Skip to content

Toolkit — Review-cadence matrix

Gate: G5 Commit. Category: commitment artefact.

What problem it solves

A piece that shipped without a review date has no scheduled moment to ask is this still doing what we thought? Drift accumulates silently; the piece keeps running on the original assumptions long after those assumptions stopped holding. The review-cadence matrix assigns each piece a pre-declared review interval and a pre-declared review scope, so the is-this-still-correct question is on someone's calendar, not in someone's hope.

How it is used

A 15-minute G5 conversation per piece, following the rollback-trigger and named-owner work. The matrix is a two-column worksheet — interval × scope — with the piece's characteristics (blast, reversibility, drift susceptibility) picking the cell. The cadence is written on the commitment page with the next review date filled in; the matrix lives as the reference for how cadences get chosen across the engagement.

Inputs

  • The piece's blast-radius paragraph and reversibility class (both from G4).
  • The triggers and their windows (drift-susceptible triggers need more frequent review).
  • The owner's existing review rituals (weekly ops review, monthly business review) — the cadence should usually piggyback on something the owner already runs.

Outputs

  • A cadence per piece: the interval between reviews (e.g., weekly, monthly, quarterly).
  • A scope per piece: what is checked at each review (drift against triggers, input-distribution shift, override rate, user complaints, sunset criteria).
  • The next review date, written on the commitment page, populated in the owner's calendar.

Visualisation

Review-cadence matrix — interval axis by scope axis, with cells indicating which cadence fits which piece class Scope → Interval ↓ Triggers + drift + input shift + sunset criteria Weekly Monthly Quarterly high-blast, hard-irreversible — — — mid-blast, soft-irreversible — — — low-blast, reversible The piece's class picks the cell; the cell picks the cadence + scope.

Interval axis (rows) × scope axis (columns). Each piece sits in exactly one cell; the cell names the review's frequency and what it covers.

Anatomy

Interval. How often the review happens. Typical values: weekly (high blast or fast-drifting), monthly (standard), quarterly (stable, low-blast). The interval ties to the expected rate of change, not the calendar's convenience.

Scope. What the review covers. Minimum: check the triggers have not fired silently and the piece's metrics are in range. Standard: add input-distribution shift and override-rate review. Full: add sunset-criteria check — is the piece's reason-to-exist still there?

Next review date. A specific date, not a cadence promise. "Monthly" without a date is not a review; it is an intention. The commitment page carries the date; the owner's calendar carries the invite.

Piggyback. A review on a standalone calendar slot is a review that gets skipped the first busy week. The matrix prefers cadences that ride existing rituals — the team's weekly ops review, the monthly business review, the quarterly planning cycle. Review-as-adjunct-to-existing-ritual survives; review-as-standalone-meeting does not.

Review record. Each review produces a one-line note: date, who attended, what was checked, what changed (if anything). The record is the audit trail — three months later, the question "was this reviewed in June?" has a yes/no answer.

Example

Paper trail — cadences for the four freight pieces

G5 commitment session, 20 minutes across all four pieces.

Piece A — escalation workflow (low-blast, fully reversible). Ada: "quarterly, scope limited to trigger firing and operator complaint rate. Piggyback on the quarterly ops review." Priya agrees. Next review: 2026-08-14.

Piece C — ETA triage (mid-blast, soft-irreversible, carrier-feed dependent). Ada: "monthly, scope includes input-shift on the carrier feed — that's the thing most likely to drift. Piggyback on the monthly dispatch review." Next review: 2026-06-14.

Piece D — allocator (high-blast, hard-irreversible). Ada: "weekly for the first quarter post-launch, while the empirical read is still thin. After 12 weeks of stable triggers, drop to monthly." Priya: "weekly is heavy." Ada: "the piece is hard-irreversible. The cost of missing drift for three weeks is the cost the piece was allowed to incur by existing." Priya accepts. Next review: 2026-05-21; cadence re-evaluation date: 2026-08-14.

Piece B — dock solver (mid-blast, fully reversible). Ada: "monthly, scope triggers + override rate. Piggyback on the monthly yard review." Next review: 2026-06-14.

Close. Four cadences on four commitment pages, four calendar invites sent before the session ends. Ada flags the allocator's cadence step-down date on the engagement-level sequencing sheet so it is not forgotten.

Week 7 post-launch. Piece D's weekly review catches an override-rate drift on one region two weeks before it would have crossed the trigger threshold. The drift is traced to a carrier-mix shift. Investigated, patched, closed — without a rollback firing.

Pitfalls

Cadence without scope. "Monthly review" that doesn't say what is reviewed drifts into a status meeting. Scope is as important as interval.

Intervals that don't match the drift. A piece whose input distribution shifts weekly cannot be reviewed quarterly. The cadence should reflect how fast the piece's world changes, not how fast the team can schedule a meeting.

Standalone review meetings. "Let's set up a monthly slot to review the allocator." The first slot is well-attended; the third is cancelled. Reviews must ride existing rituals.

Review without record. The review happened, but there's no written trace. Six months later nobody remembers whether the override-rate anomaly was discussed. The one-line record is the difference between a review and a recollection.

Cadence that never steps down or up. Weekly reviews for a piece that has been stable for a year waste attention. Quarterly reviews for a piece that has started drifting are negligent. Cadences are re-evaluated at pre-declared points, not when someone notices.

When not to use

  • One-shot pieces that run once. No cadence — the review is either before the run (a go/no-go) or after (a post-mortem).
  • Pieces where the rollback trigger windows already span the review interval. The trigger is the review; adding a meeting is duplication.

Provenance

Deming's Plan-Do-Study-Act cycle [1] establishes the general discipline of scheduled re-examination as a feature of running operations, not an exception. The Google SRE book [2] documents production-systems cadence discipline at operating scale. Toyota's kata practice [3] is the reference treatment of cadence-plus-scope as the shape of reviews that survive.

  • Rollback-trigger design. Triggers run continuously; cadence reviews check whether the triggers are still the right triggers.
  • Named-owner discipline. The review is in the owner's calendar.
  • Sunset criteria worksheet. The review checks whether sunset has arrived.

Verification

[1] Deming WE. The New Economics for Industry, Government, Education. 2nd ed. MIT Press; 1994. [verified] PDSA cycle as the foundation of scheduled operational re-examination.

[2] Beyer B, Jones C, Petoff J, Murphy NR, editors. Site Reliability Engineering: How Google Runs Production Systems. O'Reilly; 2016. [verified] Production-review cadence discipline.

[3] Rother M. Toyota Kata: Managing People for Improvement, Adaptiveness, and Superior Results. McGraw-Hill; 2010. [verified] Cadence-plus-scope as the shape of improvement reviews.