Currently, ETA is calculated based on the last 10 measurements, so it's a moving average with window size = 10. I suggest using EMA instead.
Benefits: It removes the necessity of remembering history measurements, as it can update iteratively. It also prioritizes more recent speed.
speed_now = delta_finished / delta_time
speed_ema = smooth_factor * speed_now + (1 - smooth_factor) * speed_ema
Or if you think current eta calculation is better kept as is, maybe it's a good idea to add a --ema=0.3 to support EMA as opt-in feature?