KPI dashboard that checks your data first

• Owners of micro and small firms without a controller.
• Anyone who prepares a monthly figures summary from CSV exports.

A dashboard is only as good as the export behind it. Sales and expense files from accounting software carry empty cells, impossible dates, repeated invoice numbers and VAT that does not add up. A dashboard that charts them anyway hides the problem until a total looks wrong.

1. Reads two CSV exports, sales (invoices) and expenses: comma or semicolon separated, Italian (31/03/2026), ISO or US dates, 1.234,56 or 1,234.56 amounts. You choose the date and amount formats; the app does not guess them, and tells you when another format would fit better.
2. Maps columns, with common English and Italian headers (Cliente, Imponibile, IVA, Totale, Scadenza, Data pagamento) pre-selected. You can change every mapping.
3. Checks the data before any chart. Rows with errors (empty required cells, impossible dates, non-numeric amounts, future dates, exact copies, the same ID on different rows) are left out of every number and listed with file, line, column and value. Warnings (net + VAT not equal to gross, negative amounts, due or paid date before the invoice date) keep the row and flag it.
4. Computes the KPIs from the rows that passed, with a banner saying how many rows were left out: revenue, expenses, profit and margin, cash balance, top customers, and overdue invoices by age (1-30, 31-60, 61-90 and over 90 days). Money is handled in whole cents, so totals are exact.

Local processing runs on your own computer, with owner access.

Twelve months (October 2025 to September 2026) of invented invoices and expenses, with faults planted on known lines. Measured 5 October 2026; figures from the project's evaluation summary.

• Planted faults found: each fault must be flagged by the right check on the right line.
• KPIs equal the reference: revenue, expenses, cash, overdue amounts, monthly series and the top-10 customers computed by the app from the faulty files equal, to the cent, figures computed separately from the generator's answer key.
• What this does not show: the faults are the ones we chose to plant, and the generator and the checks were written by the same developer. A perfect score shows the checks do what they claim; it does not show they catch every problem in real exports.

Python, pandas, Altair (charts), Streamlit; tests with pytest and Playwright; packaged with Docker.

• Synthetic data only; not tested on real exports from any accounting program or bank.
• Expenses are treated as paid on their date; there is no accounts-payable view.
• Partial payments are not supported: an invoice is either paid or not.
• Credit notes are kept and flagged, not matched to their invoice.
• One currency (EUR). Excel files are not read; export to CSV first.
• A reporting aid, not accounting or tax advice.

Source: snello-kpi-dashboard README and reports/summary.json, generated 5 October 2026. Code is private, so it is not linked.