What moved this week, and can I trust the numbers?

One definition a figure. Each KPI is written once, in the metric model, and every view reads it from there. Every figure is a sum over accounts, so a slice is the same sum over fewer of them: the plans add to the company, the channels add to the company, and at every month-end the company is the monthly revenue report’s to the euro. September 2026 closes at €342,458 of recurring revenue from 4,492 paying customers on both.

The drill-down. Every tile, column, bar and row opens the rows behind it: recurring revenue by plan, then by channel, then the bridge’s lines, then the accounts that moved. At each level the rows add to the figure they open from, the accounts included: past the first eight, the rest sum in one row. The path stays above the table, so one click goes back up.

The checks catch the two planted faults. With the late load on, product events stop three days before the period ends, so every period’s sign-ups are counted to the same day and the strip names the views that read them. With the duplicate IDs on, billing gives up to two of the period’s returning customers a new customer ID; matched by email, they still count as returning, and the drill-down shows both IDs. In the week the dashboard opens on, customer churn ran above its normal range: 49 accounts left, and the drill-down lists each of them.

What this shows, and what it does not

What it shows. One run of the company, 56 weeks, 13 months and four quarters of it, read the way an analytics team reads its own: the figure, its change, its normal range, and the rows behind it. A fault moves only what it touches: recurring revenue, customers and churn stay as they were with both faults on, and the strip flags exactly the faults switched on.

What it does not. A check catches only the fault it was written for: a returning customer who signs up with a new email address passes the duplicate check and counts as new. The dashboard says what moved and leaves the why to an analysis. The plan counts are the model’s; which accounts moved, and their channels, are drawn by a stated rule that keeps those counts, so the split by channel is an assumption of the demo, not a finding. The days are simulated from a monthly model whose moves are spread over the days of each month.

The full analysis. The example weekly KPI dashboard is the week to Sunday 27 September 2026 in full, with both faults planted: every source, every check, and what changed. The cohort demo prices the four acquisition channels this dashboard slices by.