---

name: comparison-and-quality description: "Guides fg-data-profiling report comparison, privacy-safe reports, metadata, data dictionaries, quality outputs, and expectation-suite caveats." disable-model-invocation: true metadata: disco-role: operating license: MIT

Comparison and Quality

Use this sub-skill when the user wants to compare datasets/reports, protect sensitive data in a profile report, attach dataset metadata or data dictionaries, inspect quality outputs, or discuss expectation-suite integration.

Read first

• Read references/comparison-workflows.md for compare([...]), profile_a.compare(profile_b), labels, constraints, and comparison output handling.
• Read references/privacy-and-metadata.md for sensitive=True, samples=None, custom synthetic samples, phone-number dtype risks, dataset metadata, column descriptions, and type_schema.
• Read references/quality-outputs.md for get_description(), JSON output, alerts, and legacy Great Expectations notes.
• Read references/troubleshooting.md for comparison errors, privacy leaks, and expectation-suite dependency failures.
• Run scripts/compare_reports_smoke.py or scripts/sensitive_report_smoke.py for safe no-network checks of comparison/privacy guidance.

Comparison quick start

from data_profiling import ProfileReport, compare

train_profile = ProfileReport(train_df, title="Train", minimal=True)
valid_profile = ProfileReport(valid_df, title="Validation", minimal=True)
comparison = compare([train_profile, valid_profile])
comparison.to_file("train-vs-validation.html")

ProfileReport.compare(other) is an alias for comparing two reports:

comparison = train_profile.compare(valid_profile)

Reports use config.title as labels. Tune report.precision and html.style.primary_colors when comparison tables are cramped.

Privacy quick start

profile = ProfileReport(
    df,
    title="Private profile",
    sensitive=True,
    samples=None,
    duplicates=None,
    minimal=True,
)
profile.to_file("private-profile.html")

For reports that need a sample section, replace real records with a synthetic sample:

profile = ProfileReport(
    df,
    sensitive=True,
    sample={"name": "Synthetic sample", "data": synthetic_df, "caption": "Synthetic rows only."},
)

This is redaction guidance, not enterprise PII detection. The open-source package does not automatically manage organization-wide PII classifications.

Metadata and quality quick start

profile = ProfileReport(
    df,
    title="Dataset profile",
    dataset={"description": "5% reproducible sample", "creator": "Data team"},
    variables={"descriptions": {"amount": "Transaction amount"}},
    type_schema={"segment": "categorical"},
)
summary = profile.get_description()
json_text = profile.to_json()

Boundaries

• Basic report creation belongs in ../profiling-workflows/SKILL.md.
• YAML/settings mechanics belong in ../configuration-and-output/SKILL.md.
• CLI command shapes belong in ../cli-and-automation/SKILL.md.
• Optional Great Expectations installation/version questions route to ../integrations-and-backends/SKILL.md.