Source quality

Triangulate sources before accepting the frame.

Published June 2, 2026. Status: evergreen source page.

A claim can have several citations and still lean on one narrow source trail. Triangulation asks whether the record, the method, the people closest to the event, the counter-source, and the comparison evidence point to the same bounded conclusion.

Use this checklist when a media story, research report, dashboard, or campaign readout feels well sourced but the sources may be repeating one another, carrying different kinds of evidence, or hiding the missing comparison. The goal is not to demand perfect agreement. The goal is to know which parts of the claim are confirmed, which parts are interpreted, and which parts still need a better source.

The triangulation pass

Check	Ask for	Why it matters	Output
Exact claim	The one sentence the reader is being asked to believe.	Sources cannot be compared until the claim is specific enough to test.	Factual, comparative, causal, predictive, or normative.
Closest record	The document, dataset, filing, transcript, method note, or raw report nearest to the fact.	A secondary summary may be accurate, but it should not become the evidence trail by itself.	Primary source visible, missing, or unavailable.
Independent corroboration	A second source that did not depend on the same quote, release, dashboard, or interested report.	Repeated summaries from the same origin are not independent confirmation.	Independent, same-source loop, or unclear origin.
Counter-source	The record owner, affected party, method expert, data owner, or analyst most likely to narrow the claim.	The absent source role is often the one that would reduce confidence.	Included, requested, summarized, or missing.
Denominator	The population, opportunity base, sample base, record universe, time window, or eligible inventory.	A true numerator can still produce a false frame when the base is hidden.	Named, estimated, implied, or absent.
Comparison class	The peer group, prior period, historical range, holdout, matched market, or expected baseline.	Frames usually become persuasive by making one comparison easy and another comparison hard.	Pre-specified, reasonable after-the-fact, weak, or missing.
Method disclosure	Sampling, weighting, model inputs, match rules, exclusions, uncertainty, or data-quality controls.	Method-bearing sources can support method-limited claims, not unlimited conclusions.	Inspectable, partial, asserted, or absent.
Language boundary	The strongest verb and scope that all visible evidence can support.	Triangulation should change the sentence, not only add caveats.	Shows, suggests, is consistent with, or does not establish.

Source distance ladder

Triangulation works best when each source is sorted by distance from the claim. A far source can still be useful, but it should not carry the weight of a close one.

Distance	Examples	Can support	Needs beside it
Closest record	Public record, filed document, transcript, data table, method note, campaign log, or report export.	What was recorded, counted, disclosed, or reported under stated rules.	Denominator, date, scope, and definition notes.
Direct observer	Participant, affected party, operator, buyer, analyst, or person with firsthand operational knowledge.	Observed sequence, practical detail, missing context, or implementation reality.	Representative scope, corroboration, and role clarity.
Method expert	Survey, public-record, economics, analytics, measurement, legal, or technical specialist.	Whether the method can answer the claim being made.	Connection to the specific source material.
Interested source	Sponsor, vendor, advocacy group, platform, campaign owner, or party with direct reputational stake.	Definitions, intent, access, product rules, or stakeholder position.	Independent denominator, counter-source, and method limits.
Secondary analyst	Commentary, newsletter, trade analysis, benchmark summary, or synthesis article.	Pattern recognition, explanation, and comparison ideas.	Links back to inspectable records and method notes.

When sources agree

Agreement is stronger when sources reach the same bounded conclusion from different origins. It is weaker when several citations trace back to the same interested release, the same dataset without the denominator, or the same untested interpretation.

Independent origin

Two sources are stronger when they observed the claim through different records, methods, or roles. A public record plus a directly relevant method expert is usually stronger than three summaries of one report.

Same-source loop

If several articles, quotes, dashboards, or posts repeat one original claim, treat them as one source until the original record and method are visible.

Aligned limits

Real agreement should include the same caveat. If one source shows a narrow record and another writes a broad conclusion, the agreement may be only on the fact, not on the frame.

