How the board works

a voice is one post, comment, or write-up we tied to a model. more voices means more people talking — not automatically a better model. a quiet 9.0 is a different signal from a loud 7.5.

vibe is 0 to 10. three quarters of it is what people say, bayesian-smoothed so thirty voices cannot outrank three thousand on raw average alone. the last quarter is how many people are talking, log-scaled so volume counts without drowning sentiment.

happy and sad on a model page are the raw mood scores. vibe is the blend we rank on.

weekly happy/sad/neutral counts, smoothed, then a badge:

• holding up — recent vibe is steady
• honeymoon fading — early praise is cooling
• aging well — the crowd is warmer than at the start
• quietly degrading — recent vibe and chatter are both sliding
• too quiet — not enough weekly voices to call a trend

the model page and /api/models read the same board snapshot for this. if they disagree, that is a bug.

one ordered list of vibe scores. we then number it a few ways:

• assistants — text + multimodal, the homepage "everything" view. this is the headline rank on a chat model page. quote it with its board, like "#12 of N assistants".
• overall — every reviewed model, including image, video, audio, and code. a compare table also shows this, labeled. do not mix it with assistants.
• category — 1..N inside pictures, video, voice, or code. output specialists lead with this.
• jobs — writing, coding, agents, research. same vibe formula, only the conversations about that job. a model needs enough job-tagged voices to appear.

"of N" is the size of that board, from the same snapshot. "#12 of N assistants" and "#40 of M overall" for the same model are different boards, not a math error.