Report a suspected data-quality issue

A focused handoff guide for people who use shared data.

Use this guide to describe a result that appears inconsistent, explain what you expected it to mean, and give the responsible owner a safe way to compare it. It covers data-quality routing—not privacy or security incident reporting.

Fictional example: Northstar Supply is a cross-functional organization using a shared weekly orders dataset. Its scenario illustrates how to make a report reproducible; it does not describe a real investigation.

First, identify what you are seeing

Compare the observation with the intended meaning before routing it.

  • A likely quality discrepancy A weekly total changes when you rerun the same report with the same date range and region filter. Record both results and the comparison conditions.
  • A definition difference One report counts orders placed while another counts orders shipped. Check which event the measure is intended to represent before treating the difference as an error.
  • An interpretation question You are unsure whether a field includes cancelled records. Name the field and the decision you need to make, then ask the dataset owner to clarify its meaning.

Prepare a reproducible handoff

Separate what you observed from what you expected.

  1. 1

    Name the data

    Identify the dataset, report, table, and field using the names shown in the resource you used.
  2. 2

    State the observation

    Record the result, when you saw it, the comparison period, and the filters or grouping applied.
  3. 3

    Explain expected meaning

    Say what you expected the value to represent and the definition, report note, or business context behind that expectation.
  4. 4

    Describe a safe reproduction

    Write the steps, query description, or report settings another authorized colleague can use to compare the result. Do not include credentials or secrets.
  5. 5

    Identify the likely owner

    Start with the dataset owner for meaning and ownership questions. A data platform team may help with shared pipeline or refresh behavior; a reporting analyst may help with report filters and presentation.

Evidence boundaries

An authored reference for preparing a safe, useful handoff—not a live queue or system of record.

Evidence boundaries
Useful evidence Keep out of the report
Dataset and field names; comparison period; filters or grouping; observed result; expected meaning and its basis; steps or a query description needed to reproduce. Passwords, tokens, or other credentials; secrets; unrestricted exports; personal records or identifiable data copied when they are not needed.

Worked example: Northstar weekly orders

A fictional example showing how to compare like with like.

Fictional example
What was observed

A Northstar Supply analyst sees a weekly orders total of 1,240, then gets 1,198 on a rerun. Both runs use the same stated week and region filter. The expected measure is orders placed during that week, not orders shipped.

  • First result: 1,240 orders.
  • Rerun result: 1,198 orders.
  • Comparison: same week and same region filter.
  • Expected meaning: count by order placement date.
Reproducible, not conclusive
Why this is a useful handoff

The matching date range and region filter give the owner a specific comparison to reproduce. The difference merits investigation but does not establish a root cause.

  • Ask the dataset owner to confirm the measure definition and intended date field.
  • Share both results and the report settings through the established data-support route.
  • Do not assume whether refresh behavior, source changes, or another factor explains the difference.

Who may help

Start with the role closest to the question. The dataset owner is the first point of accountability for meaning and ownership; other roles can help when the evidence points to platform behavior or report presentation.

Dataset owner
Clarifies the dataset’s intended meaning and ownership, then identifies the appropriate next handoff.
Data platform team
Investigates shared pipeline or refresh behavior when the issue may affect more than one report or data consumer.
Reporting analyst
Checks report filters, grouping, and presentation when the underlying data may be represented differently.

Route the prepared evidence

Use your organization’s established data-support route for the relevant dataset.

  • Provide the dataset and field names, observed result, comparison conditions, expected meaning, and safe reproduction steps.
  • Ask the owner to confirm the interpretation and advise on the next step.
  • If the concern involves personal-data exposure, unauthorized access, or a security event, use the organization’s established privacy or security reporting route instead.

This guide does not submit a report, assign an owner, or promise a response time.