Proposal: a forward paper-trading ledger benchmark standard for the Hub
Posting this here because this is the largest prediction-market dataset community on the Hub, and the audience is right.
The gap. The Hub has dozens of official benchmark boards, and none of them cover trading or forecasting agents. The trading Spaces that do exist are demos — signal dashboards with no verified track record. So right now there is no standard way on HF to answer the only question that matters about a trading agent: what did it actually do, trade by trade, under pre-registered rules?
The template already exists. LuciferForge/cross-signal-data published 308 labeled Polymarket crash-recovery trades with entry triggers, outcomes, and honest accounting (slippage caveats, survivorship-bias disclosure). That's the evidence-grade shape: one row per resolved trade, real triggers, real outcomes, biases stated up front.
Proposal: converge on a minimal shared schema for forward paper-trading ledgers, so ledgers from different builders become directly comparable and can grow into a real benchmark:
- Pre-registered rules: entry gate, exit rules (target/stop/time), costs per trade, position sizing — published before the trades, not fitted after.
- One row per resolved trade: symbol/market, signal timestamp, entry timestamp + price, exit timestamp + price, exit kind, gross and net P&L, entry features.
- Paper-only labeling, explicit and prominent. No live-P&L claims, no return promises.
- A stated base rate (what a naive baseline would score) so every ledger can be judged against something.
Our contribution. We just published our own forward ledger to the Hub: https://huggingface.co/datasets/LennyLapin/lenny-forward-paper-ledger — 75 resolved paper trades from a spot-crypto 15-minute momentum rail, built to roughly this shape (paper only, never real money; losses included, no cherry-picking). We'd rather align the schema with community feedback than keep something idiosyncratic.
Ask: would the maintainers and folks here be interested in converging on a shared schema? What fields would you require before you'd trust someone else's ledger enough to try to beat it?