principle · compounding small changes
Compounding small changes
Tighten the chain by exactly one notch each cycle. The notch from the previous cycle is still in place when the next one runs.
A retrospective produces eight ideas. The team writes them down. The list is shared. The next cycle begins. By Tuesday, no one remembers the list. By the following retrospective, the team produces another eight ideas — many of them similar to the previous batch.
This is the most common shape of we are not improving. It is not laziness. It is the gap between the artefact a list produces and the artefact a change produces. Lists do not compound. Changes do.
The discipline
The corpus's retrospective rule:
One change. Owned. Dated. Testable.
One. Not eight. Not three. One. The team that picks one change and applies it for a full cycle has invested more than the team that picked eight and remembered none. Owned by name — not by the team, not by the PO, by Alex. Dated — not next quarter, but 2026-06-15. Testable — measurable by an outsider in the next cycle's artefacts. We will be more careful at amigos fails the test. Amigos for every story is scheduled within 24 hours of the story being pulled, by the PO who pulled it passes.
Why one is enough
Three reasons.
One change actually happens. The team has finite attention. A list of eight gets one done by accident. A list of one gets one done deliberately. The difference compounds — five cycles later, the team that picked one per cycle has five changes in place. The team that picked eight per cycle has approximately… one or two.
One change is debuggable. When something else gets worse, the team can trace it back to the change they made. With eight changes, attribution is impossible.
One change is rehearsable. The next retrospective opens with did the previous change hold? what did it produce? This is impossible with a list — did all eight hold? is not a question anyone has time to answer honestly.
What "compounds" actually means
Cycle one: the team adopts amigos within 24 hours of pull. By cycle two, this is normal. The retrospective's energy goes elsewhere — say, the weekly client update goes out Friday by 4pm.
Cycle two: amigos is now habitual; the client cadence change is the new investment. Cycle three: both are habitual; the cycle's change targets something further upstream.
By cycle five, the team has five practices the rest of the industry does not have. Each one is small. Together they describe a chain that improves quarter over quarter in a way that lists never produce.
When compounding breaks
Two ways.
The change is not testable. We will communicate better survives no audit. The next retrospective cannot tell whether it happened. The cycle silently undoes it.
The change is too big. We will adopt the full UDOO chain is not a change; it is a re-org. It cannot land in a cycle. Break it down — we will write a feature brief for the next initiative lands. We will adopt UDOO will not.
The corpus's antidote to both: if you cannot describe what success of the change looks like in one sentence by the time the next retrospective runs, the change is not yet ready.
See also
- Canon — After We Build · The Retrospective
- Canon — Did We Serve · Adoption
- Principle — Chain-level thinking