super-analysis

Super Analyze

This skill is the short-form entry point for the Super Analyze workflow.

If the plugin command /super-analysis:run is available, prefer it because it is written to use structured confirmation prompts. Use this skill when the user calls /super-analysis directly.

Invocation

/super-analysis path/to/data.csv

If the user does not provide a dataset path, ask for one before continuing.

Required Workflow

You must complete these phases in order:

1. Run the deterministic scan helper.
2. Present the scan report.
3. Get user confirmation on questionnaire, ID column, condition column, design, and dependent variables.
4. Run the deterministic recommendation helper.
5. Present the recommendation cards.
6. Get user confirmation on the method for each dependent variable.
7. Generate the analysis script and companion outputs.

Helper Commands

Prefer the repository helpers over ad-hoc reasoning:

.\myenv\Scripts\python.exe .\scripts\super_analyze.py scan <path>

.\myenv\Scripts\python.exe .\scripts\super_analyze.py recommend <path>

If .\myenv\Scripts\python.exe does not exist, fall back to python.

Phase 1: Scan

The scan report must include:

• detected questionnaire
• confidence
• concrete evidence
• subject ID column
• condition column
• design label
• dependent variables
• stop conditions

Stop Conditions

Do not guess past these cases. Ask the user instead.

• Questionnaire confidence is low.
• Subject ID column is unclear.
• Condition column is unclear.
• The sheet looks like a wide or multi-block export.
• Participant condition coverage is inconsistent.
• The design is ambiguous.

Phase 2: Recommendation

After the user confirms the scan, run the recommendation helper and present one card per DV with:

• descriptive statistics by condition
• normality result per condition
• sphericity or Levene result when applicable
• recommended method
• alternative method
• rationale
• risk note

Method Rules

Use these defaults unless the user overrides them:

Design	Parametric	Non-parametric
2 conditions, within	Paired t-test	Wilcoxon signed-rank
2 conditions, between	Independent t-test	Mann-Whitney U
>= 3 conditions, within	Repeated-measures ANOVA	Friedman
>= 3 conditions, between	One-way ANOVA	Kruskal-Wallis
Multi-factor	Two-way ANOVA	ART

Output Rules

The primary output is the script:

• analyze_xxx.py

Also generate when relevant:

• cleaned_scored.csv
• summary.txt
• figures/*.png

Every major code block in the generated script must label:

• repository function(s)
• statistical method
• why it is appropriate
• fallback rule

Red Flags

Stop and correct the workflow if any of these happen:

• You are about to run omnibus analysis without user-confirmed methods.
• You are about to write the script without traceability comments.
• You are treating auto-detection as ground truth.
• You are using a parametric method after failed assumptions without user approval.