Money and markets produce more claims in a day than anyone can check. Deposits are insured up to some amount. The dollar is dying. Inflation is the highest it's ever been. Some are true, some are half true, and some can't be tested at all. They all sound equally sure of themselves.
This module gives you a five-step method and three exercises to practise it. You'll sort real claims into fact, opinion and untestable, walk one claim through every step with IMF data, and learn why finding a connection between two things tells you much less than it feels like it does.
Before you check anything, decide what kind of claim you're looking at.
A checkable fact names something a record can settle: a number, a date, a rule, a count. "The FDIC insures deposits up to $250,000" is one. It can be right or wrong, and someone keeps the record.
An opinion weighs things. "The central bank should cut rates" mixes facts about prices and jobs with a judgment about which matters more. Data can inform it. Data can't decide it.
A claim is untestable as worded when no result could ever count against it. "A crash is coming" has no date, size or market, so it can't lose. These are often fixable. Add a date, a number and a source and you have a fact claim you can check.
Try it on eight claims. Four have a record. Two are judgments. Two have no way to lose. Sort each one, then read what would settle it.
Eight everyday claims about money and markets. Pick a category, then see what would settle the claim and what the record says. The four fact claims were checked against primary sources in November 2026.
1“US bank deposits are insured up to $500,000.”
2“The dollar's share of world reserves has been falling.”
3“Inflation in the euro area peaked above 10%.”
4“The Federal Reserve's balance sheet is bigger now than it has ever been.”
5“The central bank should have cut interest rates sooner.”
6“Paying off debt is always smarter than investing.”
7“The banking system is rigged against ordinary savers.”
8“A major financial crash is coming.”
Two of the fact claims were wrong, and one of them was wrong in a way that sounds careful. "The Fed's balance sheet is bigger than it has ever been" sounds like a statement about now. The record shows a peak in April 2022 and a lower number today. The claim fails on the date.
Notice also that two of the true claims need a caveat before you repeat them. Falling share isn't falling amount, and inflation easing doesn't mean prices dropped. A claim can be true and still mislead.
1. Pin the claim. Rewrite it with a number, a date and a unit. If you can't, it's probably an opinion or untestable as worded. If you can, you may have narrowed what the speaker said. Note what you dropped.
2. Find the primary source. Go to whoever counts the thing: a central bank, a statistics office, a regulator, a court filing, the company's own report. An article about the number is a pointer, not a source.
3. Check the unit and the date. Share or amount? Percent or percentage points? Millions or billions? Month average or year end? How old is the number, and was the series rebuilt since? Most "wrong" numbers turn out to be right numbers in a different unit or from a different year.
4. Compare with a base rate. A number alone says little. Ask what usually happens. How big is a normal year-to-year move? How often would this show up by chance? A 2-point fall means one thing in a series that moves 0.5 points a year and another in one that moves 5.
5. Separate fact from interpretation. Write down what the data shows in one column and what the speaker concludes in another. Words like "collapsed", "abandoned" and "rigged" almost always sit in the second column. Judge them separately.
Here's the method on one claim. It uses the dollar's share of the world's official reserves, which is the kind of number people often quote in a bigger argument.
Step through the five steps. The chart shows the dollar and euro as a share of official foreign exchange reserves at the end of each year, 2000 to 2025.
The claim, as you might see it online: “The world is abandoning the dollar. Its share of reserves has collapsed.” Step through the method.
“Abandoning” and “collapsed” are not measurements. Turn it into something with a number, a date and a unit: the dollar's share of the world's official foreign exchange reserves is lower at the end of 2025 than at the end of 2000.
You may have lost some of the speaker's meaning. That's fine. You can now check what's left, and the leftover words (“abandoning”) are what you'll judge in step 5.
Result: the narrow claim is true and the strong one isn't supported. The share fell 13.22 points over 25 years, from 69.74% to 56.52%. That's real, and a reader who says "it's not falling" is wrong. But a long, uneven decline from a very high level isn't a collapse, and the data says nothing about who is "abandoning" what. Step 5 is where you catch that.
People usually test a claim by collecting examples that fit it. Each example feels like proof. It isn't, because examples that fit are the ones you'd see whether the claim is true or not. The useful question is the other one: what would show this is wrong, and have I looked? That pull toward agreeing examples has a name: confirmation bias.
Psychologists call a classic version of this the Wason selection task. Most people fail it the first time, so don't worry if you do. The cards below use a money claim.
Four funds, one side visible each. Choose the cards you must turn over to test the claim, then check.
Claim: “Every fund that advertises double-digit returns has hidden fees.” Each card below is one fund (invented for this exercise). One side says whether it advertises double-digit returns. The other says whether it has hidden fees. You can only see one side of each. Which cards must you turn over to test the claim?
Most people pick "Advertises double-digit returns" and "Has hidden fees". The second card can't hurt the claim. A fund with hidden fees fits whether or not it advertised anything. You learn nothing by turning it. The card that can break the claim is "Has no hidden fees", because behind it might be a fund that did advertise double-digit returns. Few people pick it.
Real claims work the same way. When someone shows you five cases that fit their story, ask for the cases that don't. If they never looked, the five tell you nothing.
There's a related trap. You find two people who went to the same school, two companies that share a lawyer, a donation that came before a decision. It feels like a discovery. Whether it is depends on a number most people never look up: how often would a connection like this appear if nothing were going on? That's the base rate.
Pairs multiply fast. Ten people make 45 possible pairs. A hundred people make 4,950. Each pair is a chance to find something shared. And if you're willing to count any kind of link (school, employer, hometown, a board seat), every kind is another chance on top.
Play with it below. Start with the birthday case: in a group of 23 people, with 365 equally likely birthdays, what are the odds two of them share one?
The formula gives the exact chance that at least two people in the group share a value. The buttons draw random groups so you can see the same thing happen. Everyone in these groups was drawn at random, so every match is chance.
Pick a group and draw one. Then ask what a “connection” would prove if it showed up in a group you cared about.
With 23 people and one kind of connection, the chance of at least one match is 50.7%. With 50 people it is 97.0%. With 20 people and five kinds of connection, it is 92.9%. In none of these did anyone arrange anything.
That doesn't mean connections never matter. It means a connection starts a question and doesn't answer one. The answer needs the base rate: how often does the same link show up among people or companies that aren't involved in what you're looking at?
You won't run all five steps on every headline. Two habits catch most problems. First, write the claim with a number, a date and a unit, and see if it survives. Second, ask what the opposite result would look like and whether anyone checked.
When you do run the full method, keep the two columns from step 5. The data goes in the left column. The speaker's conclusion goes in the right one, and it has to earn its place.
Sort it first: a checkable fact has a record, an opinion weighs values, and an untestable claim has no result that could count against it. Then pin it to a number, a date and a unit, find the primary source, check the unit and date, compare with a base rate, and keep the data apart from the speaker's conclusion.
The habit of testing a claim by collecting examples that fit it. Fitting examples turn up whether the claim is true or not, so they prove little. The Wason selection task shows it: most people check the case that can only agree and skip the one that could prove the claim wrong.
How often something happens by chance among comparable cases. Without it, a connection can look meaningful when it is normal. In a group of 23 people the chance that two share a birthday is 50.7%, and with 50 people it is 97.0%.
The data supports a narrower claim. IMF data shows the dollar's share of official foreign exchange reserves fell from 69.74% at the end of 2000 to 56.52% at the end of 2025, 13.22 points over 25 years. That is a long, uneven decline from a high level, not a collapse.