Donate a chat.
Round 02’s judging needs real chats from real readers. Drop a JSONL export and a few notes about it. We’ll anonymize before anything enters the dataset, and you decide whether to credit yourself or stay anonymous.
We accept what most chat tools export — JSONL (one message per line) or a single bulk .json file with a messages array (RoleCall full export). Per-message field shapes: role+content (ChatGPT, Claude, generic) or name+is_user+mes (SillyTavern, RoleCall). Metadata headers are skipped automatically. Min 4 turns, max 500.
What was the scene? Genre, premise, what made it interesting. Two or three sentences is plenty — this helps the judges contextualize the chat.
List any model(s) that contributed to the chat — comma-separated. Don’t worry about exact spelling or the version number; the moderators normalize these on review. Leave blank if you don’t remember.
If you used a named preset (HawThorne, JLLM, RoleCall default, etc.) drop the name here. If you wrote your own system prompt, just say so — we don’t need the prompt itself.
Before any chat enters the dataset we strip personal info: emails, URLs, phone numbers, and any persona names you used as your character (replaced with USER). The judges and the public never see those. You can also remove individual turns later by emailing us.