There Is a Riskless Trade Sitting Next to Polymarket's Bitcoin Ladder. It Pays 0.37 Percent a Year Against a 4.02 Percent Treasury Bill.

By Kresmion Research, August 5, 2026

Polymarket's event on where Bitcoin gets to before 2027 carries thirty two live contracts, twenty on reaching a given price and twelve on dipping to one. We pulled all thirty two from the venue's API at 09:05 UTC on August 5, 2026, and every rung below is quoted as of that moment.

Two pairs in the twenty rung ladder are priced in an order that cannot be right, and a twenty first contract sitting outside it makes a third case.

A twenty-first contract, market 573656, asks the same question as the $150,000 rung and trades at 3.85 percent. That is above the $140,000 rung at 3.40 and above the $130,000 rung at 3.65. Interpolating between the $120,000 and $130,000 rungs, 3.85 percent is roughly where a $127,600 strike belongs.

Bitcoin traded near $64,100 this morning, so the rungs range from a 9 percent move at $70,000 to more than a fifteenfold move at $1,000,000.

Which of these can actually be acted on

The two inside the ladder are small, and they are not the same kind of thing.

The $170,000 case is a midpoint artifact. That contract quotes 2.10 bid at 2.90 offered, which straddles the $160,000 contract's 2.20 by 2.60 entirely. Nothing there is executable and we are setting it aside.

The $190,000 case is real but tiny. It bids 2.20 against the $180,000 contract's 2.10, with both offered at 2.30, so unlike the $170,000 case it is inverted at the bid as well as at the midpoint. Those two books are 0.10 and 0.20 points wide, and the midpoint gap is five basis points. Worth naming, not worth much else.

The 573656 case is different, because there the books genuinely cross.

Market 573656 and market 701491 ask the identical question at the identical strike, with the same December 31 deadline and the same settlement source, the Binance BTC/USDT one minute candle high. They differ only in when their observation windows opened. Because neither has resolved, Bitcoin has not touched $150,000 in either window, so from here nothing can pay one without paying the other. Going forward they are the same contract.

They are quoted 1.20 points apart. And 573656's bid at 3.80 sits above 701491's offer at 3.00, so the two books cross by 0.80 points.

The trade is real, and it is free of price risk

Buying the NO side of 573656 at 0.962 and the YES side of 701491 at 0.030 costs 0.992. That position pays exactly one dollar whatever Bitcoin does. If Bitcoin touches $150,000, the 701491 leg pays and the 573656 leg does not. If it never touches, the 573656 leg pays and the 701491 leg does not. There is no price path where both pay and none where neither does.

Riskless here means free of price risk, and nothing more. The position still depends on Polymarket listing both contracts through December 31 and resolving them against the same source, and on both legs filling before the quotes move. Over a five month hold those are not nothing.

Polymarket charges takers a fee, and it publishes the formula: the size multiplied by the fee rate, multiplied by the price, multiplied by one minus the price. All the contracts here return a rate of 0.07. On this pair that comes to about 0.0046 per share across both legs, so the all in cost at the top of the book is 0.9966 for a certain one dollar. Averaged across the whole fill it is 0.9985, and the deepest tranche costs 0.9995.

Walking both books until the edge disappears, the whole trade is 340.40 shares. It costs $339.89 and returns $340.40 on December 31.

The gross edge across those 340 shares is $2.05. The fee takes $1.54 of it, three quarters. What is left is 51 cents. That is 0.149 percent over 148 days, or about 0.37 percent annualized, on capital that cannot be touched until New Year's Eve.

So two things are holding this in place, and it is worth separating them. The fee removes three quarters of the edge before anyone gets paid. The carry removes the rest of the reason to bother. Three hundred and forty dollars locked up for 148 days to make fifty one cents is 0.37 percent annualized, against 4.02 percent on a six month Treasury bill as of August 3, or about 3.98 percent interpolated to the same 148 day horizon. Any amount of capital large enough to matter would move the price before it filled.

Who pays the fee matters here. Polymarket's schedule marks it taker only and its documentation says makers are never charged, so a participant posting rather than crossing would keep the whole $2.05. That is not the same opportunity: a resting order has no guaranteed fill, and waiting to be hit is a quote rather than an arbitrage.

A third constraint may matter more than either. The net edge is 0.0015 per share on a book that ticks in 0.001, and the trade has to be assembled across five price levels because the top bid is only 26.44 shares deep. One tick of slippage on the way in removes two thirds of it. A position this thin against its own tick size is hard to capture even when the arithmetic says it is there.

