The withdrawal was public. Two-thirds of the money traded after it.

Matteo Arnaldi was due to play Flavio Cobolli in the Roland Garros men’s singles semifinal on June 5, 2026 — his first chance at a Grand Slam final. He withdrew before the match started, citing a viral illness. Cobolli advanced without playing and went on to face Alexander Zverev in the final.

Polymarket US listed a moneyline market on the match: one side pays $1 if Arnaldi wins, the other pays $1 if Cobolli wins. The exchange’s own record for the contract labels it moneyline and carries the symbol aec-atp-matarn-flacob-2026-06-05.

Three times matter, and only one of them is machine-readable. A Tennis Majors report on the withdrawal carries a datePublished of 16:39:57 UTC, verified in the raw HTML. That report says Roland Garros announced the news to spectators and media at 6:35 p.m. local time, which is 16:35 UTC; we found no official Roland Garros or ATP record of that time, so we do not count from it. The market’s last trade printed at 17:36:53 UTC.

We count from 16:40:27 UTC — the report’s own timestamp plus thirty seconds, to allow for publishing delay. That anchor counts fewer trades as post-public than either of the other two, which is why we use it.

Why the anchor is the report, not the announcement. The 16:35 announcement reaches us only as a sentence inside the report — a narrative claim, not a timestamp we can verify independently. The report’s publication time is a machine-readable field on a document that exists. Counting from the later, weaker-sounding anchor gives up 32 trades and makes every share on this page smaller.

Every trade on Polymarket US is recorded in a public daily log with a timestamp, a price and a size — the exchange calls it the Time & Sales report. We used that log directly rather than a chart drawn from it, and reconstructed this contract’s entire life: 1,035 trades across three daily files, from 2026-06-03 22:10:34 UTC to 2026-06-05 17:36:53 UTC.

In the 56 minutes from 16:40:27 UTC, 523 trades moved 296,197 contracts worth $102,735.48. That is 66.7% of every dollar this contract ever traded and 64.9% of every contract. Measured by number of trades rather than by size it is 50.5% — just over half.

The lifetime row reconciles to the cent against the totals Polymarket US published for this market. That reconciliation is what makes the split above a measurement rather than a sample — a tape missing rows would still divide into two parts and still look tidy.

The price rose after the withdrawal became public. Over the following half hour it moved from around 32¢ up to 36¢, and in the final ten minutes it fell back. We report the sequence and draw no conclusion from it about why anyone was trading.

Polymarket US publishes settlement terms for every contract it lists. This contract’s own terms end with the sentence that governed this settlement:

The phrase last-traded prices is not defined anywhere in Polymarket US’s public rules. It does not say which trade counts. It does not say whether there is a cutoff time. It does not say whether trades placed after a withdrawal becomes public count at all.

The contract settled Arnaldi’s side at 33¢ and Cobolli’s at 67¢.

Polymarket US’s Sports FAQ, in the version live before this match, says markets may remain open for a period after an announcement and that trades in that window are made at the trader’s own risk. The exchange did disclose, in general terms, that trading could continue after a withdrawal. It did not disclose for how long, or how the final price would then be set.

Which document governs is not ambiguous. Polymarket US’s rulebook says a contract’s own terms control where they conflict with the general rules, and its terms of use say the formal documents supersede other representations. We therefore treat the sentence above as the rule that governed this settlement. A separate case file sets out where that sentence and Polymarket US’s own FAQ state different rules for the identical scenario.

The literal final print of the day was 32¢ — not the settlement. But several plain readings of the closing trades produce 33¢ exactly. We tested 11 and publish all 11, including the ones that miss.

Eight of the 11 readings land on the settlement. Every one of them draws only on trades printed after the withdrawal was already public — not one of these windows reaches back across the counting anchor.

This is not a reconstruction of Polymarket’s method, and we do not present it as one. Several readings landing on the same cent is weak evidence for any particular one of them: prices in the closing minutes clustered tightly, so a range of windows was always going to round to the same place. What it establishes is narrower and firmer — the settlement is consistent with an average of the closing trades, and the closing trades happened after the news.

An earlier, unpublished draft of this analysis concluded that no reading reproduced 33¢. That was wrong. The test behind it compared floating-point values exactly and never rounded to the cent, which is how the venue itself quotes prices. The conclusion is corrected here rather than quietly dropped.

The clause in Section 03 was not drafted for this match. Polymarket US created 34 ATP and WTA moneyline contracts on 2026-06-03, the day it created this one. We read the settlement terms of all 34, and every one of them carries the identical walkover clause, word for word.

Of those 34 contracts, 32 settled at $0 or $1, as a completed match does. Two did not, and both were walkovers. One is Arnaldi’s. The other is a second-round match at the Birmingham Challenger on June 4 between Otto Virtanen and Aleksandar Vukic.

