The short answer most beginners do not want to hear: Polymarket does not have a built-in paper trading mode. There is no demo account, no virtual balance, no "sandbox" environment. Every price you see is real and every order book is the real one.
What people are usually asking for when they search "Polymarket paper trading" is some way to test a strategy without losing money in the experiment phase. There are five real ways to do that — none of them is a polished "paper trading mode," but each addresses a specific piece of what you actually need to test.
The Five Real Options
| Method | Cost | Fidelity to live | Best for |
|---|---|---|---|
| Manifold Markets (play-money community) | Free | Low | Forecasting practice, building reputation |
| Historical price backtesting | Time + data | Medium-high | Strategy research, parameter tuning |
| Shadow bot mode (log-only) | Time | Very high | Validating a bot before going live |
| Minimum-size live trades ($1-$5) | Real $ but tiny | Highest | Learning UI, testing fills |
| Prediction logging in a spreadsheet | Free | Medium | Calibration measurement |
The right tool depends on what you are actually trying to learn.
Option 1: Manifold Markets for Forecasting Practice
Manifold Markets is a play-money prediction market with a real forecasting community. You cannot withdraw real money, but the trading mechanics, calibration tracking, and Brier-score leaderboards are real.
What it is good for: - Building a public forecasting track record before you risk capital on Polymarket. - Testing whether your reasoning on a specific market type produces well-calibrated forecasts. - Learning the "feel" of moving prices, taking other people's liquidity, and watching your position get pulled by events.
What it is not good for: - Anything where execution cost matters — Manifold has no real fees and no real slippage to test against. - Anything where you need Polymarket-specific data. - Real dollar P&L attribution.
If you want to test whether you have edge on NBA games, post forecasts on a dozen Manifold NBA markets, hold to resolution, and look at your accuracy and Brier score over a month. If those numbers are good, your model has something. If they are random, scale that knowledge before testing it with money. We covered Manifold in more depth in our Kalshi alternatives post.
Option 2: Historical Price Backtesting
This is what serious quants do. Polymarket's CLOB API exposes a historical prices endpoint (/prices-history?market={token_id}) that returns price series for past markets. You can replay your strategy against those series, see what trades it would have taken, and compute P&L.
The basic loop: 1. Pull historical price series for the markets you would have traded. 2. Walk the series forward in time, simulating what your strategy would have done at each point. 3. Apply realistic transaction costs (fees + estimated slippage based on book depth). 4. Compare predicted vs realized at settlement to score the strategy.
Why this is the highest-leverage option for serious strategy work: - You can test years of trading in minutes. - You get statistically meaningful sample sizes that live testing would take months to produce. - You can sweep parameters (edge thresholds, position sizing, exit logic) to find robust settings.
Limits: - Historical orderbook depth is not available — only mid-prices. You have to estimate slippage from price-change patterns, which is approximate. - Backtests assume you would have actually executed at the historical mid; in practice you cross the spread. - The famous gap between backtest results and live results is real. Treat backtest P&L as the optimistic ceiling, not the expected outcome.
Our own bots are all backtested against this data before going live. Scripts like backtest_moneyline_wp.py, backtest_cs2_combined.py, and backtest_soccer_wp.py all pull from the same /prices-history endpoint with fidelity=5 (5-second bars) and replay strategy logic against the historical series.
For more on the structural issues with backtest-vs-live divergence, see why backtest results overstate live performance.
Option 3: Shadow Bot Mode
If you have a working bot and want to know whether its decisions are sound before the decisions cost you money, run it in shadow mode: the bot processes live price feeds, applies all its signal logic, and logs what it would have done — without actually placing orders.
Our bots all support a --mode shadow flag for exactly this. The shadow trade log is identical in shape to the live trade log — sport, fair_prob, entry_price, edge, size, predicted outcome — except no on-chain transaction was submitted. After 1-2 weeks of shadow logs you can answer questions like:
- Would this strategy have produced positive expected value at current market conditions?
- How many signals fire per day? Per sport?
- What edge thresholds is the bot actually crossing in the wild?
- How does the live signal distribution compare to the backtest distribution?
Shadow mode is the highest-fidelity test you can run short of live trading. The only thing it misses is execution — you do not learn whether your fills would have happened at the prices you assumed. For that, you need option 4.
Shadow mode is also how we validate any meaningful config change before deploying it. Change a sport's edge threshold, run shadow for a few days, compare shadow trades to what live would have done. If the new config flags fewer false positives without missing real edge, it ships.
