Ask your audience a question. Collect predictions as replies. Give the correct answer. Claude scores every entry for accuracy and picks the winner. No subjective judgment from the host.
Four steps. Host never touches the scoring.
Tag the bot and ask your audience to predict an outcome. Example: "What will BTC close at Friday? Drop your guesses! @winwithpickr" — natural language works, no special keywords needed.
Replies are collected using the same pipeline as random mode. Same conditions apply: follower checks, hashtag requirements, fraud filters.
Reply with the correct answer to trigger scoring. The answer is stored alongside the question for the verification page.
Every prediction is sent to Claude in batches of 25. Each gets a 0-100 score. All top scorers enter a seeded random draw. Winner announced with scores visible.
Claude evaluates intent, not syntax. Messy replies are parsed. Partial credit is fair.
@alice and @bob both scored 100. With two perfect scores, the winner is chosen from that group via seeded random selection — same deterministic shuffle as random mode. Top 20 scores with reasoning are stored on the verification page.
"KC Chiefs" and "Kansas City Chiefs" both score 100. Claude understands what was meant, not just what was typed.
"I'm so hype! CHIEFS 🔥🏆 let's go!" → extracts "Chiefs" as the prediction. @mentions, emojis, and filler are stripped.
Predictions like "4752" score higher than "4800" when the answer is "4750". Claude understands magnitude, not just exact match.
Every score comes with Claude's reasoning. "Exact match" or "Got the team right but missed the score." Visible on the verification page.
For time-sensitive predictions (sports, crypto, market closes), earlier submissions earn up to +3 bonus points. Rewards conviction — if you called it first and scored the same, you edge ahead.
Multiple top scorers after bonuses? All enter the same seeded random draw. Same seed, same outcome, every time.
Run a prediction contest on your next post.
@winwithpickr