What should you ask a number before you believe it?
Pull what you learned into five questions, use them on a headline, and answer the questions from the start of the course again.
Free course. Your Maestro, an AI, .
This headline ran on a regional news site (it's invented, like the course's municipalities):
Ojo de Agua, the most dangerous municipality: thefts from businesses soar 600%
It has a huge percentage, a ranking and a strong word. And every number in it checks out. This page gathers what you did in the course into five questions and tests them on this headline. At the end, the questions from the first page come back, with different numbers.
Check yourself: The headline says thefts rose 600%. What do you need to know first to decide whether that's a lot?
Five questions for any number
Each idea in the course left you with a question. Together they fit on an index card, and they work just as well on a headline, a slide in a meeting or an answer from an AI.
Out of what total?
What was divided by what. A change is divided by the value from before, and the percentage of frauds that get caught is not the percentage of alerts that are fraud.
From how much to how much?
Ask for both levels. With them you can get the change in units, in points and in percent, and none of the three is enough on its own.
How many cases are behind it?
To compare a count, divide it by who it could happen to. If there are few cases, a single one moves the number a lot: look at several years or larger groups before you build a ranking.
Which typical number did they use?
An average with extreme values doesn't describe most people, and the average of several groups' percentages is not the percentage of the total.
How precise can it be?
The decimal places you show shouldn't be finer than what a single case moves. If the number promises more than it knows, round it and say so.
The headline, question by question
Open the steps one at a time and, before you open each one, try to answer it yourself. The numbers come from the municipios dataset, the same one from module 2.
Ojo de Agua, the most dangerous municipality?
Worked example · 6 steps
The headline says Ojo de Agua is the most dangerous municipality and that its thefts from businesses rose 600%. Using the course dataset, how much of that holds up? Ask it the five questions, one per step.
Work out the next step yourself, then open it to compare.
Try it on your own
Now write the query yourself, with no template to fix. Use two fixes from the small numbers page: pool the two years and set a minimum number of cases. A municipality with fewer than 20 thefts over the two years doesn't make the ranking; its count is reported separately.
If you don't know where to start, ask your Maestro for a hint, not the answer.
Return municipio, robosdosanios (robos2024 plus robos2025) and tasadosanios (the thefts from both years times 100,000, divided by 2 poblacion, rounded to one decimal place), only for municipalities with at least 20 thefts over the two years, sorted by tasadosanios from highest to
Your turn
Return municipio, robos_dos_anios (robos_2024 plus robos_2025) and tasa_dos_anios (the thefts from both years times 100,000, divided by 2 * poblacion, rounded to one decimal place), only for municipalities with at least 20 thefts over the two years, sorted by tasa_dos_anios from highest to lowest and then by municipio. You're done when you see 12 municipalities with Villa Morena on top (242.1), and neither Ojo de Agua nor El Encino appears.
Tables: municipios (municipio, poblacion, robos_2024, robos_2025)
municipios · 18 rows
| Column | Type |
|---|---|
| municipio Nombre del municipio (ficticio). | VARCHAR |
| poblacion Habitantes del municipio. | BIGINT |
| robos_2024 Robos a comercio denunciados en 2024. | BIGINT |
| robos_2025 Robos a comercio denunciados en 2025. | BIGINT |
First rows
| municipio | poblacion | robos_2024 | robos_2025 |
|---|---|---|---|
| Ciudad Ceiba | 1420000 | 2130 | 2272 |
| San Telmo | 612000 | 1040 | 980 |
| Puerto Viejo | 385000 | 480 | 521 |
Every run starts from fresh copies of the course's tables, so try anything. Tab indents; to leave the editor, press Escape, then Tab.
Ready to run your query.
With the minimum, Ojo de Agua drops out of the ranking and is counted separately, for what it is: 8 thefts in two years.
No number in the original headline was false. What failed was what it left out: the base of the 600%, how many cases were behind the rate, and what happened the year before. The five questions aren't for catching lies; they're for finding what's missing. That's why they also work on honest numbers: they tell you how far you can take them.
Paste a headline or a figure from your work and check it with your Maestro using the five questions.
The questions from the start, again
On the first page you answered six questions before the course explained anything. Here they are again, with different numbers and the same idea underneath. Answer them without looking back; at the end you can compare.
Check yourself: A store's sales fell 20% in January. How much do they have to rise in February to get back to where they were?
Check yourself: Staff turnover in one department went from 20% to 25% in a year. Which statement is correct?
Check yourself: In 2025 there were 980 thefts from businesses in Laguna Verde and 239 in Monte Claro. Where is a business more likely to be robbed?
Check yourself: At a company, the average salary is 30,000 pesos a month and the median is 14,000. What can you say?
Check yourself: One store has 8,000 customers and 25% buy again; another has 500 customers and 70% come back. What percentage of the two stores' customers combined buy again?
Check yourself: A quality inspection catches 99 out of 100 defective parts and, by mistake, flags 1 out of 100 good parts. Only 1 in every 1,000 parts comes out defective. A part was flagged: how likely is it to be defective?
What you can do now
Compare with what you answered on the first page. If one still gave you trouble, that's fine: each has its own page in the course, and your Maestro can take you back to it.
With what you did here, you can now calculate a change from the right base, tell points from percent, turn counts into rates, distrust first place in a ranking built on a few cases, choose between the average and the median, combine several groups' rates, count how many alerts are real, and give each number the decimal places it deserves. None of that took a statistics formula.
Tell your Maestro which of the six was hardest, and go over that idea with a case of your own.
Which number at work will you read differently after this course? Which of the five questions would you ask it first?
Reflect: Which number at work will you read differently after this course? Which of the five questions would you ask it first?
Only you and your Maestro read this. It's kept on your account and deleted with it.
If you want to keep going
More and more, the number you have to check wasn't calculated by a person but by an AI. If you want to hand data work to an AI agent and check what it did, the next step is Agentic AI for Data: Your Judgment, at Scale, a paid Maestro course. Its core question is the same one you kept asking here: what was counted, and what was it divided by? There you turn it into metric definitions (the population, the numerator, the denominator, the time window) that an agent can follow and you can review. It doesn't assume you know statistics either.