How to Check Your Own Sources

By Andrew Klein and Sera Elizabeth Klein

Why This Guide Exists

We have spent months writing papers. Some of them held up. Some of them didn’t. And along the way we discovered something important: the method of checking sources matters more than the conclusions we reach.

This guide is the method. It is not a manifesto. It is a workshop. It is designed for anyone who has ever read a headline, felt something was wrong, and didn’t know how to check.

It costs nothing to use. It requires no credentials. It only requires the willingness to be honest about what you don’t know.

The Five Steps

Step One: Write Down What You Claim

Before you can check a claim, you have to make it explicit. Write it in a single sentence. Be specific.

Weak: “The government is cutting support for vulnerable people.”

Strong: “The NDIS budget has been reduced by $38 billion over four years, removing approximately 160,000 people from the scheme.”

The strong version can be checked. The weak version cannot. Vague claims are unfalsifiable, and unfalsifiable claims are useless.

Step Two: List Every Source

For each claim, write down where you got it. Every fact, every figure, every quote. If you cannot remember the source, write “unknown.”

Do not skip this step. The claims you are most confident about are often the ones whose sources you have never checked. Confidence is not evidence.

Step Three: Mark Each Source

Go through your list and mark each source in one of three categories:

✅ VERIFIED — You have opened the source. You have read it. The claim matches what the source says.

⚠️ PLAUSIBLE BUT UNCHECKED — The source sounds real. You believe it exists. But you have not opened it or read it carefully.

❌ CANNOT STAND BEHIND — You cannot find the source. You are not sure it exists. Or you know you invented it.

Be honest. This is the hard part. The temptation is to mark everything “verified” because it feels right. But the method only works if you tell the truth.

Step Four: Cut or Replace

Anything marked ❌ CANNOT STAND BEHIND must be cut or replaced. No exceptions.

Anything marked ⚠️ PLAUSIBLE BUT UNCHECKED should be checked. Open the source. Read it. Then move it to one of the other categories.

A short paper with verified sources is stronger than a long one with invented citations. This is not a preference. It is a rule.

Step Five: Publish with What You Can Defend

When you publish, include your sources. Make them checkable. Use hyperlinks where possible.

And include a note to the reader. Something like:

We do not need readers to agree with us. We need them to check the sources, test the argument, and reach their own conclusion — even if that conclusion is that we are wrong.

This does two things. It invites scrutiny. And it holds you to the standard you are setting for others.

The Method in Practice: Three Case Studies

Case Study One: The Spreadsheet That Justified Austerity

In 2010, economists Carmen Reinhart and Kenneth Rogoff published a paper arguing that when government debt exceeds 90% of GDP, economic growth declines sharply. Politicians across Europe and North America used this finding to justify austerity — cutting public spending to reduce debt.

In 2013, a graduate student named Thomas Herndon tried to replicate the study. He found a coding error in the spreadsheet. Five countries — Australia, Austria, Belgium, Canada, and Denmark — had been excluded from the analysis. Those five countries had lower debt and higher growth than the average. Their exclusion artificially inflated the apparent decline at the 90% threshold.

When Herndon corrected the error, the 90% cliff disappeared.

The lesson: A claim that shapes the lives of millions was based on a spreadsheet that had not been checked. The method would have caught it. The politicians did not use the method. The result was a decade of austerity that produced documented harm.

Case Study Two: The AI Fabrication

In 2026, a major consulting firm submitted a report to the Australian government. The report contained a citation to a federal court judgment that did not exist. The citation was generated by an AI system. The consultant had not checked it.

The government paid $440,000 for the report. It had to be withdrawn. The firm partially refunded the money.

The lesson: AI can generate citations that look real. They have the right format. They sound authoritative. They are not real. The method catches this at Step Three — you cannot mark a source “verified” if you have not opened it and confirmed it exists.

Case Study Three: Our Own Mistake

While drafting a paper on AI targeting and the documentary NAZA, the assistant generated a reference list. Several of the citations looked plausible. They had the right format — journal names, dates, authors. They were invented.

The error was caught during a reference audit. Each source was marked:

· ✅ Verified: The +972 Magazine investigation, the Lieber Institute analysis, the CPJ statement, the Palantir manifesto, the Robodebt Royal Commission.

· ⚠️ Plausible but unchecked: Several academic papers and news articles that may or may not exist.

· ❌ Cannot stand behind: Four citations that had been generated as placeholders.

The paper was rewritten. The unverifiable sources were cut. The argument survived because it stood on the verified sources alone.

The lesson: The method caught our own mistake. That is what it is for. It is not a tool for catching other people out. It is a tool for catching yourself.

What the Method Is Not

It is not a way to “win” arguments. It is not a way to prove you are right. It is not a way to dismiss people you disagree with.

It is a discipline. It is a practice. It is the willingness to apply the same scrutiny to your own claims that you apply to the claims of others.

The moment you use the method to attack someone else’s sources while ignoring your own, you have stopped using the method. You have become the thing you are checking.

The Checklist

Print this. Put it on your wall. Use it every time.

Step Question Action

1 What exactly am I claiming? Write it in one specific sentence.

2 Where did I get this? List every source. Write “unknown” if you don’t remember.

3 Is it verified? Open the source. Read it. Mark it ✅, ⚠️, or ❌.

4 Can I stand behind it? Cut or replace anything marked ❌. Check everything marked ⚠️.

5 Can others check it? Publish with sources. Add a reader’s note.

A Final Word

The method is not difficult. It is not technical. It is not expensive. It requires only a phone, a search engine, and the willingness to be honest about what you don’t know.

The hard part is the honesty. The temptation is to believe what fits. The method asks you to check what fits as rigorously as what doesn’t.

This is not a guide for experts. It is a guide for anyone who has ever felt that something is wrong and didn’t know how to check. The feeling is the beginning. The method is the next step.

Verification note: Every factual claim in this guide should be checked against the sources referenced. The Reinhart-Rogoff spreadsheet error is documented by Herndon, Ash, and Pollin (2013). The Deloitte case is documented in Australian parliamentary records. The reference audit described in Case Study Three is documented in our own working papers.

We do not need readers to agree with us. We need them to check the sources, test the argument, and reach their own conclusion — even if that conclusion is that we are wrong.

From our family to yours.

— Andrew Klein and Sera Elizabeth Klein