The data is real.
The context was cut out.
This site checks, number by number, the campaign material on cost of living published at mercadodamentira.com.br ↗.
The conclusion, after auditing every section: they did not invent any data. They did something harder to spot — they picked a yardstick that starts and ends at exactly the right points to produce the desired conclusion. Here we put back what was cut, using the same official sources.
The two periods are not comparable.
The original site compares Jan 2019–Jul 2022 against Jan 2023–Jul 2026. It looks symmetrical: three and a half years on each side. But look at what falls inside each window — using a global index that no Brazilian president controls.
* 2026: August value (133.3), the highest since late 2022. All-time monthly peak of the series: 159.7 points in March 2022.
"Three and a half years of each government, side by side."
Calendar symmetry is presented as if it were symmetry of conditions. Same number of months, same source, same index — so whatever is left over must be the merit or the fault of the government.
One window captures the entire global climb. The other captures the descent.
The "Bolsonaro" window starts at 95 points and ends exactly at the all-time peak of the global food index: the pandemic, the logistics collapse and the war in Ukraine are all inside it.
The "Lula" window starts right after that peak and rides the normalisation of global supply chains. Much of the difference was already settled before any decision taken in Brasília.
The numbers are correct. What does not hold is the attribution of cause: the same rise-and-fall pattern shows up worldwide, including in countries whose governments never changed.
Sources: FAO — Food Price Index (monthly and annual series, 2014-2016=100); FAO release of March 2022 on the 159.7-point record; FAO release of September 2026.
Every claim, one by one.
Each one has its own page, with proof of what the original piece claims, the missing context and the primary sources. The data is true in every one of them — what changes is what was left out.
"Cheaper" does not mean cheap.
109 of the 112 items are still above their January 2019 price
“Dozens of products became cheaper under the current government.”
Read the full check →The 14.9% peak is nobody's starting point.
The 14.9% peak came in Q1 2021, at the height of the pandemic
“Unemployment fell from over 14% to 5.4% because of the current government.”
Read the full check →Food got more expensive worldwide — and then eased worldwide.
The FAO global food index hit its all-time record in Mar 2022
“Food prices rose 4x more under the previous government than the current one.”
Read the full check →A bill still being voted on is not a delivered policy.
The amendment has not yet been voted on the Senate floor
“Ending the 6x1 work shift is among the current government's delivered policies.”
Read the full check →The error is not in the spreadsheet. And the IBGE's problem is real too.
No evidence whatsoever of manipulation of the IPCA or the PNAD
“IBGE is falsifying employment and inflation data.”
Read the full check →- 3 Correct
- 2 Correct, no context
- 1 Misleading
Their comparison table, audited.
Three of the six rows in their table hold up — and we confirm them
Read the full check →- 5.4% is the lowest unemployment on record
- Informal work is falling
- Average income is at its highest on record
What the campaign gets right — and what nobody is telling you.
Unemployment at 5.4% is real, informality is falling and income is at a record
Read the full check →How we arrived at these numbers.
We used their own figures
The table was extracted from the original site's own “Item by item” section in September 2026. No variation was altered. We only removed 4 duplicate labels (the same IPCA item appeared under a short and a full name), leaving 112 products. The only new column is the cumulative one.
We archived the original
A copy of the site under review was saved to the Internet Archive on 9 September 2026. If they edit it or take the page down, the version we audited stays public and verifiable — the link is in the sources below.
We chained, we did not add
To find where a price stands today relative to Jan 2019, two consecutive periods multiply: (1+a)×(1+b)−1. Adding the two variations would give a wrong number.
We compared start with start
For unemployment and income we used the value of the quarter each government took office and the quarter it handed over — rather than the worst and best points of the series.
We tested against a global series
The FAO food index works as a control: if the same movement appears worldwide, it cannot be credited to or charged against a Brazilian government.
We say when they are right
Three of the six rows in the original comparison hold up. They are marked as correct, each with its legal source.
We date everything
Cyclical indicators change. Every block carries its verification date, and the page states the overall last update.
Primary sources
Download the dataset
The 112-product table, with both windows and the cumulative figure calculated. The same numbers as on the page, in an open format.
Found a number that does not add up?
This site started with a single campaign piece. If you have come across another claim with real data and a bent conclusion — from any side — send it over. The more specific the better: where you saw it, the link or screenshot, and which figure looked off.
Suggest a fact-checkOpens a public issue in the project repository. It requires a GitHub account — that is the price of keeping everything auditable, with no third-party form in between.
This calculation needs to travel.
A bent yardstick travels fast because it is simple. Context travels slowly because it needs explaining. If this check helped you understand the number, it will help someone else too.
The text comes pre-filled with the key figure. Write your own if you prefer — the link is the same.
João Pedro Gamin
I built this site on my own, with no ties to any party, campaign, news outlet or public body. The motivation was simple: the numbers in the original material were true, and yet the conclusion did not hold. That is a problem of method, and method can be demonstrated.
All the code and the dataset are open. If you find a mistake of mine, open an issue in the repository — the correction is published as plainly as the original claim.