About This Project

TruthBased.org is an independent, data-driven research project — not affiliated with any political party, advocacy group, think tank, or media outlet. No advertising. No sponsored content. Primary sources, disclosed confidence levels, and transparent limitations.

What TruthBased.org does

We take politically charged claims that are widely repeated and test them against the strongest available evidence. We don't pick easy targets. We pick claims where smart, reasonable people disagree — and then we show what the data actually says, who it comes from, and how confident we can be in it.

If a conclusion changes because the evidence changes, we update the article and log the correction publicly on this page.

Source hierarchy

Not all sources are equal. We rank evidence in tiers and always prefer the highest-quality source available for a given claim.

Source hierarchy tiers used in TruthBased.org analysis
TierSource typeExamples
Tier 1 Primary government data, peer-reviewed research, court records, official statistics FBI UCR, CDC WONDER, DOJ prosecution records, published RCTs, appellate court rulings
Tier 2 Systematic reviews, institutional reports from nonpartisan organizations Cochrane reviews, RAND Corporation, CBO analyses, Cass Review, university research centers
Tier 3 Reputable journalism, investigative reporting, expert commentary AP, Reuters, major newspaper investigations, named expert interviews
Tier 4 Advocacy organizations, think tanks, opinion pieces Heritage Foundation, ACLU, Cato Institute, editorial boards

When Tier 4 sources are the origin of a claim (as with the Heritage Foundation's voter fraud database), we use them as the starting dataset but verify against higher-tier sources. We never treat advocacy data as settled fact.

Strongest-case testing

Every position is presented at its strongest before it’s evaluated.

A strawman is the weakest version of an argument, easy to knock down. A steelman is the strongest version — the version a smart, well-informed advocate would actually make. Each article includes a dedicated section presenting the best case for each position, using the strongest available evidence and most charitable interpretation.

This is not false equivalence. After presenting each side at full strength, the evidence is weighed and the conclusions follow the data. But the reader sees the full picture first.

Confidence levels

Not all conclusions carry the same certainty. We label our findings using three confidence tiers so readers know exactly how strong the underlying evidence is.

HIGH Multiple independent, high-quality sources converge on the same conclusion. Large sample sizes. Replicated findings. Example: the base rate of voter fraud relative to ballots cast. MODERATE Evidence points in a clear direction but has notable limitations — small samples, observational designs, or limited replication. Example: long-term projections of cancer deaths from very low radiation doses. LOW Evidence is sparse, conflicting, or methodologically weak. We report what exists but flag the uncertainty prominently. Example: long-term detransition rates beyond 5 years.

What we disclose

Every article includes the following, by default:

Standard disclosures in every article

Limitations: What the data cannot tell us. Where gaps exist. What studies haven't been done.

Falsifiability: What evidence would change our conclusion. Every article states what would need to be true for the opposite conclusion to hold.

Scope: What the article covers and what it intentionally excludes, so readers don't assume silence equals endorsement or denial.

Database details: Where the data comes from, what time periods it covers, and how we counted.

How we handle disagreement

When credible sources reach different conclusions, we don't pick the one we prefer. We report the disagreement, explain why the sources diverge (different methodologies, different populations, different time frames), and let the reader see the full landscape. Where the weight of evidence favors one side, we say so — but we never pretend consensus exists where it doesn't.