| name | data-analysis |
| description | Experiment result analysis with reproducible metrics, charts, and Korean decision summaries. Triggers: analyze results, compare metrics, visualize data, experiment analysis. |
data-analysis
Usage
Use this skill when comparing runs, interpreting metrics, or producing analysis-ready artifacts from result files.
Responsibility
This skill owns analysis workflow only:
- Scope: data loading, metric comparison, visualization, and report-ready summaries.
- Out of scope: model retraining, production deployment, or architecture refactor.
Path Policy
- temp/: temporary merged datasets and scratch notebooks/scripts
- documents/drafts/: interim analysis narrative and open questions
- documents/final/: final Korean analysis report and decision summary
- documents/reference/: reusable analysis methodology and metric definitions
Protocol
- Define analysis question
- Clarify baseline, success metrics, and comparison window.
- Collect and normalize inputs
- Load results from canonical result/log sources.
- Normalize schema and units before comparison.
- Compute metrics
- Calculate primary and secondary metrics consistently.
- Include uncertainty or variance when available.
- Visualize
- Use readable plots with clear labels and units.
- Avoid decorative charts that hide magnitude.
- Interpret
- Distinguish observation from recommendation.
- Explain practical impact in Korean for documents/ outputs.
- Publish artifacts
- Keep temporary calculation files in temp/.
- Move narrative into documents/drafts/, then finalize in documents/final/.
Output Contract
Each delivery contains:
- Data scope and baseline
- Metric definitions used
- Key plots or tables
- Main findings and limits
- Recommended next experiment or action
Quality Bar
- Reproducible metric calculations
- Explicit assumptions and filtering rules
- No conclusion without numeric evidence
