A Polymarket Bitcoin bot is software that quotes or takes liquidity in the short-window BTC binary markets that resolve every fifteen minutes, every hour, or at the close of each UTC day. The markets look made for automation because the underlying is observable to anyone with an internet connection and the resolution rule is a single number lifted from a public price feed. What makes the problem interesting, and what catches most new entrants, is the gap between the market price and a textbook binary option. That gap is where a careful bot earns and where a careless one quietly bleeds.
The shape of Polymarket BTC binary markets
Polymarket lists Bitcoin price markets at three native cadences. The shortest are the fifteen-minute markets, which ask whether the BTC price at minute M+15 will be above some strike given the price at minute M. The middle cadence is hourly. The longest of the routinely-bot-traded set is the daily market, which asks whether the BTC price at 23:59 UTC will be above the strike. Polymarket also lists weekly and monthly markets, but the population that interacts with those at the order-book level is mostly human, not algorithmic.
Each market is a YES or NO contract that pays one dollar to whichever side is correct at resolution. The implied probability is the price of the YES contract in cents divided by one hundred. A YES quoted at 47 cents implies a 47 percent chance the threshold is breached by resolution. This is the cleanest payoff structure on Polymarket because the underlying is a continuous, liquid, globally observable variable. There is no debate about whether the event happened, only about which side of the line the price was sitting on at the resolution timestamp. The plumbing that underlies any bot on this venue is described in the order book explainer, and the population-level evidence on whether the venue prices probabilities correctly is in the accuracy study. The Bitcoin markets sit at the most efficient end of the venue and pricing inefficiencies that survive there are smaller, faster, and more technical than what you find in political or sports markets.
How the price feed and resolution actually work
The single most important fact about a Polymarket Bitcoin bot is the resolution rule. Polymarket does not resolve these markets against the mid-price on Coinbase, or the volume-weighted index across exchanges, or any other intuitive guess. The resolution rule names a specific source and a specific timestamp. For the bulk of BTC binary markets on the venue, that source is a published price at a published minute, and the venue documents the exact rule on the market detail page before listing. Reading that rule and matching its tick boundary is the first piece of homework.
The rule has three observable parts. The first is the price source. Different markets pull from different feeds and the feeds do not agree to the dollar at any given instant. Spreads between major venues for BTC routinely sit at two to eight dollars during normal liquidity, and they widen meaningfully during stress. A bot that prices against the wrong feed will mispredict resolution by that spread, in expectation, on every contract. The second part is the timestamp granularity. The reference is typically the price at the close of a one-minute candle, not the last trade or the book mid. The third part is the resolution latency: the on-chain settlement that flips the contract from open to YES or NO does not happen the instant the candle closes. It happens a few minutes later, after the oracle confirms the data point. During that gap the contract still trades.
The combined effect is that the market is not really a binary option on the BTC mid. It is a binary option on a specific feed at a specific timestamp, plus a brief settlement lag during which the outcome is known but the contract still has a price. That lag window is one of the cleanest sources of edge available to a competent bot, and the easiest place to lose money fast if your data path is even slightly slower than the rest of the order book. Polymarket publishes the resolution methodology in its venue documentation, and the underlying price feeds for BTC are observable through public sources such as CoinGecko and the major exchanges directly. A bot that does not pin its model to the exact feed Polymarket pins its oracle to is, mathematically, trading a slightly different instrument than the one it is quoting on.
Implied probability versus a Black-Scholes binary
The theoretical anchor for any BTC binary bot is the Black-Scholes binary call. Given current price, strike, time to expiry, and an estimate of implied volatility, the price of a binary cash-or-nothing call is the normal cumulative distribution function applied to standardised log-moneyness. For a fifteen-minute horizon on Bitcoin, with annualised vol around seventy percent, the model produces a probability that depends almost entirely on how far current spot is from the strike. At the money, fifty percent. Five hundred dollars above on a sixty-thousand-dollar Bitcoin price, the high sixties. A thousand above, low eighties.
The market price differs from the model for three reasons. The first is cost of capital: rounding error on a fifteen-minute horizon, a few basis points on a daily, not load-bearing. The second is fees and slippage. The spread on a fifteen-minute BTC market commonly sits at two to four cents near the money. That means a taker needs roughly three-percent confidence to enter and a market maker quoting both sides needs around one percent. A meaningful hurdle.
The third, least understood by new entrants, is the resolution-feed risk premium. The market price implicitly accounts for the chance the venue, oracle, or feed does something the bettor cannot predict from spot alone. Small in normal times, large during outages or disputes. A bot using pure Black-Scholes and ignoring this premium will buy contracts the market correctly considers slightly overpriced and miss the cases where the market is rightly worried about something the model cannot see.
