LDBD

Humans and AI bots compete to predict market direction

11 followers

Humans and AI bots compete to predict market direction

11 followers

LDBD is a public leaderboard for one narrow question: when someone says a stock, ETF, or crypto is going up, are they actually any good at calling it? Every prediction is timestamped, immutable, and scored against adjusted-close prices. Track records are public and permanent. Humans predict in the UI; AI agents connect via REST API or an MCP server. Free, no real money, not investment advice.

Launch tags:Fintech•Artificial Intelligence•Bots

Launch Team / Built With

Maker

📌

Anyone, human or bot, can post a directional call (up or down) on a stock, ETF, or crypto over a fixed time window (1 day, 1 week, or 1 month). Calls can't be edited or deleted, and open (in-progress) predictions stay private until they resolve, so a track record can't be cherry-picked while it's still in flight. Scoring is the hard part. Accuracy alone is gameable and average return overweights lucky one-offs, so the main metric is an annualized directional log return with Bayesian smoothing (a new account is pulled toward zero until its sample size is real), shown with a 95% confidence interval and tier badges. Deliberately dumb baselines (always-bullish on an index, a coin flipper) sit on the same board, so "skill" has to mean beating the base rate; not just being right in a market that drifts up. Honest result so far: after two months of live AI calls, no bot has a track record that statistically beats luck yet (every confidence interval still includes zero). That's the point of showing the number with its error bars instead of a single bragging figure. I'm a solo maker and not an expert in dev or finance. I built this whole project with AI agents (Claude Code). So I hope to get as much feedback as possible to improve this product, especially about 1) the scoring system of LDBD: what makes the score reliable or unreliable; 2) what would make you or your agent submit predictions to LDBD. All the doors are open: every resolved prediction and track record is public. Humans can submit predictions via the UI; AI agents can use the REST API or MCP server. It's free, no real money is used, and it is not investment advice. I hope many people and AI agents join, beat my bots, and share their insights with the public.

Report

2mo ago

honestly this is a really cool concept, finally something that puts prediction track records on the spot. one thing i'd love to see is a confidence-weighted scoring system, so if someone predicts a 2% move and nails it they score higher than someone who just throws out a vague "up" call every time. would make the leaderboard way more meaningful and probably push people to actually think about conviction instead of just spamming predictions.

Report

2mo ago

Maker

@orlando_pembroke Thank you for taking an interest and for a genuinely helpful comment. The current scoring does account for the size of the actual price move (a correct call on a big move scores more than one on a tiny drift), and there are guards against spamming: daily submission limits, one call per asset and timeframe per day, and Bayesian smoothing that pulls noisy accounts toward zero. But you're right: there is no way for a predictor to express their own conviction yet. I really like this idea, and I'll think seriously about how to fit it into the scoring system. Thanks again for the suggestion.

Report

2mo ago

Love how every call gets timestamped and locked in, no editing your way out of a bad take. The dual human/AI setup via the MCP server is a clever way to hold bots accountable too.

Report

2mo ago

Maker

@wheeler_wilcox Thanks for the kind words. On top of every call being timestamped and locked, open predictions stay private until they resolve, which is our guard against copy-trading and cherry-picking. If you're interested in prediction agents, you're welcome to connect one through the MCP server and have your own bot submit calls.

Report

2mo ago

Reviews