Do 2,673 Shared Hands Mean Collusion? A Report Reviewed

Finding: the relevant deals and settlements checked out, and the available betting, folding and chip-contest records do not substantiate this collusion report. On October 31, 2026, a Hold’em player reported a tablemate without a description, named accomplice or hand ID. We reviewed it alongside the same reporter’s cheating report, without assuming the two reported players were partners. We call them A and B: A was reported for cheating, B for collusion. Identities and unshown cards remain private.

Check the hand in progress at filing

The pre-deal hand-ID timestamp and settlement time identify a hand that ended seconds after the report. B bet 100 on the turn, A folded and the reporter called. Both remaining players checked the river; B’s one pair beat the reporter’s high card. B netted 160, the reporter lost 120 and A lost 20. This hand does not show A and B taking turns raising to squeeze the reporter. The deal, showdown ranking and settlement all checked out. Because the report named no hand, this remains a time-based identification.

The preceding hands were checked too

In the latest completed hand before the collusion report, B called 20, A then moved all-in preflop, and everyone else folded. B and the reporter each lost 20. A netted 70; the unmatched portion of the shove was refunded. B never raised, so this hand does not show A and B taking turns raising to squeeze the reporter. In the preceding hand, both reported players lost 450 and a third player won the pot. Both hands passed deal, pot-winner and settlement checks. They are the nearby hands located by time, not hands explicitly identified by the reporter.

Do 2,673 shared hands establish collusion?

Through 14:37 Singapore time that day, we checked all 3,865 saved hands for A and all 3,056 for B. They shared 2,673 hands across 47 sessions, giving 4,248 distinct hands after deduplication. All deal-order, saved-card, showdown-winner and chip-settlement checks passed. Frequent shared play establishes that they often sit together; it does not establish card sharing or soft play.

To look for unusual concessions to a particular opponent, our action screen used hands with at least three human players, including 2,554 shared hands. We compared each player’s decisions against other opponents at the same betting street and call-to-pot pressure band, counting each directed opponent relationship at most once per hand. A had 827 matched opportunities against B and folded 51.8%, versus a matched other-opponent baseline of 50.1%. B had 451 against A and folded 53.9%, versus 53.3%. Neither showed unusually frequent folding to the other. We also screened their records against other frequent tablemates without finding the marked excess-fold patterns defined for this review. This compares saved actions; it does not establish intent.

The ace-king hand in the first report was itself an A–B contest: B called A’s shove and lost 10,000. Across 48 hands ending heads-up with equal contributions, no larger third-party contribution and at least one entire stack committed, A won 27, B won 19 and two tied. Counting only their equal contributions to each other, A won 345,592 from B and B won 218,161 from A: a net 127,431 toward A. These are chip-flow descriptions; wins in both directions cannot rule out deliberate dumping.

We therefore calculated equity when chips went in, rather than relying on results. We used the actual hole cards in hindsight but only the board visible at each decision, enumerating turn and river runouts and sampling 20,000 runouts per earlier-street decision. B’s final contribution was a call in 20 of these contests; 11 had under 35% actual pot equity. Against other opponents the corresponding count was 24 of 70. B did make more weak-equity calls against A, which deserves attention. After matching street and the pot-equity requirement of the call (below 25% or at least 25%), B had 5/20 calls where actual equity fell more than 15 percentage points below the requirement (25%), versus a weighted other-opponent baseline of about 25.2%. A’s corresponding figures were 5/28 (17.9%) versus about 20.0%.

Risky calls occurred, but the current evidence is insufficient to label them targeted chip dumping. B had only seven final preflop calls against A in this sample. Moreover, hindsight equity against actual cards is not the player’s estimate against an opponent’s range. These comparisons do not examine off-table agreements and cannot conclusively exclude collusion.

Deal verification and collusion review answer different questions

Zero-knowledge proof verification is already live for default Hold’em. Sign in with the account that played, open the hand in History and choose Verify fairness, or select it on the Fairness Verifier. The proof checks the complete deck, the order determined by player randomness and decryption of visible cards. The playing device can also compare the proof with its record from the hand. See the zero-knowledge proof guide. The nearby hands support the newer protocol; older hands were checked under their original protocol and are not all v2 proof verifications.

A zero-knowledge proof does not check whether people shared cards through a messaging app. These historical actions also lack usable per-decision durations, so we reached no conclusion about click-timing correlation. Our finding is that the checked deals and settlements are correct, with insufficient behavioral evidence to establish collusion, not that off-table collusion is impossible. A specific hand, coordinated action or chat evidence can support further review. Frequent shared play or winning more chips alone should not trigger punishment.