In the July 2026 cycle, 6 of the top 25 wallets passed the published copyability screen.

“Real edge” is a verdict about the trading, not a promise the trade is still available. Those are two different questions and this page only answers the first. We answered the second separately: on 2026-07-19 we froze all six Real-edge wallets into an append-only forward ledger at the prices a follower could actually have seen quoted — seen, not obtained: we froze the displayed price and never tested whether an order of any size would have filled at it. (Corrected 2026-08-04: this sentence, and the ledger’s own thesis line, previously read “the prices a follower could actually have gotten”. That asserted executability we never tested. The wording was first changed without a note beside it; a published claim does not get to weaken quietly any more than it gets to improve quietly, so the retracted phrase is quoted here..) Four of the six had zero follower-accessible positions at the moment of the freeze — the edge was real and already gone. At the 7-day checkpoint the two live ones stood at $1,475 (+47.5%) and $1,665 (+66.5%) on a simulated $1,000 each, the other four flat, and every copyable position had already resolved by the 24-hour checkpoint — which was itself computed about two hours late, so it bounds the window near 26 hours rather than at one day. So 6 of 25 is the verdict count and 4 of 6 is the accessibility count; they describe different things and neither corrects the other. If you arrived here from a piece that led with the second number, this is where the two meet: the prospective copyability ledger.

The Evidence column is a different question from the verdict. Verdict says what kind of trader a wallet is; evidence says how much we trust the fetch it was computed from. Verified means the fetch is complete relative to the endpoint’s own completeness signals — never absolutely; U1, where every signal read green on a fetch later measured at 1.7% of the at-least-888,820 fills since counted for that wallet’s 30-day window, is the standing proof of the difference (qualified 2026-08-14; this sentence previously read “Verified means the fetch is complete,” unqualified — a form the workpaper and the register already refuse); Provisional means the verdict stands but a known gap (a truncated fill log, an unresolved completeness measurement) means it could move; Insufficient data always applies where the verdict itself is “Insufficient evidence.” A wallet can be genuinely copyable on a Provisional fetch — that is why the two labels are printed separately rather than folded into one word. Full rule and every wallet’s derivation: the evidence-confidence gate →

Why so many “Insufficient evidence.” We only stamp “Real edge” when a wallet’s profit reconstructs from its own fills — and only when the rebuilt figure lands within about ±15% of the displayed profit. A book that over-reconstructs (145%, 164%) fails that test just as one that under-reconstructs does. On the biggest, most-active books our reconstruction can’t yet close the ledger either way — so rather than issue a confident-but-wrong verdict, we say so. That’s the whole product: we would rather refuse than guess. Coverage rises as the engine improves — and every wallet check stays free.

Check our work. The reconstruction workpaper publishes every wallet’s error for this cycle — the 8 we scored, the 15 we refused with their verbatim rationales, the 2 we classified without a usable PnL, and the pooled figure across the full 25. Median gap on the 10 wallets we claim — a set chosen by market and fill count, before any verdict, and not the same set as the 8 we scored above: 5.1 points; our reconstructions sum to 98.7% of the displayed dollars.

What we haven’t proved. On 2026-07-16 we found that our fetch for swisstony (rank 02) had captured 1.7% of the at-least-888,820 fills we have since measured for the 30-day window this page reports on — and 0.5% of its lifetime book — while every completeness signal we own read green. Rebuilt 2026-07-18: swisstony’s fill and market counts above are now the most complete total we have obtained (at least 888,820 fills, 16,258 markets — provenance in u1-swisstony-provenance.json); its reconstruction percentage and mechanism split are not rebuilt and remain unproven, for the reason logged in UNRECONCILED.md U1. Re-fetched 2026-07-18 (U2): the other 24 wallets were measured the same way. Two more (maz26, rank 03, and 0x2c335066FE58fe9237c3, rank 07) were already flagged by our own engine as capped and are now rebuilt the same way as swisstony; a third (0x076daa87, rank 17) had the identical every-signal-read-green failure swisstony did, at a smaller but still severe 5.3× undercount, also rebuilt. Fill counts for 16 of the remaining 21 wallets landed within the ±15% fills tolerance this page promises (market counts were tracked but not gated — several sit wider; see the U2 ledger), and their published fill counts stand as measured. For 5, our 30-day re-fetch counted fewer fills than published — which our own engine’s fetch is not time-scoped to 30 days in the first place, so a shortfall there does not prove either side wrong; those 5 are carried as an open question, not a pass. None of this moved any verdict, the “6 of 25 copyable” count, or the frozen workpaper headline. Full per-wallet ledger, method and provenance in the standing unreconciled-items table — U1 for swisstony, U2 for the rest.

