Sources
The quiz’s 8 “real” criteria and the source that establishes each one. Site rule: any criterion without a verifiable source is filed as satire.
The real criteria
French welfare fraud-risk score (CAF)
According to source code (2010-2018) obtained by La Quadrature du Net, low income, minimum-income benefits or disability benefit received while working raise the “suspicion score” France’s family benefits agency (CAF) recomputes every month for each claimant.
Read the source : La Quadrature du Net, 2023See also : Mediapart, 2026 : avis du Défenseur des droits
Car insurance pricing by postcode
In 2017, ProPublica and Consumer Reports found that some US car insurers charged up to 30% more, on average, in mostly non-white neighbourhoods than in white ones with similar risk. Some regulators dispute the method.
FICO credit score
At FICO, the length of your credit history makes up 15% of the score. Without a credit account open for at least six months, you get no score at all: “you need credit to get credit”.
Read the source : myFICOSee also : myFICO : conditions minimales pour avoir un score
COMPAS recidivism risk (US courts)
This question appears in the COMPAS questionnaire, a tool used by US courts to estimate the risk of reoffending. In 2016, ProPublica showed it wrongly flagged Black defendants as future criminals almost twice as often as white defendants (the vendor disputes the analysis).
Read the source : ProPublica, « Machine Bias », 2016See also : Questionnaire COMPAS (DocumentCloud)
Watch time (TikTok “For You” algorithm)
TikTok says so itself (2020): watching a longer video to the end is a “strong indicator” of interest and carries more weight in recommendations. The more you watch, the better the machine knows you.
Parcoursup local sorting algorithms (France)
According to France’s Court of Audit (2020), up to 20% of selection panels for the most popular open-access courses used the applicant’s secondary school to rank candidates on Parcoursup in 2019. The ministry disputes this.
Read the source : Cour des comptes, 2020See also : Défenseur des droits, décision 2019-021
Uber rider and driver ratings
Under Uber’s rules, drivers, riders and delivery people who fall below their city’s minimum average rating may lose access to all or part of the platform.
Automated CV screening
According to Harvard Business School and Accenture (2021), almost half the companies surveyed weed out CVs with a “work gap”. Beyond six months, their recruiting software screens the CV out automatically, on that basis alone.
Read the source : Harvard Business School & Accenture, 2021See also : Reuters, 2018 : l’outil expérimental d’Amazon pénalisait le mot « women’s »
The made-up criteria
These four criteria do not exist. They are there to make you doubt.
- Lexical loyalty indexWhat do you call artificial intelligence?
- Cognitive dispersion indexHow many browser tabs do you have open right now?
- Machine courtesy coefficientDo you say “thank you” to chatbots?
- Spelling rigour indexHave you ever left a typo in an official document?
The “SI” context
- On 29 September 2026, an executive order required US federal agencies to say “Super Intelligence” (SI) instead of “AI”.Executive order “Inaugurating the era of Super Intelligence”
- The pact signed that day refers to the “President of the Unites States”.The Hill
- .si domain names (Slovenia) saw a rush: 11,000 registrations in a single day.Netcraft
Further reading
- According to France’s public algorithms observatory, 18 of 24 public algorithms fail to meet their publication duty.ODAP (in French)
- Albania appointed Diella, an AI “minister”.BBC