data-quality
Skill0 starsChanged 43 days agoMITView on GitHub
Vera's data quality playbook. Covers profiling, rule design across seven quality dimensions, BLOCK/WARN/INFO severity, a metadata-driven PySpark rules engine with quarantine, equivalents for Great Expectations, Soda, dbt tests, Lakeflow/DLT expectations and Deequ, layer-by-layer reconciliation for root-cause analysis, and freshness/volume/schema monitoring. Use when designing DQ checks, validating data, or investigating wrong numbers.
13 other projects use a version of this.
Adopted by
Most recently taken up by
- SSameer20/data-analyst-toolkittheir own version, first seen Nov 8, 2026
- SWEStash/swe-workflow-skillstheir own version, first seen Nov 8, 2026
- Happy-Technologies-LLC/happy-platform-skillstheir own version, first seen Nov 1, 2026
- newsc2/claude-skillstheir own version, first seen Oct 20, 2026
- nimadorostkar/Claude-Skills-collectiontheir own version, first seen Oct 13, 2026
- SkillMedev/data-engineering-pipelinetheir own version, first seen Oct 5, 2026
Other versions13 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 13
- nahid-sparktales/agent-dispatcher★ 59same name
- Happy-Technologies-LLC/happy-platform-skills★ 41same name
- nimadorostkar/Claude-Skills-collection★ 33same name
- brainbytes-dev/everything-claude-trading★ 16same name
- khuynh22/agent-dev-team★ 9same name
- yeaight7/agent-powerups★ 6same name
- SSameer20/data-analyst-toolkit★ 4same name
- gmanch94/claude-code-aidlc-template★ 2same name
- SWEStash/swe-workflow-skills★ 1same name
- Abdox-menara/guard-skills★ 0same name
3 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository4 other files it gives its agents.
- ml-engineeringSkill
- pipeline-engineeringSkill
- setupSkill
- squadSkill
The file
Files
Data Quality Playbook
| Need | Open |
| Profile a table before designing rules | references/profiling.md |
| Metadata-driven PySpark rules engine + quarantine | references/rules-engine-pyspark.md |
| Same rules in Great Expectations / Soda / dbt / DLT / Deequ | references/engines.md |
| "The numbers are wrong", root-cause and reconciliation | references/investigation.md |
| Freshness, volume, schema and drift monitoring | references/monitoring.md |
Seven dimensions (plus schema)
| Dimension | Question | Typical checks |
| Completeness | Is anything missing? | not-null on keys, null % ≤ threshold, expected partitions present |
| Uniqueness | Is anything duplicated? | unique business key, no duplicate loads |
| Validity | Are values allowed? | ranges, enums/accepted values, regex formats, date sanity (not in the future, not before 1990) |
| Consistency | Do related values agree? | cross-column rules (end ≥ start, total = qty × price ± tolerance), same key means the same attributes across sources |
| Timeliness | Is it fresh? | now − max(event_ts) ≤ SLA, watermark lag |
| Volume | Is the amount plausible? | row count within a band vs trailing history, per tenant |
| Referential integrity | Do references resolve? | FK exists in dimension (orphan rate ≤ threshold) |
| Schema | Is the shape as agreed? | expected columns and types, no unexpected new columns (contract) |
Severity
| Level | Meaning | Action |
| BLOCK | Publishing would corrupt downstream data or decisions | Fail the run, don't publish, alert |
| WARN | Suspicious but usable | Publish, log, alert the owner; quarantine failing rows if row-level |
| INFO | Observability metric | Log only; used for trends and tuning thresholds |
A rule starts as WARN while it's new. It becomes BLOCK after two weeks without false positives. This keeps a new rule from stopping production on its first night.
Choosing an engine (when data_quality.engine is blank)
| Stack | Recommend |
| Databricks + Lakeflow Declarative Pipelines | DLT expectations (expect, expect_or_drop, expect_or_fail) + native results table |
| Databricks notebooks / any Spark | Native PySpark rules engine (this playbook), or Great Expectations for rich docs |
| dbt anywhere | dbt tests + model contracts (+ dbt-expectations package) |
| Mixed warehouses, analyst-authored rules | Soda (SodaCL YAML) |
| Very large Spark on AWS/EMR | Deequ / PyDeequ |
Threshold rules of thumb
- Derive from the profile:
threshold = observed p99 (or max) + margin. Record the baseline and date in the rule description. - Volume bands: use a day-of-week-aware trailing 4–8 week median ± k·MAD. It's more robust than mean ± σ.
- Say which rules are business decisions, such as whether negatives are allowed, which statuses are valid, or what SLA the business needs. Those need a human to confirm.
Output contract for every check
rule_id, table_name, dimension, severity, passed, actual_value, threshold, failed_rows, sample_keys, run_id, checked_at
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.