A reproducible search records the choices behind the results. USAspending's official data-sources guide identifies filters including action date, award type, awarding agency, place of performance, recipient, and industry. Product Service Codes have their own official taxonomy resources. Use filters as a documented research method rather than assuming one keyword captures an entire market. Read the official data-sources guide; review the official PSC taxonomy resources.
Write a search specification
Include the business question, subject, included award types, date field, date range, geography meaning, agency role, likely keywords, and candidate codes. State the output measure and the row level. If you cannot describe these choices in a paragraph, narrow the question before downloading a large file.
Fictional Coastline Engineering studies federal stormwater work. Its keyword list includes several terms found in its own service descriptions. The researcher then inspects relevant official award descriptions to discover alternative wording. Candidate codes are treated as a research hypothesis, not as proof that every matching record is relevant or every nonmatching record is irrelevant.
Run broad and narrow passes
- Discovery pass: use a few broad terms to learn vocabulary.
- Code pass: inspect candidate code definitions and representative records.
- Focused pass: combine the justified filters for a defined population.
- Challenge pass: look for known relevant examples your method missed.
- Review pass: manually classify a small sample for relevance.
Keep a search log with each pass, filters, result count, file name, and lesson learned. Change one major dimension at a time when diagnosing a surprising result. If you simultaneously alter keywords, date range, and agency filters, you will not know which change explains the difference.
Test both false matches and missed matches
Coastline samples 20 records from a fictional export. Twelve clearly concern its service area, five use a similar word for unrelated equipment, and three need more information. It records those categories instead of declaring every hit addressable demand. Next, it searches for several known relevant examples to test whether the filters are too narrow.
Use a small relevance rubric: clear match, adjacent work, nonmatch, and uncertain. Write one sentence explaining ambiguous classifications. A second reviewer can inspect the uncertain group and a few obvious matches. The purpose is to understand the search's practical limits, not to create an artificial precision score from a tiny sample.
Save the exact recipe
Keep original exports, filter notes, code definitions, date of access, and the calculation steps together. Include the unit of analysis and amount field in the file index. If a system permits a saved search or direct result link, retain it alongside the written recipe; links alone may not preserve every choice indefinitely.
Finish with a statement such as: “This method identifies a defined set of potentially relevant contract actions; manual review found these limitations.” That is a usable research result. It supports better follow-up without implying complete market coverage, current opportunity availability, or guaranteed future purchasing.
Official sources & further reading
Sources checked for this edition: October 26, 2026. Recheck the current source and applicable contract before acting.
Keep the next step concrete
Use the working tool to capture your assumptions, supporting sources, and the decision that still needs review.