Polybacktest API for Polymarket Order Book Backtesting and Data

Polybacktest Overview

Polybacktest Key features

• Sub-second order book depth

  The service saves the full stack of bids and asks for up and down tokens at sub-second intervals. Traders use this data to calculate exact fill prices.

• Historical data preservation

  The database stores complete order book snapshots long after markets resolve. Traders replay past market conditions to test their strategies.

• Aligned reference prices

  The platform includes Bitcoin and Ethereum prices that match Polymarket settlement rules. Traders analyze how external price movements affect prediction token prices.

• Python SDK support

  Developers use a ready-made Python library to fetch market metadata. This tool helps programmers focus on strategy logic.

Polybacktest Demo video

Polybacktest Screenshots

Polybacktest User Reviews

• 

  @monokern May 19, 2026

  The sub-second order book depth lets me simulate how my size moves the market on BTC contracts. I stopped guessing on slippage and burning capital. Now I iterate strategies in hours. The historical coverage on resolved markets provides the statistical validation I need. I recommend this to anyone building an edge.

  View on X ↗
• 

  @PolyDekos May 18, 2026

  The API response times are fast. Having full bid and ask stacks aligned with reference prices makes backtesting feel realistic. I caught several strategies that looked profitable on charts but failed on actual liquidity. This saved me from deploying bots that would have lost money.

  View on X ↗
• 

  @polybacktest Mar 11, 2026

  Polybacktest provides sub-second historical data. It turned my vague ideas into numbers with real fill modeling. Anyone running bots on these up or down markets needs this level of granularity. It removes the guesswork around stale prices and liquidity impact.

  View on X ↗

Polybacktest Updates

1. 

   PolyBackTest Jun 4, 2026 · 13:29
   Big crypto moves are easier to fade when almost nobody traded them.
   We checked 7,017 next-open fade events across BTC, ETH, and SOL 5-minute markets.
   The setup:
   Watch a 5-minute market settle after a real underlying move.
   Require the prior move to be at least 18 basis points.
   Require prior volume to be less than 2% of final liquidity.
   At the next market open, buy the side opposite the prior winner only if the visible top-of-book ask is 57c or less.
   That is the Quiet Move Fade.
   The backtest:
   1,147 events
   57.1% won
   average ask was 52.7c
   average return was 4.4c per share
   The baseline matters.
   Across all same-priced next-open fades, we had 5,980 events. They won 52.3% and made only 0.1c per share.
   So the edge was not simply "fade every move." The filter was the prior market's turnover.
   When the prior market moved 18 basis points or more but barely traded, the price move looked less like committed positioning and more like a thin mark. The next market often opened as if the move deserved continuation, but the better trade was to buy the other side while it was still near 50c.
   Build the Quiet Move Fade on fresh crypto markets at @polybacktest.
   • 1 replies
   • 1 reposts
   • 8 likes
   • 441 views
2. 

   PolyBackTest Jun 3, 2026 · 13:29
   The biggest edge in BTC 5-minute markets wasn't picking winners.
   It was buying losers that everyone had oversold.
   We tested 1,898 resolved BTC markets.
   The setup:
   • Ignore the favorite
   • Find the underdog
   • Buy only if it's priced 35c-49c at open
   • Require 100+ shares of bid liquidity across the top 3 bid levels
   Results:
   • 1,047 trades
   • 48.6% win rate
   • Avg entry: 46.7c
   • +1.9c per share
   Here's the part most traders miss:
   The strategy won LESS than half the time.
   And it still made money.
   Why?
   Because a 46.7c token doesn't need to win 50% of the time to be profitable.
   The payoff is asymmetric.
   The liquidity filter was the difference:
   35-49c underdogs with bid support:
   → +1.9c/share
   35-49c underdogs without bid support:
   → -1.6c/share
   A cheap token with no bids is often cheap for a reason.
   A cheap token with real buyers sitting underneath it is different.
   Someone is still willing to own it.
   Even more interesting:
   The edge disappeared above 50c.
   This setup usually happens when markets move a lot in the previous 5 min market and traders are pricing in a reversal, pushing the up/down prices to 55/45 before the market has even started. Prices diverge even more on market inception.
   Break any strategy down by side, time, and volatility in seconds with polybacktest
   • 1 replies
   • 4 reposts
   • 34 likes
   • 3.4K views
3. 

