What It Does

Audits data freshness, completeness, and blind spots before any analysis or decision-making. Prevents acting on stale data, missing metrics, or incomplete context.

The Pattern

Every data-driven decision has a hidden dependency: the data being current and complete. This skill formalizes the verification step that most people skip.

The audit runs in three layers:

1. Data Freshness. For each data source, check when it was last updated and compare against acceptable staleness thresholds. Different sources have different tolerances:

• Real-time metrics: acceptable if less than 1 hour old
• Daily aggregates: acceptable if less than 24 hours old
• Weekly reports: acceptable if less than 7 days old
• Manual snapshots: note the exact date and flag if older than expected

2. Completeness. Look for gaps in the data:

• Missing time periods (did collection fail for certain days?)
• Incomplete event tracking (are all expected events actually firing?)
• Broken pipelines (did an integration stop sending data?)

Red flags: sudden drops to zero (broken tracking), metrics that never change (stale cache), conflicting numbers between sources that should agree.

3. Blind Spots. Identify what you can't measure:

• Are there user segments with no data? (e.g., mobile vs. desktop breakdown)
• Are there behavioral patterns you're not tracking? (e.g., cohort behavior beyond day 7)
• Is qualitative feedback integrated or siloed?

The audit protocol runs before any report or recommendation:

1. Check data freshness for every source you'll reference
2. Verify completeness by looking for anomalies
3. Cross-reference sources that should agree
4. Note limitations explicitly in your output

Remediation actions are concrete:

Gap Type	Action
Stale data	Re-run collection scripts
Missing tracking	Add events to the instrumentation
No baseline	Establish a benchmark now
Conflicting sources	Determine which is the source of truth

Key Decisions

Audit before analysis, not after. Running the audit after you've already written a report creates sunk cost bias. You're less likely to scrap conclusions if you've already invested time in them.

Explicit staleness thresholds. Don't use vague terms like "recent" or "up to date." Define exact acceptable ages per data source. This makes the audit automatable.

Note limitations in the output. Every report should include a section on what the data can't tell you. This prevents overconfident conclusions from incomplete data.

When to Use It

• Before starting any analytics or reporting work
• Before publishing recommendations to a team
• When metrics seem off or unexpected
• As a monthly hygiene check on data infrastructure
• When making decisions that depend on data quality