Public records

Public records and denominator checklist

Published May 30, 2026. Status: evergreen source page.

Public records can make a media claim more inspectable, but a record is not the same thing as a denominator. Before trusting a story built from incident logs, filings, government tables, budgets, inspections, dockets, or open-data portals, check what the record covers, what it omits, and what baseline would make the number meaningful.

This checklist is for readers, editors, analysts, and researchers who need to slow down claims that sound precise because they cite a public source. The goal is not to dismiss official records. The goal is to keep the claim within the population, time period, definitions, and revision limits that the record can actually support.

One-page record check

Question	Strong disclosure shows	Weakens confidence
What is the original record?	The story links the filing, table, transcript, docket, budget line, inspection record, release, or download page closest to the claim.	The citation points to a summary, screenshot, thread, or paraphrase while the underlying record remains hard to inspect.
Who maintains it?	The agency, court, office, data owner, reporting program, or record custodian is named with its role.	The source is described only as "public records" or "government data" without naming the collector.
What is included?	Eligibility rules, geography, population, reporting units, fields, filters, and known exclusions are visible.	A count is treated as universal even though only reported, filed, inspected, adjudicated, or eligible cases enter the record.
What is the denominator?	The claim gives the total population, eligible cases, transaction base, agency workload, prior-period level, or rate base needed for interpretation.	A numerator, anecdote count, percent change, or "highest since" phrase carries the frame alone.
What time period is covered?	Record date, event date, release date, fiscal/calendar period, access date, and update cadence are separated when they matter.	A lagged, revised, partial, or event-date dataset is written as if it were final and current.
Are definitions stable?	The story notes definition changes, reporting-rule changes, boundary changes, category changes, and classification uncertainty.	A trend line crosses a definition break without warning readers.
What comparison is fair?	The article compares like geographies, similar institutions, comparable periods, rate-adjusted counts, or pre-specified benchmarks.	The comparison shifts between national, local, peer, historical, and anecdotal frames depending on which sounds strongest.
What could be missing?	The article names underreporting, nonresponse, redaction, exemptions, backlog, selection into enforcement, and records not created.	The absence of a visible record is treated as proof that nothing happened, or one visible record is treated as the full universe.

Read public-record claims in order

1. Name the record type

A complaint log, arrest record, inspection report, court filing, administrative dataset, budget table, open-data export, and press release answer different questions. Decide what the record is before deciding what it proves.

2. Separate event date from record date

Many records are created after the underlying event. A spike in filings, releases, or uploads may reflect processing, backlog, policy, or reporting changes rather than a spike in the underlying condition.

3. Convert counts into rates when the base changed

If population, transactions, inspections, staff, eligibility, or reporting volume changed, raw counts can mislead. Ask whether the right comparison is per person, per case, per dollar, per transaction, per inspection, or per eligible unit.

4. Treat missing records as a source problem

Missing records may mean no record exists, the wrong agency holds it, the record is exempt, the search was narrow, the request is pending, or the data system does not capture the event. Do not turn absence into proof without checking the record process.

5. Keep revisions near the claim

Official data often changes after initial release. If the claim depends on a fresh estimate, preliminary table, or incomplete period, the headline should preserve that limit.

Denominator translation table

If the story says	Ask for	Careful wording
"Incidents rose 40%."	The starting count, ending count, total eligible population, time window, and reporting-rule changes.	Reported incidents rose from X to Y in this record over this period; the rate and reporting context determine how large the change is.
"The agency received hundreds of complaints."	Total transactions, customers, cases, inspections, or contacts during the same period.	The record shows X complaints among Y relevant contacts or cases; complaint volume alone does not show prevalence.
"This is the highest level in years."	The full series, definition changes, population changes, and whether the latest point is preliminary.	The record reaches its highest reported level since year Z under the available series; comparability limits remain visible.
"The city spends more than peers."	Per-capita, per-household, per-case, inflation-adjusted, and service-level comparisons.	The city spends more on this line under this rate base; the service mix and accounting definitions need checking.
"Only a small share was affected."	The denominator used, who was excluded, and whether the excluded group could still matter to the decision.	The measured share is small under this denominator; a different eligible base may change the practical meaning.

Public-record gap language

When a story cannot yet inspect the best record, say that plainly. "Records were not available" is not the same as "records do not exist." "The agency did not provide records by publication time" is not the same as "the agency has no support for the claim." Clean public-record writing separates request status, legal access, agency custody, search scope, and evidence strength.

Status	What it means	What not to infer
Record requested	A requester has asked for records, but release, redaction, denial, or appeal may still be pending.	That the requested record exists or will support the frame.
Record withheld or redacted	Some information was not released under an access rule or exemption.	That the hidden material proves the strongest interpretation.
No responsive records	The custodian did not identify records matching that request and search scope.	That no relevant event occurred, no other office holds records, or no different search would find material.
Partial dataset	The record covers a subset, period, geography, or reporting program.	That the subset represents the whole population without adjustment.

Minimum citation packet

Original record link, table number, docket, filing ID, release, data portal, or request response when available.
Record owner, reporting program, geography, population, record type, release date, event date, and access date.
Numerator, denominator, time window, filters, exclusions, and whether the figure is preliminary, revised, estimated, or final.
Definition notes, reporting-rule changes, missing fields, redactions, underreporting risks, and comparability breaks.
Comparison class: prior period, peer group, population-adjusted rate, eligible-case base, or stated benchmark.
One disconfirming record or base rate that would materially weaken the public frame if it appeared.

Official references

FOIA.gov's request guide is a starting point for federal public-record requests and request status language. U.S. Census Bureau citation guidance is useful when official tables, tools, technical documents, or research need a replicable citation path. BLS data retrieval tools and the BLS information guide help readers find labor, price, productivity, and economic data sources. BEA data pages are useful for economic accounts, regional data, and national-account context.

Takeaway

A public record makes a claim easier to inspect, not automatically true in the way a headline suggests. Strong public-record writing names the source, shows the denominator, explains the missing universe, and keeps comparisons inside the record's limits.