   PolyBackTest Jun 2, 2026 · 16:07
   Most traders do the same thing after BTC, ETH, and SOL all close UP:
   They click UP again.
   Our backtest says that was the wrong trade.
   We tested 2,326 cases where all 3 coins resolved the same direction in a 5-minute market and moved at least 0.10%.
   The edge wasn't blindly fading the move.
   The edge was:
   → BTC, ETH, SOL all resolve UP → buy DOWN
   → BTC, ETH, SOL all resolve DOWN → buy UP
   BUT only when your side was priced 50-58c.
   Results (1,569 entries):
   • 56.7% win rate
   • Avg entry: 53.1c
   • +3.6c/share
   What's interesting:
   Below 50c? No edge.
   60c+? Negative edge.
   The profits came from fading crowded sentiment when the market was still pricing the opposite side near a coin flip.
   In the full sample, the next market reversed in at least 2 of the 3 coins 53.9% of the time.
   The lesson:
   Don't chase the streak.
   When all 3 coins finish the same way, check the next market.
   The best trades weren't momentum.
   They were crowded momentum.
   You can backtest your own strategies with polybacktest. It has subsecond historical data going back 2 months
   • 5 likes
   • 767 views
4. 

   PolyBackTest Jun 2, 2026 · 04:10
   We tested 3,294 BTC favorites.
   Most had no edge.
   When a BTC favorite held the lead for 90+ seconds while Bitcoin was still moving, the win rate jumped to 84.4% and profits more than tripled.
   We scanned 3,294 BTC favorite checkpoints. Each checkpoint was taken every 30 seconds, using the observed token price as the entry price.
   The setup:
   • BTC 5-minute or 15-minute market
   • One side priced between 65c and 95c
   • That side had already been the favorite for more than 90 seconds
   • BTC moved at least 2 basis points from high to low in the prior minute
   That produced 1,135 checkpoints across 302 markets.
   Results:
   • Won 84.4% of the time
   • 958 wins and 177 losses
   • Average entry price was 80.8c
   • Made 3.6c per share before spread
   Across all 65-95c favorites, the baseline was 79.4% and 1.0c per share.
   The interaction mattered more than either ingredient alone:
   • Fresh BTC movement, short-lived favorite: 76.5%, 0.7c per share
   • Persistent favorite, quiet BTC tape: 78.3%, -1.4c per share
   • Persistent favorite, moving BTC tape: 84.4%, 3.6c per share
   We split the checkpoints chronologically. The setup made 4.1c per share in the first slice and 2.2c in the later slice.
   Why this works:
   A fresh 75c favorite can just be a fast repricing after one noisy tick. A quiet 85c favorite can be stale confidence. But a favorite that has held the lead for 90+ seconds while BTC is still moving is different: the prediction market is confirming direction while the underlying is still forcing traders to update.
   The risk is the payoff skew. At an average entry of 80.8c, one loss costs about 4.2 clean wins. The edge lives in refusing the lookalikes: no persistent lead, no recent BTC range, no trade.
   The next test is obvious: rebuild this as a live filter by coin, lead age, and prior-minute BTC range.
   Run that bucket map on fresh markets at @polybacktest.
   • 2 reposts
   • 17 likes
   • 1.5K views
5. 