Corrected 2026-07-25. The retraction we published in “A trade-endpoint completeness defect behind apparent phantom positions” obliged us to re-run our own board: 10 of these 25 wallets published a reconstruction below 80%, and the defect we retracted — a fetch that silently omits maker-side fills — was a candidate cause for every one of them. Re-fetching with the corrected pipeline moved 4 wallets, each independently recomputed by a blind second implementation that agreed within 0.6 points: 0xC41D736b… −112.3% → 143.62% (blind 143.22%), maz26 −220.4% → 128.68% (blind 128.68%), Jsram −22.5% → 80.42% (blind 80.89%), and ramadamaramadam 62.9% → 81.54% (blind 81.25%). The negative and near-zero percentages those four rows carried were our data gap, not the wallets’ — the same failure class, on our own page, as the one we retracted publicly. No verdict moves: 143.62% and 128.68% over-reconstruct outside the ±15% band this page publishes as the condition for a stamp, and 80.42% / 81.54% sit under it, so all four remain Insufficient evidence; the 6-of-25 count and the frozen workpaper headline are unchanged. Two wallets (swisstony, 0x076daa87) are reclassified from “low reconstruction” to cannot be measured from the public endpoints — on books this size the API’s enumeration caps make the result run-to-run non-deterministic, which our blind verifier proved by getting two different answers from two clean runs (U5). Two more (pada, FootballFan98) still reconstruct far below their displayed profit after the corrected fetch, and the open question there now points at the displayed side — the evidence, including what may not yet be concluded from it, is logged as U6. Corrected 2026-08-14: this sentence originally named a third wallet, Sparkling8899, among those whose “open question now points at the displayed side.” A window-matched re-measurement (U7, 2026-07-26) moved Sparkling8899 to 85.6% with both of that measurement’s controls passing — the window-matched reconstruction no longer points at the displayed side, and the rest of U6’s evidence remains open. pada and FootballFan98 remain far below on the same re-measurement, whose controls failed for both. The frozen per-wallet source for this correction is scorecard-refetch-2026-07-25.json, and a build gate ties this paragraph to it.

Decision taken 2026-08-02 (U8) — the July 2026 cycle stays frozen on the fetch that produced it. The corrected fetch of 2026-07-25 was applied to the 10 wallets publishing a reconstruction below 80%. It was never run against the frozen headline set, which is a different population selected on structure. The two sets share no member — so all 10 headline wallets had never been re-measured with the method that produced our own retraction. Corrected 2026-08-03: this paragraph first read “only two of its ten members were below 80% — so 8 of the 10”. The populations are disjoint; the number is 10 of 10, and the original understated the exposure. We ran it on 2026-07-26, and we are publishing the result rather than adopting it. Five of the six Real edge wallets stay inside the ±15% band under the corrected fetch. One does not: hot2trot moves from 110.8% to 133.7% — from 10.8 points from parity, inside the band, to 33.7 from parity, which is 18.7 beyond the band’s edge. Two Insufficient evidence wallets move further out on the same measure (filthybetz 85.0% → 145.8%, wco26 84.8% → 134.7%), which does not change their status.

What we are doing about it, and why. The stamps and counts on this page are not being rewritten. A frozen cycle that is silently updated is not a record, and quietly adopting corrected figures would move a published number without an accompanying decision — which our own register lists as explicitly not a way to close this item. So the July 2026 cycle stands as published, this note stands beside it, and the discrepancy stays open in U8 with the full per-wallet table and the dated decision itself. The forward rule is the part that changes: the corrected fetch is the method of record for every subsequent cycle, so no later cohort will carry this defect. A separate correction of the same vintage — our filed CFTC comment describes the 150-market re-fetch as a pooled ranking when it was the 50 largest within each of three wallets — is logged as U9; it changes no figure or finding, and a federal filing cannot be amended, so the register row is the correction of record. Readers weighing the hot2trot stamp specifically should weigh it knowing that under our current method its reconstruction falls outside the band this page publishes as the condition for a stamp.

Label movement, 2026-08-04 (U12). The evidence gate publishes its own rule: a measured, after-the-fact truncation caps a wallet at Provisional, no matter how well its numbers reconcile — and the measurement above is exactly what that rule ingests. hot2trot, muchobliged and weatherman12 moved Verified → Provisional, their selection-time fetches measured short by 22.9, 12.1 and 5.6 points of displayed profit respectively; palegrit and highnetworth are confirmed by the same measurement, reproduced to within a twentieth of a point; DEEDDIT was already Provisional on its flagged offset cap. No verdict, no count and no frozen number moved — the gate defines the label as the part that moves. Until this correction, this page said it in three places — the meta description, the hero and the stat band, the stat band’s form being “one of the six remains provisional on a known fetch gap”; all three now read four of the six. The label correction is logged as U12.