Kestrel Fitness metric definition register
Owner: Arjun Mehta. Updated Oct 1, 2026.
How a definition is adopted
One register of what every figure in the Kestrel monthly pack means: the calculation, the source, the owner, the client approver and the date each definition takes effect. Kept by Arjun Mehta, analytics lead at Northfold Analytics, with Sofia Reyes, senior analyst, and read by Grace Whitfield, head of membership data at Kestrel Fitness. Register as of October 1, 2026.
A definition enters as Proposed with an owner, a calculation, a source field and a period basis. In Collision check Northfold tests it against every existing measure that shares its name, its population or its period basis. Grace Whitfield approves it in Client review, Arjun Mehta then sets the effective date, and an approved definition is never applied to packs that have already been released. Rejected and Superseded definitions stay in the register with their reason, so the same debate is not repeated. The position marked above is Visits per active member, which is in Collision check today.
Effective register
Seven Kestrel Fitness definitions are in force or approved with a start date: five effective in the pack and two approved. The live register below lists each with its calculation, source, owner, approver and effective date. A row dated after today, such as Cancellation rate (Oct 5) or Membership revenue per active member (Oct 26), is approved but not yet in a released pack.
Collision rules
A collision is two figures that carry the same name or the same idea but are calculated differently. Four rules apply before any definition reaches Grace:
- One name, one calculation. A figure that counts differently gets its own name.
- One period basis. The register uses four-week periods, so a calendar month figure is restated before it is adopted.
- No overlap inside a ratio. A numerator and a denominator must not count the same members twice.
- A comparison that runs past its due date goes to Arjun Mehta, and the definition stays in Collision check until it is finished.
Same idea, two numbers: Active members and the finance export
The pack counts Active members as paid or frozen memberships at period end, and the finance export counts paid members only. The two figures differed by 640 in September. Under rule 1, Paid members sits beside Active members rather than replacing it. Sofia Reyes owns the proposal, which is in Client review with Grace Whitfield.
Same measure, different period: Visits per active member
Tomas Ferreira submitted it for club operations. It uses calendar months while the register uses four-week periods, and it overlaps Average weekly visits. Adopting it as written would change two pages of the pack. Arjun Mehta asked on Sep 29 for a restatement in four-week periods, and the resubmission must also name which existing measure it replaces.
Overlap inside a ratio: Freeze rate
Frozen memberships sit inside Active members, so the numerator and the denominator overlap. Sofia Reyes is testing Paid members as the base before the definition goes to Grace Whitfield.
Overdue: the Class fill rate comparison
Club operations counts class capacity per room, and the timetable counts it per class. The comparison was due Sep 26 and is overdue because the timetable export is missing two of the fourteen clubs. Tomas Ferreira owns it. Until both counts can be compared for every club, Class fill rate stays in Collision check and no class fill figure appears in the pack.
Recent changes
Kestrel Fitness definitions that moved since August, newest first. Packs affected counts the packs a change touches, and an approved change is not restated into earlier packs.
| Date | Definition | Status | What changed | Packs affected |
|---|---|---|---|---|
| Oct 26, 2026 | Membership revenue per active member | Approved by Grace Whitfield on Sep 30. Starts with the pack for the period beginning Oct 26; earlier packs are not restated. | 3 | |
| Oct 5, 2026 | Cancellation rate | Takes effect with the October pack and replaces Monthly churn rate, which is Superseded with its history kept. | 4 | |
| Sep 30, 2026 | Average membership length | Grace Whitfield asked to exclude cancellations inside the 14 day cooling off window. Sofia Reyes is rerunning September with and without them. | 1 | |
| Sep 29, 2026 | Visits per active member | Restatement in four-week periods requested by Arjun Mehta. | 2 | |
| Sep 21, 2026 | Inactive member share | Grace Whitfield preferred one visit measure, and Visits per active member already covers member activity. | 0 | |
| Sep 7, 2026 | Peak hour check-ins | In force for staffing rotas. Guest passes are included because they use the same turnstile event. | 2 | |
| Aug 24, 2026 | Reactivated members | The 90 day gap that keeps short pauses out of joiner figures was agreed with Grace Whitfield on Aug 20. | 1 | |
| Aug 10, 2026 | New joiners | Returning members are counted in Reactivated members, so the two figures never overlap. | 3 |
Definition card
Every entry in the register carries the same card. The example is Cancellation rate, which Grace Whitfield approved and which first appears in the October pack.
The first use of this definition is held
The cancellation section of the October pack is held in draft until a full finance export is confirmed, so the first Cancellation rate figures will arrive after the rest of the pack. The definition is in force from Oct 5 regardless. Arjun Mehta has told Grace Whitfield, and the held section is released only after the export is verified.
Review cadence
The register is reviewed before every pack is assembled. Anything past its due date is raised with Arjun Mehta the same day.
| Register in use, Q3 2026 (sample figures) | Result |
|---|---|
| Figures disputed by Kestrel | 3 |
| Time to explain a discrepancy | Same day, from the register entry |
| Packs affected by approved changes | 7 (Cancellation rate 4, Membership revenue per active member 3) |
| Onboarding time for a new analyst | Not measured yet |
Who keeps the register
Demo client portal. Northfold Analytics, Kestrel Fitness, the people and every figure are fictional sample data. Built on FuseBase.