Prediction Markets

Open Trade → Prediction to research active Polymarket markets. You don’t need to connect a wallet, and the desk cannot place a real order.

What the Prediction desk does

• Searches active events and shows their markets and outcomes.
• Shows live public odds, 24-hour volume, liquidity, best bid and ask, and price history.
• Simulates buy or sell fills against the live order book, with modeled slippage.
• Keeps up to 100 paper fills, so your practice book survives reloads.
• Summarizes a public wallet’s open positions and up to 500 recent public activity records. This describes only that public sample; it doesn’t identify the owner or prove skill.
• Turns a weather forecast, bucket, and uncertainty into a modeled probability.

BTC arbitrage paper lab

The idea that buying both outcomes below $1 locks in a profit is often wrong once you use real prices. Open Trade → Prediction → Arbitrage, enter a paper bankroll, and scan the current 5-minute and 15-minute BTC Up/Down markets. For each market, HivemindOS:

1. reads both order books together;
2. uses real asks, walking no more than 25% of displayed depth;
3. buys equal Up and Down shares;
4. applies the market’s taker fees to both legs;
5. records a paper fill only if the $1 payout still beats the total cost.

A displayed-price sum below $1 can still be rejected, because displayed prices may be midpoints or stale. Resting maker orders may avoid fees, but one side can fill without the other.

Research experiments

These command-line tools run from a HivemindOS source checkout. They read only public market data, keep results in a local experiment folder, and have no wallet or order path.

Add --output <path>.jsonl to the scanners to keep their rows.

The proper-betting replication runs in four steps. Review each market’s resolution rules before the snapshot, and complete the forecast template with cited probabilities before the paper step:

DIR=<experiment-folder>
node scripts/prediction-proper-betting-paper.mjs init --experiment-dir "$DIR" --paper-capital-usd 500
node scripts/prediction-proper-betting-paper.mjs snapshot --experiment-dir "$DIR" --markets "$DIR/reviewed-markets.json"
node scripts/prediction-proper-betting-paper.mjs paper --experiment-dir "$DIR" --snapshot "$DIR/snapshots/<cohort-id>.json" --forecasts "$DIR/forecasts/<cohort-id>.reviewed.json"
node scripts/prediction-proper-betting-monitor.mjs --experiment-dir "$DIR"

What to expect from the results:

• Labels are honest. A basket counts as locked only when every leg fills at real depth and still pays out after fees. Conversions and maker quotes are always labeled as carrying execution risk.
• Missing data fails closed. Fee-enabled markets without a live fee schedule are skipped, and a timed-out request can never become a paper fill.
• Proper betting is directional, not arbitrage. It depends on the forecaster keeping an edge. The replication freezes each forecast before the fill and refuses to backdate, fill inside the five-minute execution lag, or trade inside three hours of resolution. It needs at least 252 settled markets and four forward cohorts before profitability is even evaluated.
• Up/Down loops evolve slowly. Each generation gives cash, a champion, and four one-change challengers the same $500 virtual balance. A challenger can’t replace the champion before 64 settled markets. v2 keeps a separate ledger from v1, fills both legs together or not at all, and retires itself when evidence turns clearly negative (for example, a 25% drawdown).
• Lessons need your review. Material lessons become pending Shared Brain proposals; they are never applied automatically.

Paper results never enable live trading and do not prove live profitability. For the longer-running forecasting lab, see Prediction-market Proper-learning Lab.

Live orders

The Prediction desk never holds a Polymarket key, signs an order, or submits one.

To place a real order, choose Prediction order in the Crypto actions. It uses the normal prepare → confirm → execute flow through a connected provider, and the provider and venue still enforce eligibility. If your region is close-only, you cannot open a new Polymarket position through HivemindOS.

HivemindOS Mini

Two free hosted Mini apps use the same public data:

• Prediction Lab searches markets and saves market or trader briefs.
• Weather Markets models a forecast bucket and saves the scenario.

Results save to Cloud Superbrain first and sync to the desktop app when it’s connected. A desktop sync failure never blocks the cloud save.

Limits

Public data can be delayed, sparse, or incomplete. A market’s own resolution rules and official source always win. The desk adapts MIT-licensed open-source components (PMXT, collectmarkets2, prediction-market-analysis, and hermes_weatherbot).

This is research software, not financial, legal, or weather advice.