Billed, and not there.

Method & limitations.

Counting, and one piece of arithmetic. There is no forecast and no model.

• 1
  Count what was in the garage, not what was sold. Every gate movement puts one car inside from the half hour it arrived to the half hour it left. Add them up across the year and you get occupancy by weekday and half hour, which is not the monthly average everyone reports.
• 2
  Find the peak that binds, and say which one it is. Not the average day or the average hour. The busiest half hour in the whole record is the only moment the garage has to survive, and every stall above that line is inventory nobody is using. That half hour is dated in the fact strip above, so you can go and look at it. The week chart shows something different and says so: an average weekday, which is a smaller number.
• 3
  Work out what one more permit costs in space. The permits in use produce that peak between them, so each one contributes the peak divided by their number. Permits that never open a gate are left out, because they take no space. This is conservative twice over: it assumes a new permit behaves like the average rather than the quiet tail, and it ignores that a larger, more varied population peaks lower than proportionally.
• 4
  Replay the whole year at every permit count. The slider does not extrapolate a curve. For each setting it goes back through all n weekdays and counts how many would have filled. That is why the answer is a number of days, not a probability. The recommended count is the largest one at which that number is zero: the level where the busiest half hour of the year still fits inside the garage. There is no safety coefficient anywhere on this page. The margin is that day. Whether you want more margin than one dated day of evidence is a judgment, and the slider is how you take it.
• 5
  Leave the closed days out of the averages, and count them out loud. A holiday or a snow day sits in the gate record as a weekday with almost nobody in it. Averaging those in drags down the number step 3 divides by, and most US holidays are Mondays, so the damage is uneven. A day whose peak is under a third of that weekday's median is treated as a closure and left out. It never touches the fill test in step 4, because a closed garage cannot fill.
• 6
  Judge dormancy against the gate, not the invoice, and not against a credential too young to judge. A permit billed every month whose credential has not opened a gate in ninety days is not revenue at risk in the usual sense. It is a stall that was never occupied, so it is already inside the peak figure, and canceling it frees nothing. A credential issued six weeks ago that has never been used is a new hire who has not started. It is counted separately, not sold as a finding.