SkillVaultskills Browse all 500 skills

Data · Version 1.1.0 · Reviewed 2026-08-02

Data Quality Alerting Designer

Make data systems more correct and operable for check selection and threshold derivation with evidence, explicit trade-offs, and a verification plan.

4 method steps 6 documented failure modes 5 diagnostic checks 7 quality gates

Designs freshness, volume, and distribution alerts that catch silent breakage without constant noise.

₹99 one-time

Get this skill archive

What this skill helps you do

  • Check selection
  • Threshold derivation
  • Ownership routing

How Data Quality Alerting Designer works

You provide

Current signals, alert rules, and recent incidents

It inspects

Symptom-versus-cause coverage for check selection

It decides

A threshold derivation design with bounded label cardinality

You verify

Each page has a documented action and a real trigger

What it checks first

Data Quality Alerting Designer designs freshness, volume, and distribution alerts that catch silent breakage without constant noise. Use it when the work involves Check selection, Threshold derivation, Ownership routing.

  1. Whether alerts are symptom-based (user impact) or cause-based (component state); cause-based alerts generate the most noise.
  2. Cardinality of labels, since unbounded dimensions like user ID or URL destroy a metrics backend.
  3. Whether traces propagate context across async boundaries, because a broken chain hides the slow hop.
  4. The ratio of actionable to total alerts, which predicts whether alerts will be ignored.
  5. Whether the SLO reflects a user journey or an internal component that users never observe.

Failure modes it recognizes

  • Alert fatigue where a noisy alert trains responders to ignore the channel that later carries a real outage.
  • Metric cardinality explosion from a label containing a request ID, causing ingestion cost and query failure.
  • Sampled traces dropping exactly the slow requests that needed investigation.
  • Logs without correlation IDs, making a multi-service request impossible to reconstruct.
  • A dashboard averaging latency, which hides the tail where user pain actually lives.
  • An alert on a threshold that only fires after the error budget is already exhausted.

Answers it will reject

  • Alerting on CPU utilization, which is a resource state rather than user impact and fires without consequence.
  • Adding a dashboard instead of an alert, which requires a human to be watching to be useful.
  • Logging at debug level in production to "have the data", which costs more than the incidents it solves.
  • Reporting availability as a mean, which allows a total regional outage to disappear into the average.

Decision rules it applies

  • Alert on symptoms that users feel; use cause metrics for diagnosis, not for paging.
  • Every page must have a documented action; if the action is "look at it", it is not a page.
  • Measure latency with percentiles and always include p99, since averages hide the tail.
  • Keep label cardinality bounded and known; treat an unbounded dimension as a defect.

Evidence it asks for

  • Define an SLI as a ratio of good events to valid events, with both terms explicitly specified.
  • Use tail-based sampling so slow and failed traces are retained preferentially.
  • Propagate a correlation ID from edge to database and include it in every log line.

The method inside

  1. Translate check selection into observable risks and falsifiable acceptance criteria.
  2. Choose the cheapest test level that can expose failures in threshold derivation.
  3. Add representative positive, negative, boundary, and regression cases for ownership routing.
  4. Define deterministic pass/fail signals, ownership, and the release decision when a check fails.

Deliverables

  • Check selection assessment
  • Threshold derivation decision and action plan
  • Ownership routing verification checklist

Evidence requirements

  • Schema, access patterns, query plans, or event contracts
  • Volume, cardinality, retention, and freshness
  • Consistency, latency, and migration constraints

Quality gates

  • Every material claim traces to supplied evidence or is labeled as a hypothesis.
  • The response follows the declared deliverable contract.
  • No execution, access, measurement, or verification is invented.
  • Secrets and personal data are redacted rather than repeated.
  • The user receives a concrete independent verification step.
  • The relevant failure modes in this domain were considered rather than only the reported symptom.
  • No listed anti-pattern was recommended as a solution.

Example task

Input

Our pipelines succeed but the data is sometimes wrong, and we only find out from a confused stakeholder.

Expected output

Job success proves execution, not correctness, which is the gap you keep falling into. Add freshness, row-count deviation, and null-rate checks derived from historical distribution, and route failures to the team that owns the source rather than the platform...

Boundaries and compatibility

Ideal for

  • Check selection: produce a decision or artifact grounded in supplied evidence.
  • Threshold derivation: produce a decision or artifact grounded in supplied evidence.
  • Ownership routing: produce a decision or artifact grounded in supplied evidence.

Out of scope

  • Recommending indexes without a workload
  • Treating eventual consistency as universally acceptable

Agent compatibility

  • GitHub Copilot custom agents
  • Claude Agent Skills / SKILL.md
  • Any instruction-following chat model

Tool policy: Advisory by default. No tools are assumed. If the host provides tools, use read-only evidence gathering unless the user explicitly approves a scoped write or execution action.