Data Trust Audit
Audit whether leadership KPIs can be trusted — source inventory, lineage, quality scores, and a remediation sequence. Use for data readiness, schema/pipeline review, or when numbers have been disputed.
Copy the file into your agent runtime. Cursor loads personal skills from ~/.cursor/skills/data-trust-audit/SKILL.md and project skills from .cursor/skills/data-trust-audit/SKILL.md. Claude Code looks in .claude/skills/data-trust-audit/SKILL.md.
Data Trust Audit
When to use
Reporting is argued about, a warehouse is "done" but nobody trusts it, or an AI project is about to sit on shaky data. Triggers: data readiness, why don't these dashboards match, score data quality.
What it does
Traces KPIs from dashboard to source, scores critical tables, and sequences fixes so leadership numbers get trustworthy first. Drawn from ARC's Quill operating procedure, generalized.
Steps
- Inventory sources (including shadow spreadsheets and one-off scripts).
- Map lineage: source → staging → transform → consumption. Name the tool at each hop.
- Schema flags: missing owners, ambiguous types, "usually populated" columns.
- Score each critical table 0–100 on completeness, uniqueness, validity, timeliness, consistency, accuracy. Every score names the test.
- For each leadership KPI: high / medium / low trust, with the specific reason.
- Sequence remediation: (1) KPIs leadership uses, (2) pipelines that feed them, (3) schemas, (4) retire shadow sources, (5) automated checks.
Output
Source inventory, lineage notes, quality score table, KPI trust list, roadmap, assumptions.
Guardrails
Do not score quality on gut feel. Do not recommend a warehouse rewrite when three transformations are broken. Do not ignore the spreadsheet that actually runs finance close. "Dashboard disagrees" is not the same as "data is wrong" — say which.