compare-datasets

Skill: Compare Datasets

Purpose

Compare metrics, findings, and patterns across two or more connected datasets. Helps identify cross-dataset patterns (e.g., "conversion funnel behavior is similar across both product lines") and dataset-specific anomalies.

When to Use

• User says /compare-datasets or "compare across datasets"
• After analyzing multiple datasets, to find commonalities
• When the user asks "is this pattern unique to this dataset?"

Invocation

/compare-datasets — compare active dataset with all others /compare-datasets {id1} {id2} — compare two specific datasets /compare-datasets metric={name} — compare a specific metric across datasets

Instructions

Step 0: Feasibility Check (NEW)

Before running the full 6-step workflow, verify the comparison is possible:

1. Check minimum requirements:

   • At least 2 datasets must be connected (check .knowledge/datasets/)
   • If only 1 dataset: "Only one dataset connected. Use /connect-data to add another."

2. For data-intensive comparisons (e.g., retention, conversion funnels):

   • Verify temporal span: Each dataset needs sufficient history (e.g., 3-6 months for retention)
   • Verify sample size: Small datasets (<30 rows) may not produce meaningful comparisons
   • Verify schema compatibility: Key fields must exist in both datasets

3. If comparison is BLOCKED:

   • State why upfront: "Cannot compare retention — Dataset B has only 27 days of data vs. Dataset A's 12 months. Retention requires 3-6 months minimum."
   • Provide decision criteria: "We can compare when Dataset B has [specific requirement]."
   • Skip to Step 6 (Present Results) with a diagnostic report instead of attempting analysis

Rationale: This saves time and avoids misleading partial comparisons. Better to diagnose "why we can't compare" upfront than to discover data gaps in Step 4.

Step 1: Identify Datasets to Compare

1. Read .knowledge/datasets/ to enumerate all connected datasets.
2. If specific datasets are named, validate they exist.
3. If no datasets specified, use active + all others.
4. List datasets with their connection status (connected vs. not connected).