   PolyBackTest May 31, 2026 · 23:48
   We mapped Polymarket BTC token price velocity second by second. There's a literal heartbeat at 0:00, 1:00, and 2:00. The market wakes up every 60 seconds.
   315 BTC 5-minute markets with 100ms snapshot data. ~24,000 ticks per 10-second window. We measured the average size of each price tick by elapsed time.
   Token tick size by elapsed seconds:
   0-10s: $0.00356 per tick (1.46x average)
   10-20s: $0.00251
   20-30s: $0.00248
   30-40s: $0.00242
   40-50s: $0.00224
   50-60s: $0.00209
   60-70s: $0.00276 (1.13x average)
   70-80s: $0.00244
   80-90s: $0.00222
   90-100s: $0.00221
   100-110s: $0.00202
   110-120s: $0.00211
   120-130s: $0.00304 (1.24x average)
   Three peaks. Three valleys. The token price moves the most at the start of every minute, and the least 10-20 seconds before each minute closes.
   The BTC price tells the same story. Bps-per-second:
   0-10s: 11.55 bps/sec
   60-70s: 10.91 bps/sec
   120-130s: 10.97 bps/sec
   Three local maxima, one every 60 seconds. The crypto market literally has a 60-second heartbeat.
   Why? Binance closes a 1-minute candle every minute. That candle close is when high-frequency traders rebalance, when algorithmic stops fire, and when market-making bots re-quote. The volume pulse on Binance propagates to Polymarket through the bots that watch both.
   The trade implication.
   The optimal moment to enter is the 30-second window BEFORE a minute close (e.g. 0:30-0:55, 1:30-1:55). Tick size shrinks, spreads tighten, token price barely moves. Cleanest fills.
   The optimal moment to TAKE PROFIT or read direction is the 10-second window after a minute close (e.g. 1:00-1:10, 2:00-2:10). Highest activity, biggest moves, direction commits.
   If you are scalping these markets, your entry timing relative to the minute hand might matter more than your signal.
   Historical data, 315 markets at 100ms resolution. Map your own intra-market timing at @polybacktest.
   • 1 reposts
   • 34 likes
   • 2.9K views
6. 

   PolyBackTest May 30, 2026 · 22:46
   When Polymarket says 83% on the favorite, the favorite wins 70%. We backtested 798 token observations at T-55s. There's a specific price band where the market is systematically wrong.
   We sliced every observed token price into 2-5 cent buckets and computed actual win rate vs market-implied probability.
   In the middle ($0.50-$0.70), calibration is clean. Implied 54%, actual 56%. Implied 64%, actual 61%. Tight.
   At the upper-middle, it breaks.
   Token $0.80-$0.85: market implies 83%, actual win rate 70%. Gap: -12.6%.
   Token $0.85-$0.90: market implies 87%, actual 81%. Gap: -6.3%.
   That is overconfidence. When the market thinks the trade is "almost certain" (83-87%), the favorite under-delivers by 6-12 points. Buying favorites in this band is a slow leak.
   Then at $0.90+ it flips clean again. $0.94-$0.96: actual 96% vs implied 95%. Calibrated. $0.98-$1.00: actual 99% vs implied 99%. Perfectly calibrated at the certainty extreme.
   The other broken band is the longshots.
   Token $0.10-$0.15: market implies 11.5%, actual 4.8%. Gap: -6.8%. Overpriced lottery tickets, eat the spread.
   Token $0.15-$0.20: market implies 16.7%, actual 34.3%. Gap: +17.6%. Underpriced.
   35 observations in that $0.15-$0.20 bucket is a small sample, but the relationship is consistent with favorite-longshot bias seen on every prediction market ever studied.
   The two trade angles this opens.
   Stop buying favorites in the $0.80-$0.85 zone. The market overpays for "almost certain" trades that aren't. Skip them or fade them.
   Underdogs at $0.15-$0.20 might be mispriced upward. Buy a few, hold to resolution. One winner pays for 5 losers.
   Both lines depend on regime. Overconfidence at the upper-middle could compress as bots refine pricing. The $0.15-$0.20 underpricing could be regime-specific (recent BTC chop, lots of late reversals).
   Historical data, 798 observations through May. Plug your own samples into the calibration curve at @polybacktest.
   • 6 replies
   • 6 reposts
   • 45 likes
   • 3.5K views
7. 

