Verified local facts
The black-box bench uses two blind reconstructions and a third-reader transmission test to expose hidden inputs, conventions and rounding before an example is trusted.
Evidence reviewed
Help a Mauritius household audit a worked example using attributable evidence and a stop condition.
Decision checklist
The black-box bench uses two blind reconstructions and a third-reader transmission test to expose hidden inputs, conventions and rounding before an example is trusted; review marker 106 applies this device only to “audit a worked example Mauritius” and the evidence boundaries named on this page.
Freeze the displayed example
Save the complete example with owner, address, product, date and all nearby qualifications. Do not copy only the headline payment. Record what the provider states the example is intended to illustrate. MCB and MauBank exhibits remain separate; no figure crosses between them. If the example changes during review, preserve both versions and complete the audit on one frozen display.
Extract an input register
List capital, useful cash if shown, rate label, term, number and frequency of payments, fees, insurance, taxes stated, start timing and rounding clues. Mark absent rather than guessing. A value implied only by the displayed result is not an independent input. The register distinguishes explicit facts, derived quantities and unresolved conventions. This prevents circular reconstruction in which the expected answer is used to manufacture its own ingredients.
Build the first blind reconstruction
Reviewer A calculates using only the register and records every operation in order. They keep intermediate precision and state each rounding point. A formal rate measure is not recreated unless the necessary method is disclosed; cash flows can still be checked. The reviewer cannot consult the displayed answer until the worksheet is sealed. This makes the first result evidence of method rather than a fitted explanation.
Build the second blind reconstruction
Reviewer B receives the frozen exhibit and builds a new worksheet without seeing A's sequence. They may organise flows differently but must use only admitted inputs. Any required convention that is not documented receives an exception. Independent work is important because two people copying the same hidden assumption would create false confidence. B seals inputs, formula order, rounding and output before comparison.
Classify the divergence
Compare A, B and the displayed result line by line. Classify differences as transcription, omitted charge, payment count, timing, formula convention, rounding or unresolved. Correct a clear copying error openly; do not tune an assumption simply to reach the provider number. If an absent convention explains multiple possible results, the example remains unreconciled and produces a precise question for its owner. Give each divergence an identifier so its correction can be traced through both worksheets and the public explanation.
Run the third-reader transmission test
Give a third reader the frozen exhibit, final worksheets and divergence record but no oral explanation. They must locate every input, reproduce subtotals and understand why each difference closed or stayed open. Failure identifies the missing reference, unit or convention. The test guards against expert intuition hiding a step that ordinary readers cannot inspect. It also shows whether the educational explanation genuinely teaches rather than merely announces correctness.
Separate the three verdicts
Issue distinct boxes for arithmetic recreation, household affordability and provider decision. The first can pass while the second fails after payments meet essential spending. Both can pass as analysis while the institution still declines or changes terms. Conversely, approval does not repair an unreconciled example or guarantee household fit. Keeping these verdicts separate prevents an accurate illustration from becoming a sales promise.
Publish the audit trail, not a prediction
Close with frozen version, input register, worksheet hashes, divergence classification, third-reader result and unresolved questions. State whether the example is reproducible, partly explained or held. Do not reuse its amounts for another reader or call them typical beyond the provider's own qualification. A changed exhibit reopens the audit. The bench's success is transparent arithmetic and teachable uncertainty, never a forecast of price, approval or personal outcome. Add a test-vector appendix so the method survives beyond one display. Create small synthetic input sets whose answers are known because every flow is explicitly defined: no charge, one initial deduction, a different last payment, periodic insurance and a rounding boundary. Label these as editorial test cases, never Mauritius offers or typical values. Run both worksheets through them. If the reconstructions disagree on a fully specified vector, the calculation method—not the provider example—needs correction. Next, return to the frozen publication and distinguish a method defect from missing provider disclosure. The appendix records which formula step each vector exercises and prevents a later code or spreadsheet change from silently altering results. A reviewer repeats one vector manually and checks that machine precision, displayed rounding and final MUR presentation remain separate. Then the household layer uses only the personal quotation, not the synthetic figures, for affordability. The institution's assessment stays outside every vector. A perfectly tested calculator cannot know identity, income evidence, commitments or provider criteria absent from its inputs. The closing statement therefore reports three boundaries: the arithmetic method passed its vectors, the published example was or was not reproducible, and no approval or personal suitability inference was attempted. This appendix turns a one-off check into a maintained audit discipline without converting test data into financial claims.