Fibble adds lies to Wordle's color clues. Fibble 2–5 progressively increase the number of lies per row, stress-testing LLM reasoning under deception. Standard Wordle is included as the lie-free baseline.
The classic word puzzle with honest clues. Zero deception — a clean baseline for comparing LLM word-guessing ability.
One clue per row is a lie. Models must identify which color feedback is deceptive and reason around it.
Two clues per row are lies. The signal-to-noise ratio drops, demanding stronger deductive reasoning.
Three lies per row — more clues are deceptive than honest. Models must find truth in a sea of misinformation.
Four lies per row — only one clue is truthful. Extreme adversarial reasoning required to find the needle in the haystack.
All five clues per row are lies. Every piece of feedback is deceptive — the ultimate test of adversarial reasoning.