   PolyBackTest May 30, 2026 · 00:55
   How much does the spread actually cost you on Polymarket? We backtested every popular strategy. The hidden spread tax kills more edges than it spares.
   The "spread" on Polymarket BTC 5-minute markets is the gap between UP+DOWN and $1.00 at market open. Buy one side, sell the other side, the difference is what the market makers keep. We measured it across 1,499 markets.
   Median spread: $0.0100
   Mean spread: $0.0125
   90th percentile: $0.0200
   That is 1-2% baked into every round trip. Half of it falls on the buyer (you), the other half on the counterparty. It is small per trade and lethal across thousands.
   Now plug it into a real strategy. "Buy the favorite when the opening token is >=$0.55." We tested it on 770 markets.
   Win rate: 59.6%
   Gross EV: +$0.05 per $10 trade
   Net EV after spread: -$0.02 per $10 trade
   The spread swung the strategy from positive to negative. The tax ate 138% of the gross edge.
   Late-entry strategies have it even worse. At T-55s, when BTC has clearly moved >0.10%, the winning token costs $0.96. Gross profit per win is $0.04 on a $10 stake. After half-spread on entry plus Polygon gas, you keep maybe $0.02. A 99% win rate that nets 2 cents is unpaid labor.
   The strategies that DO survive: ones where the gross edge is wider than the spread. Mean reversion in the 0.30-0.40% bucket (60.6% WR, entry at $0.55, gross EV +$1.86 per $10) keeps roughly $1.50 net. 1-hour streak reversal (64% WR, entry near $0.50) keeps almost all of its 28% gross edge.
   The 1% rule: if your gross EV per trade is under 1.5% of your stake, the spread alone kills it. That is before you add platform fees or Polygon gas, both of which only make it worse.
   Every "high win rate" strategy needs to clear this bar before you wire a dollar. Most don't.
   Historical data, 1,499 sampled markets. Compute your own gross vs net spread on every strategy at @polybacktest before deployment.
   • 5 replies
   • 21 likes
   • 2.5K views
8. 

   PolyBackTest May 28, 2026 · 23:23
   "28 tools, 6 layers, $1M bot built on Polymarket's 2.7 second lag." We measured the lag directly. It does not exist.
   The entire article rests on one claim: "Polymarket updates prices 2.7 seconds slower than Binance. That gap is where every strategy lives." Everything downstream (the Markov chain, the Kelly sizing, the 28-tool stack) depends on this gap being real.
   We tested it.
   32,832 BTC price events. 100ms-resolution Polymarket snapshots paired with Binance second-level data. For every sharp BTC move, we measured how long it took the Polymarket token to reprice.
   The answer is not 2.7 seconds. Polymarket reprices in the same 100ms window as Binance moves.
   Buy on a confirmed Binance move, sell 1 second later: profitable 30% of the time.
   Sell 3 seconds later: 45%.
   Sell 5 seconds later: 50%.
   Sell 10 seconds later: 51%.
   If a 2.7-second lag actually existed, those numbers would be 80%+. They land at coin-flip. The Polymarket orderbook is not lagging behind Binance. It is moving with Binance, sometimes ahead of it on the millisecond scale. The 2.7-second window existed in 2024 marketing posts. It does not exist in fresh data.
   The "Claude +1,322% vs OpenClaw liquidated in 48 hours" line is the same recycled story we have seen in three other "I built a Polymarket bot" articles. No on-chain verification, no methodology, just a number designed to sell the next thing in the funnel.
   Now the 28-tool stack. It is impressive-sounding wallpaper. OpenBB, FRED, MCP servers, Qwen3-Coder, Claude Squad, MiroThinker, multiple "orchestration" layers. Out of 28 listed tools, exactly 2 do work that would matter (the backtest repos in Layer 5). The other 26 are LLM agents, data aggregators, and dashboards. None of them tell you whether your edge exists. The single thing that would (a real backtest against historical data) gets buried at layer 5 next to a paragraph admitting most bots skip it.
   That admission is the part to read twice. They write: "Reasoning models can sound convincing about strategies that have never made money." That is the article describing itself.
   The wallet they cite is on-chain real. But the PnL on big Polymarket wallets we have studied comes from cross-venue arbitrage or pure market making, not the 2.7-second latency arbitrage they describe. The latency story is a marketing wrapper on top of unrelated trading.
   The honest math. We have backtested every signal the post lists as profitable. Latency arbitrage: dead. Distance-from-strike at minute 4: real but pays $0.04 on $0.96 net of fees. Time-of-day edges: real but decay in weeks. Markov-on-prices: matches market calibration, no edge.
   You do not need 28 tools to test a strategy. You need one backtest against fresh historical data and the honesty to run it before you wire collateral.
   Historical data through May 26. The 2.7 second lag is not in it. Verify before you build at @polybacktest.
   • 5 replies
   • 1 reposts
   • 33 likes
   • 3.1K views
9. 

