similarity label is misleading — score is a percentile rank, not cosine similarity

[FlorentPoinsaut]

Description

All chat messages use the word "similarity":

'chien': 72% similarity
'chien' has already been suggested (72% similarity).

However, the score is a percentile rank — it means the guess is closer to the target than 72% of all words in the vocabulary. It is not a direct measure of cosine similarity between the guess and the target.

Viewer impact

Players naturally read "72% similarity" as "my word is 72% similar to the target word". This creates two false expectations:

1. Two words at 72% and 68% are perceived as semantically close to each other.
2. Players don't understand why some words score surprisingly high or low.

Proposed fix

• Replace "similarity" with "score" or "proximité" in chat messages.

  "Score = percentage of the vocabulary your word beats in proximity to the target."

Files

• bot/bot.py — chat message strings in guess(), hint(), status()