Run a ten-minute award-data quality check

Published by Federal Bid Partners LLC · October 26, 2026 · Data quality

Public award records are valuable evidence, but a dataset still needs a quality check for the question at hand. FAR 4.606 describes reportable actions and exclusions, including classified information and other categories. A blank search result therefore does not establish that an agency never buys a service. Define the coverage you can support before interpreting absence. Read the reporting scope and exclusions.

Check the file before the chart

Record the source, download date, filter choices, row count, and documented row level. Inspect the first and last relevant dates. Check whether the selected period is complete and whether the source's disclosures affect your use. Do not compare a partial current year with an entire prior year without making the difference visible.

In a fictional review, Ridgeway Consulting compares two exports and sees a steep decline. The first export covers twelve months; the second covers only six. After matching the windows, the pattern is less dramatic. This is a research-design correction rather than evidence that either source was wrong.

Use four quick tests

• Missingness: count blanks in the identifiers, dates, amounts, and organization fields needed for your question.
• Uniqueness: test the documented identifiers and inspect repeated-looking records before deleting anything.
• Consistency: compare units, field labels, date meanings, and agency roles.
• Reconciliation: reproduce one small total and inspect several records at the official source.

Separate a missing value from zero. Zero may be a real reported amount; blank may mean unavailable, not applicable, or a processing issue. Preserve the original value and record your treatment. If a missing organization code prevents a mapping, place it in an unknown group instead of silently assigning it to the largest agency.

Be cautious with joins

Joining public records by names alone can introduce false matches. Use documented identifiers and test the result with a small sample. A 2026 GAO report discusses interoperability challenges across federal eligibility and award-related data. That is a practical reason to inspect linkage quality rather than assuming two public systems match perfectly. Read GAO's interoperability findings.

Ridgeway's fictional join increases 100 source rows to 160. The researcher checks whether that increase represents a valid one-to-many relationship or duplicate matching. She does not immediately remove 60 rows to restore the expected count. Instead, she documents the relationship and selects the aggregation rule appropriate to the question.

Write a limitation that changes interpretation

Useful limitations are specific: “Twelve records lacked the office code used for this ranking,” or “This comparison uses matched six-month action-date windows.” Avoid generic caveats that tell the reader nothing. State whether the problem could change the ranking, amount, coverage, or recommended action.

Finish with a short quality note: what passed, what remains uncertain, and which conclusions are still supported. If the uncertainty could reverse the decision, gather more evidence before recommending action. If it affects only a small detail, narrow the claim and proceed with that boundary clearly recorded. The goal is proportionate confidence grounded in the actual records.