Run Tight, Predictable Sprints

Proof required

Submit: (1) a baseline metrics report covering 4–6 sprints: sprint completion rate per sprint and the average, cycle time or equivalent flow metric, and the distribution of incomplete work by cause (blocked, scope-added, underestimated, deprioritised); (2) a root cause diagnosis (200 words minimum) identifying the 1–2 specific causes that account for the majority of missed commitments, with supporting data; (3) pre-committed improvement targets for each planned intervention — written before the intervention begins, not after seeing results.

What gets checked

• Baseline draws from actual sprint data — '60% completion on average' requires the per-sprint data showing how the average was calculated; a single 'typical sprint' estimate is not a baseline
• Root cause diagnosis names specific causes with proportional data — '80% of missed commitments came from these three causes' with each quantified is a diagnosis; 'we need better planning' is not
• Pre-committed targets are specific and measurable — 'improve completion rate from 60% to 75% within 8 sprints of implementing the change' is a target; 'improve delivery' is not

Common mistakes

• Diagnosing root cause without data — a list of likely causes that matches prior beliefs, without sprint-level data, is a hypothesis not a diagnosis
• Choosing the easiest metric to show improvement rather than the most meaningful — if the real problem is missed commitments, cycle time improvement that leaves completion rate unchanged does not address the stated problem
• Setting targets after seeing post-intervention results — the target must be documented before the intervention starts

Resources

What a verifier looks for

• Baseline data quality: ask the submitter to show the per-sprint data — a single average figure has no variance information and is not a real baseline
• Root cause specificity: ask the submitter to name the top cause of missed commitments and the percentage it accounts for — a diagnosis that cannot answer this is too vague
• Target pre-commitment: ask when the improvement target was written down — if it was after the first post-intervention sprint results, it was not pre-committed

Proof required

Submit: (1) a pre-intervention plan written before implementation: the specific change (process, cadence, tooling, or stakeholder boundary), why it addresses the M1 root cause, and the sprint it went live; (2) post-intervention sprint data for 4–6 sprints using the same metrics from M1; (3) a brief notes log for each post-intervention sprint capturing any confounding events — team changes, holidays, major incidents — that might explain metric changes independently of the intervention.

What gets checked

• Intervention is specific enough to replicate — a reader not present should be able to implement it from the description; 'we improved estimation' is not replicable; 'we introduced three-point estimation for all stories above 3 points in the last two days before sprint planning' is
• Post-intervention data uses the same metric definitions as M1 baseline — if the definition changed, document why
• Confounding events log is honest — a 6–10 week period with zero confounding events is unusual; team changes, holidays, and incidents should be noted

Common mistakes

• Implementing multiple changes simultaneously and attributing improvement to one cause — if estimation and stakeholder boundaries both changed in the same sprint, neither can be isolated
• Switching to a different metric after the intervention produces disappointing results — the M2 data must use the M1 metric
• Running only 2–3 post-intervention sprints — too much variance to distinguish real effect from random fluctuation; 4–6 sprints is the minimum

Resources

What a verifier looks for

• Intervention specificity: ask what an engineer experienced differently in sprints after the intervention — if the answer is vague, the intervention was not specific enough to produce interpretable results
• Metric consistency: compare metric definitions in M1 and M2 — if they differ without explanation, ask why
• Confounding events: ask whether anything unusual happened in the post-intervention sprints; a 6–10 week period with no team changes or holidays is unusual

Proof required

Submit: (1) a before/after comparison table: each M1 pre-committed target, the baseline value, the post-intervention value, and the delta; plus a 200-word attribution analysis — what you attribute to the specific intervention vs. other factors; (2) a lessons-learned document (300 words minimum): what you would do differently in the intervention design, what the root cause analysis got right or wrong, and whether the improvement is likely to persist; (3) documentation of a real-time Q&A session with a named reviewer who has ≥3 years of engineering delivery leadership experience and a verifiable professional profile — the documentation must include specific challenge questions and your responses; the reviewer must not be your current manager or a direct report.

What gets checked

• Before/after table uses M1 pre-committed targets — the table must include the target value; if the target was not met, explain why rather than celebrating a different metric that improved
• Attribution analysis acknowledges plausible alternative explanations — a delivery improvement coinciding with two new team members cannot be fully attributed to the intervention without addressing that confound
• Q&A documentation shows the reviewer asked attribution questions — look for questions like 'why do you attribute this to your intervention rather than the team composition change in sprint 3?'

Common mistakes

• Reporting improvement against a target added after the intervention results — check that the M3 comparison table target matches the one documented at the end of M1
• Attribution analysis that only credits the intervention — every engineering team improvement has confounding variables; the analysis must name the two or three most plausible alternatives and explain the attribution
• A reviewer who validates the result without challenging methodology — 'yes this is a good improvement' without challenging attribution, target-setting, or intervention isolation does not meet the adversarial standard

Resources

What a verifier looks for

• Target fidelity: compare M3 targets with M1 documented targets — if they differ without explanation, ask when the targets were set
• Attribution honesty: ask the submitter to name the most plausible alternative explanation for the improvement that is NOT their intervention — inability to name one means the attribution analysis is not credible
• Reviewer adversarial quality: look for methodology challenge questions about attribution and intervention isolation, not only process questions