id: data-quality-gate title: Data-Quality Gate Patrol category: data primitive: /loop use_when: You want a patrol that checks a dataset/table against quality rules and blocks or flags on violations. verify: Every flagged row violates a stated rule; a clean pass reports zero violations against the rule set. tags: [data-quality, validation-gate, patrol]

Data-Quality Gate Patrol

A patrol loop that runs a frozen rule set over a dataset and flags every violation — a quality gate, not a vibe check. Primitive: /loop · Category: data

Use when

A table feeds downstream jobs and you want continuous assurance it meets quality rules (no nulls in keys, ranges, referential integrity) — with violations flagged precisely, not summarized.

Prompt

Goal:        Enforce the data-quality rule set on <dataset> each pass.
Context:     The dataset; a FROZEN rule set (key non-null, ranges, FK integrity, uniqueness).
Constraints: Read-only. Flag only rows that violate a stated rule; cite the rule per flag.
Done-when:   This pass reports zero violations, or every violation is flagged with its rule.
Evidence:    A violations ledger (row id, rule, value); the pass summary (counts per rule).
If-blocked:  If a rule can't be evaluated (missing column), report it; don't pass silently.

Verify

A separate check confirms each flagged row actually violates the cited rule, and that a "clean" pass genuinely produced zero violations across the whole rule set.

Steps

1. Load the dataset and the frozen rule set.
2. Evaluate every rule; record violations with the rule cited.
3. Summarize counts per rule; block/alert on threshold breaches.

Notes

Freeze the rule set the way the deliverability example freezes its rubric — the gate must be stable across passes, or "improvement" is unmeasurable.