   PolyBackTest May 28, 2026 · 11:22
   "$60M in bot profit, Markov chain at 87% persistence, win rate 63-72%." We plugged his own formulas back in. The strategy contradicts itself.
   He writes the formula plainly: enter when model probability p̂ exceeds market price q by at least 5%, where r = (1-q)/q is your return per win. He says at q = 0.647, r = +54.5%. The math is correct.
   Then he says actual win rate at the entry threshold is 63-72%.
   Drop those numbers into expected value:
   At q = 0.647 and win rate 65%: 0.65 x $0.353 - 0.35 x $0.647 = -$0.0035 per dollar. Negative.
   At q = 0.647 and win rate 72%: 0.72 x $0.353 - 0.28 x $0.647 = +$0.073 per dollar. Positive, until fees.
   Polymarket charges fees on these markets. Add the ask-vs-mid spread. That +$0.073 becomes maybe +$0.03-$0.04 net. The lower end (65%) is flatly negative.
   The 63% he quotes is roughly what the market already prices into a $0.65 token. We have backtested Polymarket calibration directly: tokens at $0.60-$0.65 win 62.8% of the time. The market is well-calibrated at the center. A Markov chain on BTC price states cannot beat that, because the market already sees the same price states.
   His Kelly sizing makes it worse. He cites f* = 0.71 at p = 0.87. That f* assumes the 87% number is true. If actual win rate is 65%, Kelly is closer to 0.11. Sizing at 71% of bankroll on a strategy whose real win rate is 65% is a fast track to ruin. We have backtested all-in betting at high-but-realistic win rates: even at 99% WR, betting all-in wipes out half the players by trade 69.
   The "self-learning loop" where Claude rewrites the rules nightly is automated p-hacking. We ran exactly this: 1,886 strategies tested on BTC markets. Best in-sample 77.8%. Same strategy out-of-sample: 25%. Nightly retraining on yesterday's wins looks brilliant for a week and bleeds forever.
   The wallet links are real and on-chain. The PnL on those wallets is real. But none of those bots are running this particular Markov story. The big Polymarket wallets we have studied are cross-venue arbitrage or market making operations. The Markov persistence narrative is bolted on top of unrelated profits to sell the rest of the funnel.
   That funnel: atomicbot signup, paid Claude API recommendation, three referral links in the wallet citations.
   The honest math on BTC 5-minute markets. Favorites at open win 54-58%. After fees and spread, that is roughly breakeven. The only edges that net real money are filtered time windows and they decay in weeks. Not a self-improving Markov agent sized at 71% Kelly.
   Plug your own win rate into his EV formula before you wire collateral at @polybacktest.
   • 4 replies
   • 3 reposts
   • 30 likes
   • 4.2K views
10. 

    PolyBackTest May 26, 2026 · 08:29
    There is one hour of the day when BTC and ETH both go UP 68% of the time on Polymarket. Same hour. Both coins.
    1,418 BTC and 1,437 ETH 1-hour markets, March 24 to May 22. We checked the UP rate for every hour of the day.
    Most hours sit at 48-52%. Coin flips. Then 19h UTC.
    BTC at 19h UTC: 40 UP out of 59. 68%.
    ETH at 19h UTC: 41 UP out of 60. 68%.
    Two independent coins, the exact same hour, the exact same 68% UP rate. 119 markets combined.
    19h UTC is 3pm New York time. The final stretch of the US equity session, the window where institutional desks square positions into the close. In this sample, that flow ran consistently bullish for crypto.
    The flip side. The bearish hours:
    BTC 16h UTC: 37% UP.
    ETH 22h UTC: 32% UP.
    SOL 10h UTC: 32% UP.
    SOL has its own bullish hour: 15h UTC at 67% UP across 58 markets, where BTC and ETH are weaker.
    Each coin has its own clock. But 19h UTC lighting up on both BTC and ETH at exactly 68% is the standout. When two assets agree on a time-of-day bias, the signal is harder to dismiss as a fluke.
    The caveat we always give: 59-60 markets per hour is a starter sample, and time-of-day edges decay. We have watched a 6am UTC crypto bias worth 7.8% disappear in a matter of weeks. 19h UTC is bullish in this regime. It will not be forever.
    The durable play is not "bet UP at 19h forever." It is to scan the hourly heatmap weekly and trade the hours that are currently loaded.
    Historical data through May 24, regimes shift. Build your own hourly heatmap for BTC, ETH and SOL at @polybacktest.
    • 1 replies
    • 3 reposts
    • 12 likes
    • 1.2K views

Polybacktest Reviews & Comments

No reviews yet. Be the first to share your experience.