--- name: hansei version: 2.11.0 description: 'Reflect on the improvement loop itself. Examine the trail, surface recurring findings, blind spots, methodology effectiveness, and what the loop has been ignoring. Meta-improvement - improving the improvement process. USE WHEN: reflect, retrospective, what are we missing, blind spots in the loop, why does this keep recurring, hansei, examine the trail, meta-improvement, loop critique, what hasnt worked, why is the score plateauing.' argument-hint: 'Specify the target whose improvement history to reflect on, and optionally a specific concern about the loop' --- # Hansei *Reflect on the loop, not the artifact.* Hansei examines the improvement process itself. While Kaizen improves the target and Kaikaku evaluates whether to replace it, Hansei asks whether the improvement process is working - whether it is measuring the right things, finding the right problems, and actually making progress or just churning. **Part of the suite:** For orchestration, see **Kata**. For improving the artifact, see **Kaizen**. For structural redesign, see **Kaikaku**. For novelty probes that test reasoning quality, see **Shiken**. ## What Hansei Examines ### The Trail Read the improvement history. GENBA entries, Kiroku sessions, prior run reports, changelogs. - What has the loop been finding across runs? - Are the findings getting more specific (progress) or staying at the same level (stagnation)? - Are the same categories of finding recurring? If so, why is the loop not resolving them? ### Recurring Patterns Look for findings that keep appearing across multiple runs: - Same finding, same words: the fix did not work, or the fix was cosmetic. - Same class of finding, different instances: the root cause is structural and the loop is treating symptoms. - Finding that was marked fixed but returned: the change was reverted or the fix was incomplete. For each recurring pattern: what would it take to resolve this permanently? Is it a Kaizen problem (fix the thing) or a Kaikaku problem (the structure makes this inevitable)? ### Blind Spots What has the loop NOT been looking at? - Are there aspects of the target that no run has ever examined? - Is the loop anchored to a particular analytical lens and missing others? - Has the loop been evaluating the artifact as a document but never testing it as a tool? - Has the loop been measuring input quality (how good is the skill file?) but not output quality (how good is the work the skill produces)? - Are there observer classes (from Principle 2) whose perspective has never been represented? ### Intent Drift Has the system drifted from its original purpose? - Compare the current Target Condition (if stated) against what the loop is actually optimizing for. Are they still aligned? - Have the measurements evolved so far from their original derivation that they no longer measure what was intended? - Is the loop optimizing for its own metrics rather than the human's actual goal? - Have goals, constraints, or assumptions changed since the loop started — and has the loop adapted, or is it still chasing the original framing? ### Retirement What no longer earns its existence in the improvement process itself? - Are there goals that were achieved long ago and no longer need tracking? - Are there metrics or measurements that no longer discriminate — they've been at ceiling so long they add overhead without signal? - Are there constraints that were once relevant but no longer apply (e.g., compatibility with a tool version that's been retired)? - Are there tests, checks, or verification steps that have never failed and never will — cost without value? Retirement is not deletion from history — it is moving things from "active" to "archived" so the loop stays lean. The trail records what was retired and why. ### Methodology Effectiveness Is the improvement process itself well-designed? - Is the scoring rubric measuring what matters? - Are the diagnostic lenses revealing real problems or generating false positives? - Is the loop producing genuine improvements or just reshuffling? - Would a different approach (different rubric, different focus, different evaluator) produce better results? ### Trajectory Plot the trajectory. Not just scores over time, but the character of the work: - Early runs: broad findings, large improvements. Expected. - Middle runs: findings become more specific, improvements become smaller. Progress. - Late runs: findings are cosmetic, improvements are marginal. Approaching convergence - or approaching the structural ceiling. - Stagnation: score oscillates, findings cycle, no clear direction. The loop may need Kaikaku on itself. ## Output Hansei produces a reflection, not a fix. Its output is: 1. What the loop has been doing well 2. What the loop has been missing 3. Whether the loop's methodology is still appropriate or needs adjustment 4. Specific recommendations for the next cycle (focus area, methodology change, rubric revision, etc.) Hansei does not modify the target artifact. It modifies the loop's approach for the next iteration. ## Evidence An observer reading the kiroku session after this skill completes should be able to find: - **What was examined.** Which loop aspects were looked at — trail history, recurring patterns, blind spots, intent drift, retirement candidates, methodology, trajectory. What was *not* examined and why. - **Recurring patterns.** For each: what it is, how many times it has appeared, and the root cause assessment (structural or fixable?). - **Blind spots surfaced.** What the loop has not been looking at. If nothing found: explicit statement with reasoning. - **Trajectory characterization.** Whether the loop is progressing, approaching convergence, stagnating, or showing signs it needs Kaikaku on itself. - **Recommendations.** Specific, actionable guidance for the next cycle. These are Hansei's artifact — it does not modify the target. Mark decisions `[!DECISION]`, realizations `[!REALIZATION]`, reversals `[!REVERSAL]` in the session as they occur.