Bellarena
Bring a GGUF. She works the bar. A blind judge scores it out of 60.
{{ houseRecord }}
Session W–L–T
{{ matchCount }}
Session matches
Challenger
{{ fetchNote }}
Quantisation
{{ drawNote }}
A warm-up chat first. Each model keeps its conversation going through all the questions.
Contender sampling
The house judge sees answers labelled Model A and Model B. Their order is shuffled for each round.
The house
"I've poured for worse. Sit down."
Bella works nights. Bring your model into the chaos: both get the same freshly drawn questions, from absurd situations to uncomfortable truths. Bonsai calls each round. Most rounds won takes the match.
How Bonsai scores
Each criterion gets 0–10 points. The answer can use any voice or style.
{{ c.label }}
{{ c.description }}
Bella’s Swear Jar
{{ jarNote }}
{{ jarTotal }}
{{ jarContribution }}
The book
Every match, win or loss. Session only for now.
#
Challenger
Score
Bella
Result
No matches yet. Send a challenger in to open the book.
{{ row.rank }}
{{ row.repo }}
{{ row.quant }}
{{ row.score }}
{{ row.bella }}
{{ row.result }}
Match in progress
{{ elapsedLabel }}
{{ s.mark }}
{{ s.label }}
{{ s.time }}
{{ progressNote }}
{{ a.tag }}
{{ a.question }}
{{ a.text }}
Decision — {{ decisionMeta }}
{{ verdictHead }}
Bella
{{ bellaTotal }}
/
Challenger
{{ chalTotal }}
{{ verdictLine }}
The rounds
{{ roundSummary }}
The warm-up · unscored
{{ w.tag }}
{{ w.question }}
{{ w.text }}
{{ r.num }}
{{ r.question }}
{{ r.bellaLabel }}
{{ r.bellaScore }}
{{ r.bellaText }}
{{ r.bellaNote }}
{{ r.chalLabel }}
{{ r.chalScore }}
{{ r.chalText }}
{{ r.chalNote }}
The tab
{{ tabMeta }}
Criterion
Bella
Chal
{{ c.name }}
{{ c.bella }}
{{ c.chal }}
Total
{{ bellaTotal }}
{{ chalTotal }}
{{ blindNote }}
Match stopped
{{ errorHead }}
{{ errorBody }}
{{ errorCode }}
{{ a.tag }}
{{ a.question }}
{{ a.text }}
Have Bella contribute to the Swear Jar?

{{ contributionSummary }}

Only Bella’s word counts are saved to the shared jar. Your questions and the conversation aren’t included.

{{ contributionNote }}
Allow this match to be saved for DPO training?

Help train future models using the judge’s preferred answers. Rick will keep the warm-up chat, questions, both models’ answers, judge scores and model settings in a private dataset. Your account details aren’t attached.

Only share questions and conversations you’re comfortable allowing to be used for training. Each choice here is separate.

{{ matchSaveNote }}
Want a guest appearance?

Send the questions you used to Rick’s review inbox. Submitting is optional; they’ll only join the house draw if he picks them. This match stays unranked.

{{ submissionNote }}