dead minutes — do i drop them or flag them?

so i read the research post here about how much of a recorded session is basically empty minutes (nothing traded, quote never moved) and i've been staring at my feature pipeline ever since feeling kind of dumb. my setup: small fixed watchlist, 1 minute bars, i build rolling features (returns, volume z-score, spread) and train a little gradient boosted thing to predict the next few minutes. the post's argument, as i understood it, is that a big share of those minute rows on a fixed list aren't observations at all, they're the previous observation with a new timestamp on it. which if true means my rolling windows are being padded with copies of themselves, and my train/test splits look way bigger than they really are. what i don't know is what to actually DO about it: 1. drop those rows entirely and let the clock be event based instead of wall clock? feels cleaner but then my "last 20 bars" is sometimes 20 minutes and sometimes most of an afternoon, and i have no idea if the model can handle that without me feeding it elapsed time as a feature. 2. keep them but add a flag column saying "nothing happened here"? that keeps the shape but i suspect the model just learns the flag and i've solved nothing. 3. resample to something like volume bars / dollar bars, which people keep mentioning to me and i keep not doing because i don't fully understand them. also the part that stuck with me was that a scanner-driven universe had way fewer of these empty minutes than a fixed list. is that just because scanners pick things that are already moving, i.e. it's selection and you've now got a survivorship problem baked in? because that feels like trading one bias for another, not fixing anything. and honestly the bit i'm least sure about: does any of this even matter for a slow strategy? if i'm holding for hours, do i care that the minutes in between were dead? or is the damage all in the evaluation stage, where a backtest counts a stale minute as a real chance to be right? happy to be told i'm overthinking a beginner problem.