Shared blind spot

Sources can agree because they share the same missing data: unmatched outcomes, excluded cases, nonrespondents, unmeasured inventory, or unavailable control groups.

When sources conflict

Conflict	First question	Better next source	Careful wording
Different counts	Are they counting the same universe, dates, eligibility rules, and revisions?	Record owner, table note, docket, release history, or data dictionary.	"The count depends on the record universe and release date."
Different rates	Are the numerator and denominator paired correctly?	Official table, subgroup base, population estimate, or opportunity base.	"The rate changes under a different denominator."
Different survey results	Do population, sample source, wording, field dates, weighting, and subgroup bases match?	Full survey disclosure and questionnaire.	"The results are not directly comparable without method alignment."
Different causal stories	Which source shows timing, mechanism, alternate explanations, and counterfactual?	Timeline, pre-period trend, matched comparison, or holdout evidence.	"The timing is consistent with the claim but does not establish cause."
Different campaign results	Which report separates delivery, exposure quality, outcome quality, attribution, and incrementality?	Campaign readout fields, holdout logs, match rules, and uncertainty intervals.	"The report shows observed or credited activity under this method."

Advertising measurement use

Campaign evidence often fails triangulation because each party can see a different part of the path. The publication, platform, advertiser, agency, and vendor may all be telling the truth while none can alone prove causal impact.

Source	Can usually verify	Cannot verify alone	Pair with
Publisher	Eligible pages, placement IDs, package logic, creative delivery, and reader context.	Advertiser-side outcomes or incremental lift.	Advertiser outcome data, test design, and reporting fields.
Ad server or platform	Served impressions, clicks, viewability fields, frequency, and platform-visible conversions.	Unmatched outcomes, cross-channel exposure, and true counterfactual.	Outcome-quality review and attribution-window checks.
Measurement vendor	Match rules, model outputs, survey differences, attribution windows, or lift estimates.	Whether the measured universe represents the whole decision.	Method disclosure, uncertainty, leakage checks, and calibration evidence.
Advertiser or CRM owner	Lead status, sales stage, revenue, duplicate handling, and follow-up coverage.	Media exposure quality or audience selection before exposure.	Placement records, audience eligibility, and comparison rules.
Agency or analyst	Budget context, report structure, decision threshold, and next-action recommendation.	The claim without source records, method notes, and outcome definitions.	All source trails, with the final verb matched to the weakest necessary link.

Downgrade rules

Downgrade when every source traces to the same release, dashboard, stakeholder quote, or report without an inspectable primary record.
Downgrade when the counter-source is present only as a sentence of reaction while the central claim gets documents, charts, and expert interpretation.
Downgrade when the source closest to the number cannot show the denominator, sample base, inclusion rule, or revision date.
Downgrade when a method source can explain a metric but cannot observe the outcome the headline claims.
Downgrade when a conflict is resolved by choosing the more vivid story rather than the closer record or cleaner comparison.

Meeting prompts

Which part of the claim is directly observed, and which part is interpreted?
What is the closest record, and who can independently confirm its scope?
Are multiple citations actually independent, or do they come from the same origin?
Which missing source role would most likely narrow the claim?
What denominator or comparison would make the claim less dramatic?
What final verb fits the weakest necessary source in the chain?

Use with

Pair this checklist with the worked media claim example to see one source trail converted into supportable wording, the source role and incentive map to classify sources, the disconfirming evidence checklist to name the strongest weakener, the media claim audit worksheet for a full article review, the quote weight and response standard checklist when balance is unclear, and the campaign readout QA checklist when a measurement report is being used as evidence.

Takeaway

Triangulation is not a demand for every source to say the same thing. It is a way to keep each source in its proper lane: records establish what was recorded, method notes define how numbers were produced, participants add practical context, experts test interpretation, and counter-sources show what the frame has survived.