Aftermath
Monitoring. The first repair landed on 6 October 2026. The trigger below fired on 8 October, though not as expected: the gap was spend no door saw at all. The third repair landed the same day.
Reopen if the dashboard’s fal figure for a day lands more than 20 percent off fal’s own billing page, or it shows “fal spend NOT included” for more than a day.
| Read | How | When |
|---|---|---|
| Session’s billed spend | the status line in the terminal | at most once a minute |
| Week’s billed spend | the console’s usage dashboard | each refresh |
| Gap | dashboard against fal’s billing page | once a week while monitoring |
The rule that changed. Check the model page against the pricing API before trusting an estimate. A spend ceiling is only as true as the estimate under it. And read the provider’s own bill: an estimate only counts the calls that went through your doors.
Next read. The same class of mistake with a database instead of a bill: a throwaway test run rewrote our live database. And the workshop step that teaches the habit behind both: measure a mod’s effect, write the line that removes it, and date it.