A displayed figure is the end of a calculation, not the beginning of an audit. To decide whether that figure is reviewable, start with the result and walk backwards. The goal is to recover an unbroken provenance chain: the result's unit and date, the operation that produced it, every derived value it depends on, the assumptions that shape the case and the source values entered at the beginning. Viotus publicly describes these elements across Calculator and Valuator; this guide turns them into one repeatable audit task.
1. Freeze the result you are actually reviewing
Write down the displayed result before changing anything. Capture its value, unit, relevant date and scenario. A bare number is not enough: 5 can mean a percentage, a currency amount or a duration, and a value for one date may not represent another. The public Viotus framework states that a figure without its unit, date, assumption or source is difficult to trust. Treat those fields as part of the result, not as optional notes.
Also identify the exact case: base case, alternative scenario or sensitivity view. Valuator's public design keeps assumptions, scenarios and calculated ranges connected precisely because a change in the case can explain a change in value. If you cannot name which scenario produced the figure, pause the audit. You do not yet have a stable object to review.
2. Follow the execution trace one operation backwards
Next, find the operation that emitted the result. Record its method, inputs, output and any warning shown with the run. Calculator's public direction keeps inputs, outputs, warnings and calculation traces together, and its engine trace is intended to make a run reproducible. The trace should answer a simple question: what happened immediately before this number appeared?
Rerun the same inputs and rules. Matching output does not prove that the model is appropriate, but it establishes a stable calculation to inspect. A deterministic method can still contain a poor formula, a wrong input or a material omission. Its value to an auditor is narrower and essential: the same run can be reproduced, so disagreement can be located rather than hidden by a changing answer.
3. Separate derived values from source values
For every input to the final operation, label it either source or derived. A source value was supplied to the workspace; a derived value was produced by an earlier calculation. Then repeat the same question for each derived value until no unexplained intermediate remains. This reverse walk prevents a calculated subtotal from being mistaken for original evidence.
At every link, check unit compatibility and execution order. Viotus publicly describes Calculator blocks that can be connected only after their units and order are validated. That matters because a plausible final number can still rest on incompatible quantities or on a downstream value computed too early. If an upstream value changed, verify that dependent calculations belong to the same completed run rather than a mixture of old and new states.
4. Inspect each assumption as a testable choice
Now list the assumptions attached to the chain. Do not hide them inside labels such as conservative or expected. Record the actual value, unit and scenario in which each assumption applies. Valuator's public design connects inputs, scenarios and ranges so a user can understand why a value changes; use that relationship as an audit test.
Change one assumption deliberately, run the calculation again and compare the resulting range or scenario. The objective is not to discover a universally correct assumption. It is to confirm that the result responds to the named assumption, that the comparison remains visible and that you can explain the direction of the change without inventing a story after seeing the output.
5. Verify the origin, unit and date of every source value
The reverse walk ends at the source values. For each one, ask who or what supplied it, which unit it uses, which date or period it represents and whether it belongs to the audited scenario. The available public Viotus evidence does not claim that software can certify an external source as correct. It says the origin must remain visible. Provenance lets a reviewer find and challenge an input; it does not turn that input into truth.
A date deserves special attention. Two correctly entered values may describe different periods and therefore fail to support the same result. Likewise, a unit conversion must appear as an operation rather than disappear between fields. If the source cannot be identified, the date is absent or the unit changes without a trace, mark the chain incomplete instead of guessing.
6. Close the audit with a provenance record
Summarise the chain in one line of descent: displayed result, final operation, derived values, assumptions and dated source values. Attach the execution history and warnings that belong to that run. Calculator is designed around persistent workspaces, connected results and revisitable execution history, while versioned workspaces provide a way to preserve work as work rather than as a screenshot of an answer.
A result passes this audit when another reviewer can identify its case, traverse every dependency, reproduce the run and see where judgement entered. It fails when context must be reconstructed from memory, when source and derived values are indistinguishable or when the trace cannot show which run produced the output. That conclusion is intentionally modest: it establishes reviewability, not financial correctness or advice. Interpretation and the eventual decision remain with the user or an appropriate professional.
A compact financial result audit checklist
Before relying on a result, confirm: the output has a unit, date and scenario; the final operation and warnings are visible; the same inputs and rules reproduce the run; every intermediate is marked as derived; units and execution order are valid; each assumption is explicit and can be varied; every source value has an origin, unit and date; and the saved execution history belongs to the result under review.
If one answer is missing, stop at that link. Do not compensate with a confident explanation. The useful question is not merely whether the number looks reasonable, but whether you can travel from that number back to evidence without losing context along the way.