That second contract is the only independent test available of how this clause behaves. It settled Virtanen’s side at 71¢. Its literal last print was 77¢. It was a small market: 33 trades, 1,977 contracts, $1,324.22 across its whole life.

So we swept every trade-count window both tapes can support — 33 widths, under a simple average and a volume-weighted one — and asked which reproduce both settlements.

A consistent reading exists, and it is narrow: the average of the last 14 or 15 trades reproduces both settlements, under either method of averaging. One trade wider or narrower and it fits one contract or neither. None of the 12 time-based windows we tested fits both, because Virtanen–Vukic traded so thinly that every window shorter than a day holds only 2 of its 33 trades.

Two settlements and a two-window band is not a method, and we are not presenting it as one. We are not saying Polymarket US averages the last 14 or 15 trades. With two data points, a band this narrow is as consistent with coincidence as with a rule, and we have no third walkover to test it against. What the sweep does settle is the weaker claim in the other direction: it is not true that no consistent reading exists.

An earlier, unpublished draft of this work asserted exactly that — that no single window fits both contracts. It was wrong, and it was wrong the same way Section 04’s corrected claim was wrong: the window that fits was tested on one contract and never on the other. The sweep above is symmetric by construction, which is the only reason we trust it.

One thing the band does establish for this page. On the Arnaldi contract, all 15 of those trades printed after the withdrawal was already public — the earliest of them at 17:30:28 UTC, nearly an hour past the counting anchor. Whichever reading Polymarket US actually used, every window that reproduces this settlement lies entirely inside the post-announcement window.

• We are not saying the 33¢ settlement was calculated incorrectly. We cannot tell from the public record which trades Polymarket US used.
• We are not saying trades placed after the withdrawal became public determined the settlement, or that the settlement was fixed before those trades happened. Neither can be established from the public record.
• We are not saying any trader knew the eventual settlement price while trading, or that anyone made or lost money because of how this market was priced.
• We are not alleging manipulation, exploitation, or wrongdoing of any kind.
• We are not claiming that Polymarket’s Contract Outcome Review Process, a discretionary process described in its own rulebook, was used for this market. We searched the exchange’s notice archive and found no record either way.
• We are not claiming a regulatory violation.

The questions this leaves are for Polymarket US, and they are narrow:

• How was the 33¢ settlement calculated, and which trades went into it?
• What does “last-traded prices” mean in this contract and in the other 33 that use the same wording?
• Was there a defined cutoff time for the trades used to determine settlement?
• The Sports FAQ says markets may remain open after an announcement. Who decides for how long, and why did this one stay open for 56 minutes?
• How was the 71¢ Virtanen–Vukic settlement calculated, given its final print of 77¢?
• Has the “last-traded prices” language been clarified or replaced?

Source. Polymarket US publishes one Time & Sales file per calendar day, covering every contract on the exchange, with a transaction time, symbol, price and quantity for each execution. We pulled the files covering this contract’s life and filtered them to its symbol, aec-atp-matarn-flacob-2026-06-05. The tape published alongside this page is that filtered extract, unmodified apart from a header row.

Completeness. The reconstruction reconciles to the cent against the lifetime totals Polymarket US published for this market — 456,605 contracts and $154,002.74. An earlier pass was short 66 early rows and did not reconcile; we pulled the prior day’s file and it did. We treat a reconciliation as the precondition for reporting any subtotal, not as a bonus check.

Arithmetic. Every price in this tape is exactly two decimals and every quantity is a whole number of contracts, so every figure on this page is computed in integer cents. Nothing here passes through a floating-point number, and the builder halts if a row violates either assumption.

Anchor. 2026-06-05 16:39:57 UTC is the earliest moment at which we can independently confirm the withdrawal was public, and it is the datePublished of a document that exists. We add thirty seconds. Earlier anchors would raise every share on this page; we did not use them.

Population. The 34-contract batch is enumerated from Polymarket US’s own market catalogue by creation date, not assembled from a hand-picked list of contract ids. An earlier draft of this work claimed a different population and a different count, and was wrong; that is why the enumeration ships as a file rather than as a sentence.

The window sweep. Section 05 sweeps every trade-count width both tapes can support, under both a simple and a volume-weighted average, and reports every width that fits both settlements together with its immediate neighbours. It is symmetric by construction: no width is tested on one contract and not the other. That asymmetry is exactly how the earlier draft reached the opposite conclusion.

Limits. Two walkover settlements is not a census of how this clause behaves, and a band that fits two data points is not a method. This page tests a contract’s wording against its own tape; it does not establish what Polymarket US intended, what it computed, or what any trader knew. Where the public record does not answer a question, this page says so rather than inferring.