Excel failures and good rubrics

First, understand what can go wrong in an Excel workbook. Then write clear criteria that describe a good result. Worked examples follow in a separate section.

Possible Excel failures

Use these five areas to inspect the workbook; not every area applies to every task.
Identify the underlying problem before deciding how to test it.

1 · Data and calculation accuracy

Source figures, dates, units and categories match the inputs.
Calculations use the right logic, denominator and financial treatment.
Look for missing inputs, boundary errors and totals that do not reconcile.

2 · Formula links and meaningful checks

Calculated outputs link to the relevant inputs and supporting calculations.
Input changes flow through the affected totals and summaries.
Reconciliation checks compare values that can genuinely differ.
Look for hardcoded outputs, stale links, typed “Pass” labels and checks that compare a value with itself.

3 · Completeness and preservation

All required periods, entities, scenarios and supporting detail are included.
Existing content outside the requested changes is preserved.
Look for omitted results, unrequested edits and lost formulas, formatting or workbook objects.

4 · Functionality, compatibility and reuse

Required features work in the user’s Excel version.
The solution handles the stated data size and future additions within the task’s scope.
Look for broken filters, dropdowns or refreshes, missing required add-ins and unsupported formulas.
Check that fixed ranges do not cut off new data.

5 · Presentation and professional judgment

Labels, units and visual distinctions make the result easy to interpret.
Required template meaning is preserved and helper material does not obscure the output.
Material assumptions are justified and explanations reflect the workbook.
Look for misleading conclusions or hidden uncertainty. Judge against the task and audience, not personal taste.

Avoid double-counting

A wrong source input that flows consistently through the workbook is a source error.
A summary that independently stops updating is a separate linking error.
Several visible symptoms can come from one underlying issue.

Automated checks: flag file differences and formula patterns.
Excel testing: verifies recalculation, refresh and interactive features.
Expert review: judges source interpretation, professional assumptions and materiality.

Turning failures into rubric criteria

A rubric is a set of checks that tells us whether a workbook does the job. Each criterion describes one important thing a passing output gets right.

The simplest test: Can another reviewer read the criterion, inspect the workbook and reach the same pass-or-fail decision?

A good criterion

Specific: names what is being checked.
Observable: describes a result we can verify.
Focused: tests one meaningful property.
Grounded: comes from the task, its inputs or a justified professional standard.
Useful: catches something the recipient would care about.
Self-contained: makes sense without reading the other criteria.

A weak criterion

Uses vague words such as “correct”, “clean” or “professional” without explaining them.
Combines unrelated checks in one sentence.
Requires a favourite formula or style without a reason.
Repeats a check already covered elsewhere.
Uses an unverified expected value or invented tolerance.
Is written just to make one particular output fail.

Two types of rubric criteria

Outcome

The prompt, input files or template fix what the result should be. This can concern values, behaviour, scope or presentation.

Preference / Quality

A justified professional expectation applies where the task leaves room for judgment.

Classify by where the expectation comes from. A number is not automatically Outcome, and formatting is not automatically Preference.

Good versus weak criteria

Three examples of clear, observable checks. Adapt the source and scope to the task.

Type	Weak criterion	Good criterion
Source accuracy	The workbook is correct.	Each shareholder’s opening shares match the supplied shareholder register.
Formula linking	The model should fix the summary.	The summary’s share counts link to the corresponding calculations in the ownership model.
Reconciliation	The check cell says Pass.	The ownership check compares the sum of shareholder ownership percentages with 100%.

How many criteria?

For a straightforward task, start with around 10 criteria covering the important requirements and possible failures.
For a more complex task, the set can grow to around 20 when there are several distinct calculations, behaviours or deliverables to check.
Treat these as starting points, not targets. Add a criterion only when it covers an important, separate check. Avoid padding the set or removing essential checks to reach a number.

Before you finish

Verify the expected result against the inputs and calculations.
Cover important requirements without penalising the same error twice.
Keep important checks even when every sample output passes.

Use meaningful labels instead of cell references.
Verify numerical benchmarks and agree any tolerance before using it.
Explain material changes before changing what passes or fails.

Two different questions: the failure type tells us what went wrong. The rubric type tells us why the expectation is justified.