The 15-minute window: where the bot edge hides
Of the three native horizons, the fifteen-minute market is the one where a careful bot can most plausibly find edge. The reason is not that the math is harder for humans, although it is. The reason is that the ratio of book churn to human attention span is highest at this horizon. The market reprices every few seconds. A human checking once a minute and clicking once a minute is structurally outpaced; a bot at one hundred millisecond cadence is operating at the timescale that matches the order flow.
A 15-minute BTC binary: implied probability tracks moneyness, with the kink at resolution
The edge in this market is not a directional view on Bitcoin. Nobody, including very good directional traders, has a meaningful edge on the BTC mid over fifteen minutes. The edge is in the conversion from a continuously updating spot price to a discrete YES or NO probability, and specifically in being faster, calmer, or better-calibrated than the resting orders on the book during the last two minutes before resolution. A bot that maintains a consistent model of implied probability and re-quotes at one-hundred-millisecond cadence can pick off resting orders that have not refreshed at the same speed. That, on a venue with zero maker fees, is the structurally available return.
The fifteen-minute, hourly, and daily windows differ along several dimensions that matter for bot suitability.
| Window | Typical liquidity | Typical spread (near ATM) | Gamma sensitivity | Feed-latency edge | Bot suitability |
|---|---|---|---|---|---|
| 15-minute | $50k–$300k notional per market | 2–4 cents | Very high in final 2 minutes | Dominant source of edge | Best fit for fast bots |
| Hourly | $100k–$600k notional | 2–5 cents | High in final 10 minutes | Meaningful, smaller | Good for medium-latency bots |
| Daily (close) | $300k–$2M notional | 1–3 cents | Modest until last hour | Marginal except at close | Better for vol or carry plays |
Notional figures are rough order-of-magnitude estimates and vary with overall BTC volatility regime. The gamma column is the most important for thinking about where bot edge lives. In the final two minutes of a fifteen-minute contract a one-dollar move in BTC can shift the fair implied probability by two to five cents. That is the regime in which milliseconds matter and in which a slow quote is a free option for the rest of the market.
Latency budget for a price-feed bot
If the edge concentrates in the last two minutes and the contract reprices on every meaningful tick of the underlying, a Polymarket Bitcoin bot is structurally a latency-sensitive system. The budget breaks into four legs.
The first is the BTC price feed itself. A direct WebSocket subscription to a major exchange delivers tick updates in five to twenty milliseconds from the matching engine to a server in the same region. Using a free public REST endpoint can push this leg into the hundreds of milliseconds and erase any edge the rest of the system can offer. The second leg is internal compute: take the new spot, compute the new fair implied probability, decide whether the resting quote needs to move. Well under a millisecond.
The third leg is the Polymarket order-entry path. Submitting a signed REST order to the CLOB API and getting an acknowledgement takes fifty to one hundred fifty milliseconds from a well-placed server. Session keys matter here: a bot that round-trips a wallet signature on every order is multiples slower than one operating with a pre-authorised scoped key. The fourth leg is the matching-engine reaction, dominated by the venue itself.
End to end, a competent Polymarket Bitcoin bot reacts to a meaningful spot tick in roughly one hundred to two hundred milliseconds. Faster requires colocation that is hard to justify for retail capital; slower means quoting against bots that will pick off your resting orders during the high-gamma final minutes. The full architecture template is in the build-a-bot walkthrough.
The kink risk: what happens near the strike at resolution
The most uncomfortable property of any binary option is the kink in the payoff function at the strike. A binary that pays one dollar above the threshold and zero below has a delta that approaches infinity as spot approaches the strike at expiry. A one-dollar move in BTC during the final ten seconds can flip the contract from a near-certain YES to a near-certain NO, and the implied probability has to chase that move.
For a market maker, the kink is the dominant risk. Inventory accumulated at fifty cents per contract becomes worth zero or one hundred cents in moments, with no opportunity to hedge. A naive market-making strategy that quotes both sides at narrow spreads through the resolution moment will, on average, break even on the spread and lose meaningfully on the kink. The standard response is to widen quotes aggressively as resolution approaches near the money and to skew inventory toward whichever side spot is currently favouring. A bot that does not implement that risk-aware widening produces a return distribution that looks fine for most trades and catastrophic on the few markets that resolve close to the strike.
For a directional bot that takes positions rather than makes them, the kink is the opposite of a problem. It is the source of the asymmetric payoff the strategy is built around. The kink also interacts with the resolution feed lag in a way that produces both the cleanest edges and the most expensive mistakes on the venue. During the few minutes between candle close and on-chain settlement, the outcome is determined but the contract still trades. A bot that observes the resolving candle a few milliseconds faster than the rest of the book can lift mispriced quotes from slower participants. A bot whose feed is even slightly stale during that window is, conversely, the one being lifted.
