My Claude plan drained overnight but Token history shows 0: where the usage really went

Your Claude plan says a chunk of the allowance went overnight, Otto was only supposed to do a brain check and a backup, and Token history shows nothing ran. This article explains the three honest reasons those numbers can disagree, and how to read the history so you trust it again.

First: the brain check and the backup are free

The two overnight presets in the Runs on its own card in Settings, Back up my data and Weekly brain check, never touch Claude at all. The backup copies your files, and the brain check is a scan the app does with its own code. Both cost exactly zero AI usage. So a night where only those two ran and Token history shows 0 is the app being accurate, not blind. The drain came from somewhere else.

Reason 1: an older app dropped stopped runs from the history

On versions before 1.11.71, a run that was cut short, whether it hit your plan's usage limit, crashed, or was stopped by hand, recorded nothing at all. It spent real allowance and then vanished from the ledger. That is exactly the "40% gone and history says nothing ran" picture. Update the app: current versions count an unfinished run from its moment-by-moment samples, and the run appears in Recent runs with a note like stopped at your plan limit or stopped before it finished.

Reason 2: the ledger only counts what Booked ran

The Token history page says this about itself: it counts what Booked ran, and only that. Anything else signed in to the same Claude account spends from the same plan and never reaches this ledger: Claude Code in a terminal, claude.ai in a browser, the phone app. If any of those were active, their spend shows on your Claude plan but not here.

How to check properly

1. Open Advanced in the top bar, then under History click Token history.
2. Switch the range to 1 day and flip between Whole runs and Moment by moment.
3. If you see "Booked ran nothing in this window. There is spend recorded outside it - widen the range to find it," do exactly that.

Good to know

Raw token counts overstate the load: most of the app's input tokens are cache reads, which are the cheap kind. So the app's token number and your Claude plan's percentage are not the same scale, and comparing them directly will always look wrong.

Still stuck? Email bookedsolid@kivimedia.freshdesk.com and a person will help.