Where the revenue went.

1 · Churn by month — accelerator + product

Churn is an account going to zero CARR in a month. Contraction is an account that shrank but still pays. Both come from the same waterfall behind the CARR report, so these numbers reconcile to it by construction. Rate is churn against that month's beginning CARR.

2 · Every account that churned — 109 events

One row per account per month where CARR went to zero. Source id is the Sage customer id, so this works as a queue: copy the id, open the record. Newest first.

3 · Contraction — 13 events

Accounts that shrank without leaving. Worth watching separately: a downgrade is often the month before a churn, and it never shows up in a logo count.

4 · Reconciliation — does the detail equal the warehouse?

The left pair is what ANALYTICS.CARR_MONTHLY reports. The right pair is the sum of the account rows above. They should match to the cent, because both use the same definition. Any red row means this report and the waterfall have drifted and one of them is wrong.

All 21 months reconcile exactly.

5 · Recognition basis — effective vs booked — accelerator, trailing 8 closed months

Two calendars for the same churn events. Effective is the month the customer actually left — the ratified, investor-grade basis every other number on this page uses. Booked is the month back office processed the cancellation in Sage — the basis the Sage SaaS-Intelligence dashboards show. Same events, different calendars: a cancellation effective July 31, 2025 was keyed December 9, 2025, so Finance’s panel books that churn in December and this report books it in July. Over Feb–Jun 2026 the two calendars agree cumulatively within $2,400 — the monthly wobble between them is timing, not missing revenue. A month’s booked figure can include events effective before the window (a processing backlog clears), which is exactly what the 2025 panels show.

Booked months key on Sage’s WHENMODIFIED audit stamp, which moves if a line is ever edited again — good enough for a tie-out view, never an official basis.

6 · Cohort retention — vintage triangle + the July-1 snapshot

Two different retention metrics share this panel; read them separately. The triangles are vintage retention: each row follows only the logos whose first contract began that month, and each column asks how many still pay N months in. Each category gets two — logo retention (how many stayed) and revenue retention (how much of their month-0 CARR stayed). Same survival test, different weighting, and the gap between them is the reading: a vintage that keeps its small accounts and loses its largest looks healthy on logos and thin on revenue. The small +n recovered counts appear on the logo triangle only — recovered revenue books as its own waterfall line, and showing it twice would double-count it. The cards are the July-1 snapshot cohort — everyone active on July 1 tracked forward, whatever their signup date. That snapshot is Kristen’s GLR KR; the two must never appear on one chart unlabeled. Non-renewal is recognized at contract end, so a step-down around month 12 is annual terms coming due — the metric working as designed, not a sudden loss event. Once a logo churns inside a curve it stays churned: a re-sign shows in the small +n counts (recovered), never back in the retention figure — resurrecting re-signs inflates survival, the same class of error as the ~46% stacking figure, pointed the other way.