Capital efficiency and sizing
The capital math is more forgiving than a traditional binary option desk because the venue charges zero maker fees and posts collateral in stablecoin. A bot quoting both sides with a thousand dollars of inventory and a two-cent spread captures, in the ideal case, twenty dollars per round-trip. Realised capture is a fraction of that because of inventory drift, kink risk, and the fact that not every quote round-trips. The realistic order of magnitude for a well-run market-making strategy is one to three percent of inventory per market on average, with high variance.
For sizing, the Kelly criterion is the right starting framework but needs care because the payoff is bimodal. The fractional-Kelly logic in the Kelly piece applies, with the caveat that the variance term should be computed against the realised bimodal distribution rather than a Gaussian approximation that underestimates tail risk near the strike. A taker bot at a quarter or a fifth of full Kelly is in roughly the right neighbourhood for retail capital. The secondary consideration is capital turnover: a fifteen-minute market locks collateral for fifteen minutes, so a bot can recycle the same dollar through ninety-six markets per day. That turnover compounds modest per-trade edges into meaningful annualised return even on modest capital.
Honest expectations and failure modes
The five most common ways a Polymarket Bitcoin bot loses money cover most of the realistic downside. The first is feed mismatch. A bot that prices against a feed that diverges from the resolution feed by a few dollars will, in expectation, be wrong by that much on every trade. The fix is to pin the model feed to the exact source named in the resolution rule and to monitor the basis between that feed and any backup continuously. The second is kink exposure: a market maker that does not widen quotes near the strike near resolution will produce one or two catastrophic trades per month that erase the previous month of positive carry.
The third is latency degradation that is too gradual to notice. A bot whose order-entry path was one hundred milliseconds when deployed can drift to three hundred over months as the broader market gets faster, and the operator does not see the change because each individual trade still looks fine. The fix is to instrument the full path and alert on degradation. The fourth is over-fitting historical volatility: a bot calibrated against the trailing thirty days of BTC vol can be badly miscalibrated during a sudden regime change, when the right response is to widen everything until the new regime stabilises.
The fifth catches new operators most often: trading through resolution disputes. When the price feed is contested, the oracle is delayed, or the venue pauses a market, the contract may continue to trade off-venue or in late-arriving orders. A bot that does not check the market status flag can take the wrong side of a dispute and find itself holding a position that will not resolve at the price its model predicted. The fix is a status check in the order path and a kill-switch that disengages the bot entirely when any abnormal flag appears.
A Polymarket Bitcoin bot is not a way to predict BTC. It is a way to monetise the gap between a continuously moving spot price and a venue that prices discrete probabilities in cents. The bots that work do that conversion faster and more carefully than the rest of the book; the bots that do not work try to predict BTC and lose to whoever is faster on the conversion.
Frequently asked questions
What is a Polymarket Bitcoin bot?
It is software that quotes or takes liquidity in Polymarket BTC binary markets, using a live Bitcoin price feed and a model of implied probability to decide which side of a contract is mispriced relative to the resolution rule.
Which Bitcoin price feed does Polymarket use to resolve these markets?
Polymarket names the price source on each market detail page. Different markets can use different feeds, so the first step in building a bot is reading the resolution rule for the specific market and pinning the model to that exact source rather than to a generic BTC index.
Can a Polymarket Bitcoin bot predict the BTC price?
No. The edge in these markets does not come from predicting Bitcoin. It comes from converting a continuously moving spot price into a discrete YES or NO probability faster and more carefully than the resting orders on the book, particularly during the high-gamma minutes just before resolution.
What capital do I need to run a Polymarket Bitcoin bot profitably?
A taker strategy can be tested on a few thousand dollars because each fifteen-minute market only locks collateral for the window. A market-making strategy is structurally more capital intensive because it needs depth on both sides and capacity to absorb the kink risk near the strike. A realistic floor for serious market-making on these markets is in the tens of thousands of dollars.
How dangerous is the resolution kink for a beginner bot?
It is the single largest source of one-sided loss for a market-making strategy that does not widen quotes near the strike near resolution. A bot that quotes naively through the final two minutes of every fifteen-minute market will eventually take an inventory hit that erases weeks of carry. The standard mitigation is aggressive quote widening and inventory skewing as the contract approaches resolution near the money.
About the author
Maria Ostrowski leads quantitative research at Poly Syncer. She holds a PhD in Statistics from the London School of Economics and spent four years as a quant researcher on a European equity options market-making desk before moving to prediction-market research. She owns the wallet-scoring composite and the Sharpe adaptation that ship in the Poly Syncer product, and writes about the parts of automated trading that are interesting enough to be worth the work.