Financial Model Audit Checklist and Scorecard - 61 Tests, Severity Weighting, Grade and Findings Log (Excel)

Description
This is not a model. It is the checklist you run against a model — somebody else's before you rely on it, or your own before you send it. Sixty-one tests in eight categories, each carrying a severity weight from one to three, a verdict you pick from a dropdown and a column for the evidence. Four tabs, 359 live formulas.

Why a checklist and not a scanner. There are add-ins that scan a workbook and produce a list of anomalies. They are useful and they are not this. A scanner finds inconsistent formulas; it cannot tell you that the growth rate was applied to the wrong base, that the covenant definition in the model is not the definition in the credit agreement, or that the file carries another client's name in its document properties. Those are the findings that end deals, and they are found by a person who knows where to look. So every test here is written to be run by a human in under a minute, and every one names the thing to click.

Severity is built into the arithmetic. Three means a defect that makes the output wrong or the file unusable. Two means a defect that will cause an error under maintenance. One means it makes review slower without changing an answer today. The score is weighted points earned over weighted points available, so a failure on "the balance sheet balances" costs three times what a failure on "the print area is set" costs.

Not applicable removes a test from both sides. Marking a test Not applicable takes it out of the numerator and the denominator, so skipping tests cannot flatter the score. A separate coverage figure reports what share of the tests you actually ran.

One severity-three failure caps the grade at C. Whatever the percentage says. In the worked example in the guide, a model scores 86.5 per cent across 57 tests run and still grades C, because four of the failures are severity three and one of them is the balance sheet. A model whose balance sheet does not balance is not a model with a small problem, and an average is the wrong way to say so.

The eight categories. Structure and navigation (8 tests) — inputs, calculations and outputs separated; one input tab; inputs buried inside formulas; hidden rows, columns and sheets. Formula integrity (10) — circular references and iterative calculation; formulas consistent across a row; hardcoded overrides; error values; error suppression; volatile functions; approximate-match lookups; external links; broken names. The three statements (8) — whether the balance sheet balances exactly and whether the check is an identity rather than a plug; cash from the cash flow statement; retained earnings rolling forward; asset and debt schedules. Time and periods (6). Assumptions and drivers (6) — units, compounding, the base a percentage is applied to, scenario switches, the same assumption used twice with two values, sensitivity grids agreeing with the base model. Financing and covenants (6). Tax, valuation and returns (7). Presentation and control (10) — including another client's name in the document properties, which is the leak that ends relationships.

Tabs: Read Me, Audit Checklist, Scorecard, Findings Log.

How to run it. Unhide everything first — right-click any tab, Unhide, then select all and unhide rows and columns; what was hidden is usually what matters. Work down the checklist setting each verdict. Write the evidence, not the opinion: a finding that says "Balance Sheet!H42 is out by $1,240 in year six" survives a disagreement with the person who built the model, and one that says "the balance sheet looks wrong" does not. Read the scorecard by category first, then the grade. Hand over the findings log, sorted by weight.

How it is built. Microsoft Excel (.xlsx). No macros, no add-ins, no external links, no password protection and no locked cells — which are four of the things it tests for. It ships blank on purpose; the verdict and evidence columns are yours to fill, and the completed example lives in the guide.

What it is not. Not an add-in and it scans nothing. Not a substitute for understanding the business — a model can pass all sixty-one tests and still be wrong about the market. Not audit in the statutory sense, and nothing in it constitutes an opinion on financial statements. Not tied to one model type: it runs against a three-statement model, a project finance model or a real estate proforma without changing the tests.

A 7-page guide and preview PDF, including the completed worked example, is included with the download.