Data quality & reconciliation with exception

**Quick judgement:** Strong, well-scoped skill for structured data reconciliation where stable identifiers exist and auditability is critical. It is particularly suited to payroll/compliance-style domains that demand deterministic joins, explicit exception reasons, and hard pipeline gates. **Strengths:** - Clear matching priority (Pay Number → Driver Card → Driving Licence → DQC) with deterministic, non-fuzzy logic. - Exhaustive categorization of records (matched, missing, duplicate, mismatch, invalid) with standardized reason codes. - “No silent failure” design: unmatched/variance thresholds and count checks can stop pipelines rather than quietly degrading data. - Good usability signals: concrete input requirements, output schema, and example prompts. **Key risks / limitations:** - Not suitable when you lack stable identifiers or when fuzzy, open-ended matching is required (e.g., messy name/address-only linkage). - Requires the user to specify tolerances, normalization rules, and ID priority where multiple identifiers exist; poor configuration will degrade results. - Performance and scaling characteristics are not specified; very large datasets may need additional tuning or infrastructure considerations. **Recommended scenarios:** - Weekly or scheduled reconciliation between payroll, HR, and compliance/driver registers, with audit-ready exceptions reports. - Building governance-focused data pipelines where any record loss, unexpected duplicates, or unexplained variance must be explicitly flagged. - Creating operational data quality scorecards with clear thresholds, red flags, and exception routing for human review. - Environments (e.g., regulated industries, transport/logistics, finance) where silent join failures or untracked mismatches are unacceptable.