The gap converged once already

We have both contracts in our capture on 19 days, beginning July 15. These are daily averages of every observation we took that day, so they differ slightly from the 09:05 snapshot quoted above.

This matters more than the current reading. The gap was 1.30 on July 15, closed almost completely to 0.09 by July 25, and reopened to 0.57 within two days. It crossed the other way on August 1, widened through August 4, and narrowed slightly on August 5. Today's 1.21 is the third widest of the 19 days, not the widest, and this pair has closed a gap this size before. Anyone describing the last two weeks as a one way divergence would be reading a window that starts at the minimum.

The readings from July 15 to July 22 are thinner than the rest, one to seven observations a day against 13 to 34 afterward, so we weight them less but keep them in. Dropping them would have produced a cleaner trend than the data supports.

One caution on the series itself: 701491 turned $1,056 in the last 24 hours against 573656's $26,074, and quotes 2.30 by 3.00, a seventy basis point spread. A good deal of the movement in that column is a quiet midpoint wandering inside a wide book.

What else could explain it

Very cheap contracts are known to trade rich against their true odds, and that would push the whole low end of this ladder up. It is a reasonable candidate for the $170,000 and $190,000 cases.

It cannot explain 573656 against 701491. Those two carry the same strike at almost the same price, so any bias that depends on how cheap a contract is applies to both of them equally.

A second candidate is concentrated flow. Market 573656 turned $26,074 in 24 hours against $7,316 for the busiest of the fifteen contracts Polymarket listed together on November 24, a ratio of 3.6 times. Measured against only the eleven of those that ask about reaching a price, where the busiest is the $100,000 rung at $2,045, the ratio is 12.8 times. The second comparison is the more flattering one and we would rather show both. We measured the volume either way. We did not establish the reason for it: we did not examine the venue's page layout, and the two contracts sit in different event groups in Polymarket's own metadata.

The limits of this

The ladder is mostly right, and the twelve contracts we have not discussed are cleaner still: the twelve contracts asking whether Bitcoin dips to a given price are ordered correctly at every step, from a 78.50 percent chance of dipping to $60,000 down to 1.85 percent at $5,000. On the up ladder, twenty rungs give nineteen adjacent pairs, and seventeen of those nineteen are ordered correctly. The two that are not are ten and five basis points apart at the midpoint, one of them visible only there, and the one crossing that can actually be traded involves a duplicated contract. Nothing here says these venues are broken.

On the questions that carry real money the books are not thin. Polymarket's market on whether the Clarity Act is signed into law in 2026 was quoted at 14.5 percent at 09:05 UTC on $673,220 of 24 hour volume, about 26 times the volume of the contract we have spent this article on. That market is itself a lesson in freshness: its 24 hour volume was $125,659 at 07:05 and five times that two hours later.

The other major venue is quieter in a way worth naming. Of Kalshi's seven contracts on whether Bitcoin trades above a given level at any point before the end of 2026, five traded nothing in the past 24 hours, and the venue's own quote timestamps on all seven read April 9, 2026. They still carry bid and ask sizes and between 73,042 and 329,935 contracts of open interest, so they are stale quoted rather than dead. Their lifetime volumes run from 265,370 to 1,249,237 contracts, which is why a book nobody has quoted since spring can look heavily traded to anyone reading the lifetime field instead of the daily one.

Where our own numbers have the same problem

Kresmion publishes a prediction market consensus layer. At 08:15 UTC this morning it carried:

As of that 08:15 UTC read, our $150,000 line sat above our $140,000 line. That was the same inversion, in our own product. See the update at the end of this article: it no longer does.

It is not inherited, and the cause is narrower than a weighting problem. Our $150,000 row is built on 573656, the orphan duplicate this article has spent its length arguing is the same contract as 701491. Our $140,000 row is built on 701492, the ladder sibling. Substitute 701491's 2.65 percent for 573656's 3.85 and the blend falls below our $140,000 row and the inversion disappears. We picked the wrong one of two identical contracts.

Both rows do blend two venues, and on mid prices Kalshi has the pair the right way round, 4.50 percent on $140,000 against 3.50 on $150,000, so the Polymarket leg is deciding the output. The leg we selected is what produced the inverted number.

A third fault while we are here. The two legs we blend do not settle the same way: Kalshi resolves against the CF Bitcoin Real Time Index and Polymarket against the Binance BTC/USDT one minute candle high. We applied a settlement identity test carefully to the Polymarket pair earlier in this article and never applied it to the cross venue blend we actually publish.

The honest complication is that the Kalshi quotes we blend are the same ones we describe above as unquoted since April. So one leg is live and wrong, the other is stale and right, and we published a number that took the worst of both.

