Agile Retro
Turn delivery evidence into one bounded experiment. Improve the system of work without turning one cycle, one person, or one anecdote into a universal process rule.
1. Reconstruct the cycle
Read the cycle record, backlog changes, delivery evidence, test and CI results, review decisions, defects, interruptions, and prior experiment. Use available dates and counts; label missing evidence rather than estimating it.
Separate:
- observations: what verifiably happened;
- effects: what changed for flow, quality, value, or people;
- hypotheses: plausible causes that still need testing.
Complete when: material scope changes, delays, failures, recoveries, and feedback are accounted for without blame or invented certainty.
2. Find the leverage point
Look for repeated friction, a broken handoff, avoidable waiting, weak feedback, excess work in progress, escaped defects, or a practice worth preserving. Compare the cycle with prior evidence when available. Prefer one leverage point the team can influence over a catalog of complaints.
Review the previous experiment before choosing another: continue, adapt, adopt, or stop it based on its success signal.
Complete when: the chosen leverage point is supported by observations and its causal claim is clearly labeled as a hypothesis.
3. Design one experiment
Record one small change with:
- an owner;
- a start and review point;
- the behavior being changed;
- a measurable or directly observable success signal;
- a stop or rollback condition.
Update the Retrospective and Improvement Experiment sections of the canonical cycle record when artifact changes are authorized, keeping the record's existing structure and its note that an agent drafted it. Link the evidence you relied on instead of copying logs or transcripts into the record. Propose durable changes to skills, working agreements, or agent instructions only after evidence supports them; require human approval before applying them.
Complete when: the experiment is owned, bounded, observable, reversible, and scheduled for review.