Verify every task

This is the check that matters. A provider can pass on its own while the task pointed at it quietly falls back to something else — this row tests what the task will really use, after fallbacks.

Task	Provider	Model	Result

Usage — last 30 days

Provider	Model	Calls	Tokens in / out	Cost	Used for
Unrecorded	`deepseek-v4-flash`	40	12,296 / 8,683	$0.0129	—
Total		40		$0.0129

Token counts come straight from each API response, so cost is calculated from what was actually used. A custom gateway resells at its own rate, so no cost is shown for those rows — quoting the official figure for a reseller would be worse than showing nothing.