What it does
Anomaly Detection (built as the Validation Checks Builder) adds a validation sheet to any workbook. Each check compares an expected value with an actual value calculated by formula and returns a status automatically, so problems surface the moment the numbers change rather than at review time.
When to use it
- A workbook has no error checking, or its checks are typed text that never changes.
- You are handing a model over and want its integrity to be visible.
- After building or extending a model, as a standing health check.
How it works
- Pick the relevant categories for the workbook type:
- Structural: formula errors, blank key cells, sheet naming, whether a change log exists.
- Financial integrity: balance sheet balances, cash flow reconciles, P&L foots, revenue is positive, retained earnings tie.
- Data consistency: period continuity, row and column counts, duplicates, detail tying to summary.
- Model-specific: ACT/FCST boundary, sequential counter, valid named ranges, scenario toggle works, no hardcodes in the forecast zone.
- Build the sheet with the standard columns: check number, description, expected, actual (formula), variance and status, plus a notes column.
- Set materiality: $1 for financial checks and zero for count checks, so rounding doesn’t trigger false failures.
- Summarise PASS, FLAG and FAIL counts at the top.
For EXL Cloud models, Skill 02 also lists model-specific checks to enable, such as date validation, data refresh status, unmapped revenue and expense values, working capital splits summing to 100%, a pre/post-processing P&L check and a negative cash check.
What you get
A Validation Checks sheet with 10 to 15 meaningful checks. Every status cell is a formula and every failing check has a remediation note. In EXL Cloud models the total error count feeds the error message shown on each sheet header.
Good to know
- Adds checks only. It does not fix issues, clean data or build models.
- Typing “PASS” as text is the most common failure. Every status here is a formula.
- Text comparisons use an equality test, not a numeric variance.
- Checks are terminal: they never feed back into the model, which avoids circular references.
- Fifty checks means none get attention. The skill aims for 10 to 15 that matter.
- Run Skill 14 first so the validation sheet can confirm the Change Log exists.