The BNB 5-minute wallet: reconstructed to the cent.

On July 24, 2026, a post from @DextersSolab put this wallet in front of a wide audience: a claimed ~$917k profit in about three hours trading Polymarket’s five-minute “BNB Up or Down” quick markets, on a claimed 23-for-24 record. We pulled the wallet’s complete on-chain activity and reconstructed every fill, redemption and merge. In this dataset, one of those two numbers holds almost exactly. The other does not.

In this dataset, the wallet placed 139 buy fills and zero sell fills across 24 of 39 consecutive five-minute “BNB Up or Down” quick markets, in a single 3.1-hour session (09:12–12:19 UTC, Jul 17 2026). That session is the wallet’s entire recorded trading history on Polymarket. It has not traded again since, for reasons this dataset does not tell us.

Net realized excluding the rebate leg is $917,311.53 — the figure that ties to Claim 1 above. The rebates are a separate, later settlement (Polymarket’s own reward/rebate schedule), not proceeds from any individual fill, which is why the ledger breaks them out as their own line rather than folding them into the 139 buys or 22 payout events.

What the fills show is not one mechanism but at least four distinct ones, stacked in a single session.

A non-material pricing discrepancy affecting neither this finding nor any published total is logged in Technical notes (U14) →.

All 24 markets the wallet traded, in order. “Cash out” combines redemptions and the one merge; a dash means the losing position expired worthless and nothing was collected. This is the complete population — nothing here is a selected subset.

The Result column below and the machine-readable ledger disagree on four rows, with no dollar figure in dispute — see Technical notes (U13) →.

Full machine-readable ledger, the same 24 rows shown here: bnb-5min-ledger.json. This table is presented for per-market detail and is not itself the summed reconciliation — the tied totals are in Section 02 and Section 05.

$40,894.05 of on-chain token cost across these same three markets expired worthless. $1,626.55 of that total is the losing side of the hedge market above, and because it was never formally redeemed, it does not appear anywhere in the wallet’s own open-position feed as a live balance — unlike the other two markets’ worthless cost, which does. Both figures are components of the one $40,894.05 total, not an addition on top of it.

• Manipulation: neither evidenced nor excluded. The decisive price moves behind the late-window fills had a median of roughly 5.13 basis points in this dataset; ordinary five-minute moves in the same span had a median of roughly 5.15 basis points — the two are statistically indistinguishable. We have no visibility into the wallet’s trading activity off Polymarket, so manipulation can be neither confirmed nor ruled out from this data; the size of the moves needed to explain the profits does not require it.
• Whether the pricing quirk still exists: not tested. This report covers a single historical session and does not re-check current quick-market pricing behavior.
• Why the wallet stopped after 3.1 hours: unknown. The wallet has not traded again as of this report’s publication.

Two open discrepancies between this page and the published ledger. Neither affects the dollar reconciliation, the 21-of-24 result, or the verdict.

Open discrepancy, added 2026-08-13 — the “only exception” sentence in Section 03 (U14). The first card in Section 03 says the single non-round print across all 139 fills is $0.789622964. Market 23 in the published ledger (8:10AM–8:15AM ET) carries one fill, $40,062.5631 of cost and 50,000 units redeemed, which implies $0.801251262 — also non-round. We are not substituting one figure for the other: at least four readings survive the published records, including a per-fill fee of about 1.47%, under which both figures are right, and units bought differing from units redeemed. The page does not say which market its print belongs to, so even attributing it to market 23 is an inference. Which claim fails, and how, is not determinable from what we have published. No published total moves either way. Full entry: search the open-discrepancy register for U14 (that file carries no internal anchors by design).

Open discrepancy, added 2026-08-13 — the Result column in Section 04 (U13). Section 04’s table and the machine-readable ledger it links publish the same 24 markets, and their characterisation fields disagree on four rows: 08, 12, 15 and 17. On two of them the JSON’s own field disagrees with the sign of its own net P&L — market 12 is labelled WIN beside −$1,606.64, market 15 is labelled OPEN(loss-so-far,$0) beside +$26,069.03. We are not offering an explanation for that here; the register records what was tested and what remains open. No dollar figure is in dispute: every cost, cash-out and net P&L column ties between the two artifacts, and the 21-of-24 result is unaffected. The file has no generator in our repository, so the only legitimate repair is re-deriving it from the raw fills — correcting the field in place, by hand, is the silent-fix class our own gates exist to prevent, so we will not do it. Until that is done, read the Result column in Section 04 and the result field there as two different statements. Full entry: search the open-discrepancy register for U13.