Two further problems. We publish 3.83 percent and 2.65 percent for the same event under two separate keys, because we inherited the venue's duplicate instead of collapsing it. And of the 1,817 rows in that layer at the time of that read, 1,808 drew on a single venue, so for most of them the word consensus was describing something that is not there.

We are collapsing the duplicate keys, adding an ordering check so an inverted rung cannot publish, gating stale venue legs out of the blend, and relabelling single venue rows. None of that had shipped when this was written this morning.

What would change this read

If 573656 and 701491 converge again the way they did into July 25, then this is a recurring gap that closes on its own and the current reading is a wide instance of a normal pattern.

If they keep separating while volume stays concentrated in 573656, then nothing in the current structure has a paying reason to close it.

The observable things are the bid and offer on 573656 against 701491, whether 701491 stays listed, whether Polymarket's fee schedule or rebate tiers change, and whether the $170,000 and $190,000 rungs sort themselves out. All of them are public.

Key takeaways

Frequently asked questions

Is this actually an arbitrage?

Yes, in the sense that the payoff is identical in every outcome. At the top of the book it costs 0.9966 after fees for a certain one dollar, and about 0.9985 across the full 340 shares. It is worth 51 cents on $339.89 tied up until December 31, about 0.37 percent annualized. Both legs still settle through separate oracle proposals, so it is free of price risk rather than free of all risk.

Why has it not been arbitraged away?

Three things sit on it at once. The fee takes 75 percent of the gross edge. What survives is 0.37 percent annualized against 4.02 percent on a six month Treasury bill, on money that cannot be moved until December 31. And the net edge is smaller than two ticks on a book whose top level holds 26 shares, so a single tick of slippage while assembling it removes most of what is left. The gap did close once, to 0.09 points on July 25, before reopening.

Does the fee explain the mispricing?

Partly. The fee takes three quarters of the gross edge, which is most of it. But Polymarket charges takers only and exempts makers entirely, so someone posting rather than crossing pays nothing and the fee cannot be the whole reason the price is wrong.

Does this mean prediction markets do not work?

Twenty rungs give nineteen adjacent pairs and seventeen of them are ordered correctly, and the venue's Clarity Act market traded $673,220 in a day. The problems here are concentrated in duplicated and neglected contracts.

Does Kresmion have this problem too?

It did. At 08:15 UTC our consensus layer published the same inversion, and not because we inherited it: we selected the orphan duplicate over its ladder sibling. All four fixes shipped later the same day and the ladder now reads in the correct order. The update at the end of this article records the change.

Correction, August 5 2026

An earlier version of this note reported, alongside those quote timestamps, that all seven Kalshi contracts show zero posted liquidity. That figure was accurate and it was uninformative, and it should not have been set next to evidence that does carry information. Checking afterwards, Kalshi's liquidity field reads zero on every market we sampled, including one that traded 257,470 contracts today, so it is an unpopulated field rather than a measurement of those seven books. The clause has been removed from the body and from the sources note. The zero 24 hour volume, the bid and ask sizes, the open interest and the April 9 quote timestamps are unaffected, and none of the article's figures or conclusions depend on the removed clause.

Update, August 5 2026

The four fixes described above shipped later the same day, and the inversion in our own layer is gone. Recording it here because the disclosure above is written in the present tense of this morning and would otherwise describe a defect that no longer exists.

Our year end Bitcoin ladder now reads, as of 16:18 UTC:

Correctly ordered. Market 573656, the orphan duplicate, now feeds no row at all, and the second key that published a separate number for the same event is gone. Every row above is labelled single venue rather than consensus, and the Kalshi legs are recorded as withheld for staleness rather than blended in. The layer holds 1,608 rows of which 1,605 draw on a single venue, so the relabelling covers almost all of it.

What has not changed is the reason any of this was worth writing down. The ladder was wrong because we picked the orphan duplicate over its ladder sibling, which is a one line selection fault rather than anything deep, and it published for months without anyone noticing because nothing checked whether a ladder we publish is monotonic. The ordering check is now that something.

