceased to exist.
Absorption becomes unreadable. A side holding against heavy flow and a side that simply has not pulled its quotes yet look identical for a few seconds, and a few seconds is the whole event.
Levels built before the release describe a different auction. They do not stop being facts, but they stop being comparable to what is happening now.

None of that is a failure of the software. It is what happens when you measure a comparison across a discontinuity.

The quiet hours are a different problem

Overnight is the mirror image. Nothing is unusual because almost nothing is happening, so an adaptive measure drifts down until ordinary prints start clearing the bar.

This is worth separating from the news case, because the fix is different. Around a release the reading is unstable and should be suspended. In thin hours the reading is stable but the sample is small, and small samples produce confident nonsense — which is the same trap as calling a fifteen-trade result promising, on a shorter clock.

Honest limits

One instrument. The schedule above is crude. Other markets have their own, and the reasoning transfers while the times do not.
This is not a claim that stopping is more profitable. I have not measured the counterfactual of trading through releases, and I am not going to assert one I have not measured.
Not advice. A description of what I do and why the measurement stops being meaningful — not a recommendation about what anyone else should do around a news release.