after we build · part five
The Retrospective
Three questions, in order. One change — specific, owned, testable. The discipline of compounding rather than listing.
The retrospective is calendar-locked at the cycle's close. Trio plus anyone whose work is being examined. Inputs: the signal reading and any postmortems. The retro doesn't re-litigate them — it learns from them.
This is where meaning's journey through the cycle is evaluated honestly — not as a feelings session, but as a comparison between what was predicted and what happened.
Without written predictions before the cycle, this is impossible. A reconstructed prediction is a description. Only a prediction written before the cycle can be honestly evaluated.
Three questions, in order
1 · Which chain links held?
Name the practice, not just the outcome. "The amigos session caught the concurrent submit edge case" is a link holding. "Things went well" is not. Name what worked so it is consciously preserved.
2 · Which chain links broke?
Trace to a level, not a person. Scenario gap? Observation mismatch? ADR drift? Prediction not checked? Each broken link names a structural gap with a structural fix.
3 · What is the one change we are making?
One. Not a list. Specific enough to test, owned by a named person, with a measurable outcome and a check date. "We should communicate better" is not a change. "A template prompt added by Thursday, checked at the next retrospective" — that is.
Enough to compound rather than list.
One change is named, owned, and dated. A retrospective that produces more than one change produces zero changes — because five action items implemented zero is feelings. One action item implemented is compounding. The volume's argument hangs on the difference.