Methodology

1. Where the data comes from

The ranking isn’t set by an editorial team: it comes from the votes of the people who play. In each duel, two behaviors are shown and the player picks the one they find more problematic. Those choices, repeated over and over, are what move a behavior up or down the ranking.

There are currently 183 behaviors in the running.

No account is needed to vote. Two optional pieces of information are asked before playing — age bracket and gender — only to make the comparisons described below possible. They aren’t tied to any identity.

2. The Elo score

We use the Elo system, originally designed to rank chess players. What makes it useful here: it doesn’t just count wins, it weighs each duel by how strong the opponent is. Beating a behavior already judged very problematic is worth more than beating a harmless one.

Every behavior starts at 1,000 points. After a duel, the expected probability of winning is calculated like this:

E = 1 / (1 + 10^((opponent score − score) / 400))

The score is then adjusted by K × (result − E). The K factor shrinks as a behavior collects votes: 40 under 30 duels, 32 up to 100, then 24 beyond that. In other words, a new behavior moves fast, while a well-established one only moves with many consistent votes. A floor of 100 points keeps a score from sinking indefinitely.

3. Comparisons by group

A separate Elo score is kept for each group players declare: by gender, and by age bracket (16–18, 19–22, 23–26, 27 and over). That’s what lets us measure disagreements — the same behavior can be judged harshly by one group and much less so by another.

These sub-rankings necessarily rest on fewer votes than the overall ranking. The narrower the group, the bigger a gap has to be to mean anything.

4. What this ranking is not

It’s worth being clear about what these numbers can and can’t tell you.

• It’s not a representative survey. The sample is self-selected: these are people who chose to play, not a designed panel. The results describe our community, not the French population.
• It’s not a diagnosis. A high score reflects a consensus that a behavior is problematic, not a verdict on a person or a real relationship.
• The game’s tone shapes the answers. Some prompts are deliberately lighthearted or provocative. They’re voted on in a playful setting, which affects the choices people make.
• What players declare isn’t verified. Age and gender are self-reported.

5. Moderating the prompts

The behaviors put to the vote are written and reviewed in advance, not generated automatically. Prompts in the “Love & Sex” category are not shown to players who select the 16–18 age bracket.

6. See the data for yourself

The full ranking can be browsed and filtered by category, gender and age bracket on the red flag rankings page. The self-assessment tools, which rely on official scales rather than our votes, are described in our sources.