# AI spreadsheet brief

## The question the workbook answers

- Decision this workbook supports: [decision]
- Person who will open it: [role]
- Period and grain: [e.g. Jan 2025–Jun 2026, monthly]
- Currency, units, and time zone: [scope]
- Where it goes next: [emailed, uploaded, pasted into a deck]

## The data

- Source file(s): [filenames]
- Rows and columns to keep: [list]
- Field definitions: [column = meaning, type, unit]
- Known dirt: [merged headers, blank rows, text numbers, mixed date formats,
  duplicate ids, "N/A" placeholders, inconsistent category spellings]
- Rows to exclude and why: [refunds, test orders, incomplete final month]

## The workbook

- Sheets: Inputs, Model, Dashboard, Checks (or state your own structure)
- Every derived value is a live Excel formula, not a pasted result
- Assumptions live on Inputs only; no business input is retyped on two sheets
- Dashboard KPIs reference the exact period named above
- Charts: [one claim per chart, with axis labels and units]
- Formatting: number formats, column widths, filters, frozen header row

## The checks sheet

Column A is a human-readable label, column B is a formula. A numeric check
passes at exactly 0; a boolean check passes at TRUE. Include at least:

- grand total minus the sum of its components
- closing balance minus (opening + movement)
- each dashboard KPI minus the model cell it claims to show
- row count in the model minus expected row count from the source
- a period check: the last modelled period equals the requested end period

## Rules

- No hardcoded number where a formula belongs.
- No blanket IFERROR that hides a broken reference; handle the real case.
- Say "not established" instead of inventing a value for a missing field.
- List every row dropped during cleaning, with the reason and the count.
- Report the recalculation and check results before summarizing the findings.