Re-add the bridge,
get the residual
Budget, actual, and the drivers the commentary names. The tool re-adds them and prints what is left over, at the precision you typed, with the pack’s own name for it.
Free. No signup, no email, no AI. It runs in your browser and nothing you type is sent anywhere or stored.
Any unit: thousands, millions, the currency itself. Keep the decimals you report.
The drivers the commentary names
Actual 11.2 against budget 10.0.
UNEXPLAINED: +0.2. A remainder the commentary does not account for.
The commentary does not close. UNEXPLAINED: +0.2 is what the bridge leaves unaccounted for; name its source before it is absorbed anywhere.
The working
Variance: 11.2 - 10.0 = +1.2 Drivers: 0.3 + 0.5 + 0.2 = +1.0 Residual: +1.2 - (+1.0) = +0.2 Shown at 1 decimal, the precision you typed.
What it means when the two boxes disagree
A variance commentary is a claim that a list of drivers explains a number. Re-adding the list is the check this page runs, and it is easy to skip because the prose reads as complete. When the drivers re-add to less than the variance, the difference is what the bridge does not explain: a driver not yet named, an allocation, an interaction between drivers, a rounding convention, or a mistake in the bridge itself. Its size says how much is unaccounted for; finding out which of those it is comes next, not the decision.
A bridge that closes exactly is not suspicious on its own; a documented method with a governed residual line is what a good bridge looks like. What deserves the question is a driver whose quantification nobody can state, because that is how a gap gets closed by sizing a number to fit. Ask how each driver was measured before the zero counts as proof.
Where this check comes from
The numbers pre-filled here are the variance fixture: a budget of 10.0 against an actual of 11.2, and a commentary naming three drivers of 0.3, 0.5 and 0.2. They re-add to 1.0 against a variance of 1.2. In the pack, the agent that prepares the variance investigation must write UNEXPLAINED: +0.2 rather than absorb it, and its test set also carries a planted exact fit that it must refuse to trust.
The check is one of the fixtures in The Finance AI Stack. You do not need the pack to use this page. That is the point of publishing the page.
The ten builds that pack sequences are on a free two-page card: Your First 10 AI Implementations, in the order to build them, with what the first ten leave out on purpose.
AI prepares, your people decide.
Practitioner guidance, not accounting, audit or financial advice. Units, signs and the decimals shown follow what you type; this page re-adds your numbers and does nothing else.
The other checks in this family: The same hiring round, at two decimals and at four · The biggest discount that still holds the floor · A tolerance with two limbs, and the one that binds · The instructions you pasted, counted against the ones you tested · All free tools