co-growth-learning-synthesis

name	co-growth-learning-synthesis
description	Learning synthesis skill. Experiment result interpretation with Scale/Iterate/Kill decision framework, effect size review, and next hypothesis generation. Use when INTERPRETING experiment results, deciding to scale or kill a test, reviewing effect sizes, or generating next-iteration hypotheses.

Learning Synthesis

Experiment result interpretation and Scale/Iterate/Kill decisions.

Use This Skill When

• Interpreting completed experiment results.
• Deciding whether to Scale, Iterate, or Kill an experiment.
• Reviewing effect sizes and confidence intervals.
• Generating next-iteration hypotheses from learnings.

Workflow

1. Load experiment results (primary, secondary, guardrail metrics).
2. Check statistical significance (α ≤ 0.05) and practical significance.
3. Review effect size and confidence interval width.
4. Check guardrail metrics for degradation.
5. Apply decision framework:
   • Scale: significant lift, no guardrail violations → ship to 100%.
   • Iterate: directional signal but CI crosses zero → refine hypothesis.
   • Kill: no signal or guardrail violation → stop and document.
6. Generate next hypotheses based on learnings.
7. Save decision and rationale to files.

Deliverables

• report.md: result summary, decision, and next steps.
• results/experiment-results.md: metrics, CIs, decision rationale.

Quality Gates

• Statistical significance checked with correct α threshold.
• Effect size and 95% CI are reported (not just p-value).
• Guardrail metrics reviewed for degradation.
• Decision uses Scale/Iterate/Kill framework explicitly.
• At least one next hypothesis is generated.

If any gate fails: identify the issue, fix, and re-validate.

Gotchas

• A p-value of 0.049 with a wide CI is not a strong signal. Always report CI width alongside significance.
• "Iterate" is not "re-run the same test." Each iteration must change at least one variable.
• Novelty effects inflate early results. If runtime was <2 weeks, flag potential novelty bias.

Validation Loop

1. Generate synthesis and decision
2. Check: significance, effect size + CI, guardrails, Scale/Iterate/Kill applied
3. If any check fails → revise analysis
4. User approval before shipping or killing ⏸️