Option 4: Minimum-Size Live Trades
The cheapest way to test execution mechanics is to actually trade — at the smallest size Polymarket allows.
Polymarket's minimum order size is around $1 of notional (a few shares at typical prices). At that size, even a complete loss costs you a dollar. The information you get is real: did your fill happen at the price the orderbook showed? How long did it take to confirm? Did the WebSocket fill notification arrive cleanly? Did the settlement and redemption flow work end-to-end?
This is the only way to test the parts of the trading loop that no simulation captures — your network latency to Polymarket's servers, your wallet's signing speed, the actual API rate limits you experience, and the UX flow at settlement time.
For someone deploying a bot for the first time, we recommend at least 24 hours of minimum-size live trades before raising position size. Find the bugs while losing dollars, not while losing real money.
Option 5: Prediction Logging (the Cheap Calibration Test)
Before you trust any model, you should know whether it is calibrated — when it says 70% probability, do those events occur 70% of the time? You can measure this without trading at all.
Take a spreadsheet. Every day before games begin, write down your model's predicted probability for each game. Don't trade — just log. After two weeks, you have ~100 predictions with known outcomes. Bucket them by predicted probability (0.50-0.55, 0.55-0.60, etc.) and check the actual win rate in each bucket.
If your 0.60-0.65 predictions hit 62% of the time, your model is well-calibrated in that range. If they hit 47%, the model is overconfident at that level and you should not size against its raw output without recalibration.
This is the cheapest, lowest-friction test of all. It does not test execution. It does not test sizing. But it answers the most important question — can the model be trusted at all — without spending a dollar.
For more on why calibration matters more than accuracy, see our calibration beats accuracy post.
Which One You Should Use
The right option depends on what you do not know yet:
- Don't know if your forecasting reasoning is any good? → Manifold (option 1).
- Don't know if your historical strategy idea has edge? → Historical backtest (option 2).
- Have a bot that works in backtest and want to validate live decisions before risking capital? → Shadow mode (option 3).
- Confident in the strategy but worried about execution mechanics? → Minimum-size live trades (option 4).
- Don't yet know if your model is calibrated? → Spreadsheet logging (option 5).
For someone going through the full strategy-to-production pipeline, you will end up using most of these in sequence. Build the model and check calibration (option 5). Backtest it on historical data (option 2). Run the bot in shadow mode against live prices (option 3). Validate execution with tiny live trades (option 4). Then scale.
What "Paper Trading Polymarket" Is Not
A few things people search for that do not really exist in 2026:
- A Polymarket demo account with a virtual balance. Polymarket has not built this. Probably will not.
- A simulator that exactly replays past market conditions. Polymarket's data export is mid-price, not full order-book depth — so any simulator is approximating slippage rather than replaying it.
- A third-party tool that simulates Polymarket trading. Some tools claim this; what they actually do is wrap historical mid-price data and apply assumed transaction costs. That is just option 2 with a UI layer.
If you find yourself wanting a polished paper trading app for Polymarket, the underlying need is usually one of the five real options above.
A Practical Onboarding Sequence
If you are coming in cold and want to learn Polymarket trading without losing money in the learning phase, do this in order:
- Week 1: Predict outcomes for ~20 markets per day on Manifold or in a spreadsheet. Track accuracy and calibration. (Free.)
- Week 2: If your calibration is reasonable, do option 4 — deposit $50 on Polymarket, place 10-20 minimum-size live trades to learn the UI and the execution flow. (Cost: typically $2-5 in fees plus spread on liquid markets, assuming roughly even outcomes.)
- Week 3+: Build a backtest of your strategy against historical price data (option 2) to confirm the historical edge before scaling size. (Cost: time.)
- Production: Run shadow mode (option 3) on any bot for a week before letting it place real orders.
This sequence takes about a month and costs under $20. You end with a calibrated model, a validated strategy, working execution code, and the confidence to size up.
Related deeper reads: - How to Bet on Polymarket: Step-by-Step for First-Time Users — for the live-trade execution path. - Kalshi Alternatives in 2026 — Manifold context plus other prediction-market venues. - Calibration Beats Accuracy — why prediction logging matters before sizing. - Backtesting Sports Betting Strategy — the backtest-vs-live gap in more detail. - Polymarket Fees Explained — costs to factor into any test.