Sources

Polymarket gamma API, event 89502, thirty two live contracts of which twenty ask whether Bitcoin reaches a given price by December 31 2026, fetched 09:05 UTC on August 5 2026, with the $170,000 rung at 2.50 percent above the $160,000 rung at 2.40 percent on a book quoting 2.10 by 2.90 that straddles the $160,000 book entirely, and the $190,000 rung bidding 2.20 against the $180,000 rung's 2.10 with both offered at 2.30, an inversion on the bid rather than the midpoint: https://gamma-api.polymarket.com/events/89502

Polymarket gamma API, market 573656 at 3.85 percent, bid 3.80 offered 3.90, on $26,073.81 of 24 hour volume, unresolved, sharing the December 31 2026 deadline and the Binance BTC/USDT one minute candle high settlement with market 701491 at 2.65 percent, bid 2.30 offered 3.00 on $1,055.97 of 24 hour volume: https://gamma-api.polymarket.com/markets/573656

Polymarket CLOB order books for the 573656 and 701491 YES tokens, fetched 09:05 UTC on August 5 2026. Buying the NO side of 573656 and the YES side of 701491 clears 340.40 shares for $339.89 including fees, against a certain payout of $340.40 on December 31 2026, a profit of $0.5063, 0.149 percent over 148 days or about 0.37 percent annualized: https://clob.polymarket.com/

Polymarket fee documentation and the feeSchedule returned by the gamma API on all four contracts, giving feesEnabled true, feeType crypto_fees_v2, rate 0.07 and takerOnly true, with the published fee equal to size multiplied by the rate, the price and one minus the price, and makers never charged. Across the 340.40 share trade the gross edge is $2.0510 and the fee is $1.5447, 75.3 percent of it: https://docs.polymarket.com/trading/fees

Polymarket rebate programmes. The feeSchedule returned by the gamma API carries rebateRate 0.2, which the fee page describes as the crypto maker rebate, while a separate taker rebate programme runs to 50 percent at the highest thirty day volume tier. No figure in this article applies either, and every number here is computed at the full 0.07 rate: https://docs.polymarket.com/programs/taker-rebates

Polymarket gamma API, the twelve live contracts in event 89502 asking whether Bitcoin dips to a given price by December 31 2026, fetched 09:05 UTC on August 5 2026, ordered correctly at every step from 78.50 percent at $60,000 to 1.85 percent at $5,000: https://gamma-api.polymarket.com/events/89502

Federal Reserve Economic Data, six month Treasury constant maturity DGS6MO at 4.02 percent and three month DGS3MO at 3.91 percent, observation date August 3 2026: https://fred.stlouisfed.org/series/DGS6MO

Polymarket gamma API, the fifteen live contracts created on November 24 2025 inside event 89502, of which the busiest over 24 hours is market 701501 on Bitcoin dipping to $55,000 at $7,316.29 against $26,073.81 on market 573656, a ratio of 3.56 times. Restricted to the eleven that ask about reaching a price, the busiest is the $100,000 rung at $2,044.52 and the ratio is 12.75 times: https://gamma-api.polymarket.com/events/89502

Kresmion capture history for markets 573656 and 701491, 19 days from July 15 2026 to August 5 2026, daily average implied probability: the gap was 1.30 points on July 15, closed to 0.09 on July 25, inverted to minus 0.19 on August 1, and reached 1.31 on August 4 against 1.21 on August 5, the third widest of the 19 days. Readings from July 15 to July 22 rest on one to seven observations a day against 13 to 34 afterward: internal query on the polymarket_history table

Polymarket gamma API and Kresmion capture history, market 1163699, Clarity Act signed into law in 2026, quoted at 14.5 percent on $673,220 of 24 hour volume at 09:05 UTC on August 5 2026, against $125,659 of 24 hour volume at 07:05 the same morning: https://gamma-api.polymarket.com/markets/1163699

Kalshi trade API, series KXBTCMAXY, the seven contracts closing December 31 2026, fetched 09:05 UTC on August 5 2026: five of seven with no volume in 24 hours, all seven carrying bid and ask sizes and venue quote timestamps of April 9 2026, open interest between 73,042.22 and 329,935.03 contracts, lifetime volume between 265,369.94 and 1,249,237.44 contracts, and mid prices of 4.50 percent on the $140,000 contract against 3.50 percent on the $150,000 contract: https://api.elections.kalshi.com/trade-api/v2/markets?series_ticker=KXBTCMAXY

Kresmion prediction_consensus table read at 08:15 UTC on August 5 2026: 3.94 percent on the $129,999.99 threshold, 3.83 percent on $149,999.99 built from Polymarket 573656 and the Kalshi $149,999.99 contract, 3.53 percent on $139,999.99 built from Polymarket 701492 and its Kalshi counterpart, a separate single venue key at 2.65 percent for the same $150,000 event, and 1,808 of 1,817 rows carrying a venue count of one: internal query

Coinbase spot price API, Bitcoin near $64,100 on the morning of August 5 2026: https://api.coinbase.com/v2/prices/BTC-USD/spot