data-quality-observability-reviewlisted

# Data Quality Observability Review Produce one **Data Reliability Contract** that tells producers, consumers, and operators whether a dataset or projection is fit for its declared decisions and actions, what has degraded, and how to recover without silently publishing stale or incorrect truth. ## Atomic boundary Own generic dataset/pipeline identity, producer-consumer contracts, lineage, freshness, completeness, validity, uniqueness, referential and semantic invariants, distributions, reconciliation, quality states, alerts, quarantine, backfill/replay, correction, consumer impact, and recovery proof. Own only the lineage needed to determine fitness and impact; do not expand this artifact into a general provenance or custody system. Read [references/data-reliability-contract.md](references/data-reliability-contract.md) for quality states, check selection, reconciliation, and backfill patterns. ## Workflow 1. Inventory the critical data products, producers, transformations, stores, consumers, decisions/actions, owners, latency expectations, retention, sensitivity, and blast radius of missing, late, duplicated, or wrong data. 2. Define grain, keys, schema, units, timestamps, ordering, null/unknown meaning, source authority, lineage, version, compatibility, and consumer contract. 3. Select checks from the failure modes that can change a consumer decision: freshness, volume/completeness, validity, uniqueness, referential integrity, distribution, semantic inva