Displayed PnL can't tell you which leaderboard wallets you could copy. The fills can.

Scope note, revised 2026-08-14. An earlier title and thesis on this page asserted more than the evidence below supports — a population claim ("mostly bots") this site's own scorecard does not establish, and the displayed profit figures stated as fact rather than Polymarket's own unreconstructed numbers. The title and thesis above reflect the narrower, supported claim: the per-wallet signature classification, and the finding that displayed PnL alone cannot separate real edge from artifact. Version history →

Every leaderboard and copy-trading tool ranks wallets by profit. But profit doesn't tell a copier the one thing that decides whether they'll make money: is this edge reproducible, or is it a bot doing something you structurally cannot?

Sourced to a committed artifact. The Portugal fill's $125,950 and the USA line's $2,849.38 both tie to on-chain usdcSize — the exact on-chain USDC amount per fill — in debased-receipts.json (both markets re-fetched per-market, paged to a clean under-cap finish, every fill behind all five lines hashed and committed). Every dollar amount in this receipt is checked against its declared rounding model by tools/audit.mjs, which fails the build if one drifts. Correction record: Version history →

What this receipt is sourced to, and what it still isn't. The wallet's full address is 0x24c8cf69a0e0a17eee21f69d29752bfa32e823e1 — independently look-up-able, matched against three files in the decode engine that agree on it. The fills behind all five lines above are the committed, hashed artifact linked in the correction directly above, not a hand-typed 2026-07-11 pull with nothing kept. Unchanged: debased's exact net PnL is still not decision-grade — at 444 markets / 74K fills / heavy merges, reconstruction needs deeper accounting than a five-line receipt attempts — and the strategy classification still rests on the prices and sizes shown, not on these dollar totals. The card’s summary stats above the receipt (74,345 fills · 444 markets · 65% / 67% / $0.78) are not sourced to that artifact: the fills count disagrees with the engine’s own saved output by 757 (ours high, ~1%) — logged as U4 in the same table, magnitude and direction only. The same is true of the card’s ~$503M volume and ~$1.67M displayed-profit figures and Section 01’s $48M buy-volume figure: hand-typed from the 2026-07-11 pull, outside the U4-audited set and tied to nothing in this repo (scope disclosure added 2026-08-14). Logged as RECONCILED in the unreconciled-items table, U3.

We pulled the fill history for each of the next five — 14,000 to 469,000 fills apiece — and ran the same classification debased got. The story holds: the biggest names are market-makers, arbitrageurs, and lottery players whose edge is structure or variance, not calls a retail account could follow.

Why no reconstructed-PnL column here: at hundreds to 10,000+ markets, exact cashflow reconstruction off the public feed is not decision-grade — merges, redemptions, and per-market API limits dominate the cashflow view. That limitation doesn't touch the copy verdict: every signature above is computed from the fill history we pulled and is enough to classify whether a retail user could realistically copy the wallet. Displayed profit is Polymarket's own leaderboard figure, pulled 2026-07-11 — this line said 2026-07-12 until 2026-07-16, contradicting the five other places on this page that date the same pull, including the column header directly above it.

Not everyone is a bot. When we reconstruct a genuinely directional wallet, our engine ties out to Polymarket's own number — proving the method, and letting the difference between "real" and "artifact" show cleanly.

• Theo4 — real edge — the 2024 election whale. $13.8M+ into Trump/Harris presidency & popular-vote markets, held to resolution. Directional conviction. Copyable in principle.
• myzbsq — one lucky bet — reconstructed to 99.6% of its displayed $1.22M. But that $1.22M is a single bet: bought "No Argentina" on one World Cup match at ~$0.15, rode it to $0.99, sold. Not a strategy — a coin flip that landed. Copying "myzbsq +$1.2M" copies the outcome, not an edge.

Your leaderboard ranks these wallets identically. Your users copy the number and lose. The closest anyone has measured this is in crypto, not prediction markets: across 1,486 copy-trading leaders on Binance, Bybit and MEXC, 97.04% were green on their own PnL and only 43.61% produced positive PnL for their followers.¹ Different venue, different instrument — but it is the same gap this page is about, and we have not found a published prediction-market version.

MSR Decode produces a Replicability Score: a per-wallet verdict — real / artifact / farming — with the on-chain evidence behind it. We are evaluating machine-readable access that could sit as a column next to your PnL ranking, so your leaderboard stops misleading and your copiers stop torching money; among the tools we inspected, none ships that combination today. No self-serve product, commercial data license, or redistribution right is offered on this page — see the pilots page.

Want the full decode of your current #1 wallet, free? We'll send the receipts. If it's useful, talk to us about a scoped pilot — machine-readable delivery is under evaluation.

1. Yieldfund, “Is Copy Trading Profitable? A 90-Day Multi-Exchange Study” (Vlad Hatze, published 24 Nov 2025): 100,236 copier outcomes across 1,486 lead traders on Binance, Bybit and MEXC, over a 90-day window to 24 Oct 2025. Both percentages are over all leaders in that sample, which is the study's own denominator — an earlier version of this page said “~56% of even profitable leaders,” a figure the study does not support: it is 100 − 43.61 re-labelled onto a smaller population, and the honest range for that population is 55.1–58.1%, not a point estimate. Corrected 2026-07-16. This is self-reported vendor research and the leader-level data is not published, so we cannot recompute it; we cite it as the closest measurement that exists, over a population that is not ours.

Material revisions to this page, dated. Full detail and the open items they reference: the register.

• 2026-08-14 — headline and thesis narrowed. The original title (“The top of the leaderboard is mostly bots you can’t copy”) and thesis (“Their profits are real”) asserted more than this page measures — a population claim the site’s own scorecard does not establish, and the displayed figures stated as fact rather than Polymarket’s own unreconstructed numbers. Retitled to the per-wallet signature finding the fills actually support.
• 2026-07-16 — receipt arithmetic corrected, Section 02. Two of five dollar amounts in the debased receipt were internally inconsistent with their stated shares × price. 2026-07-18 — sourced to a committed artifact. A per-market re-fetch (debased-receipts.json) found the 2026-07-16 arithmetic-only pass had itself moved both figures away from their true on-chain values; both now tie to the artifact.
• 2026-07-16 — swisstony’s fill count withdrawn, Section 03. The table’s “complete fill history” language was false: a full re-fetch measured 2,918,281 fills against the 14K printed, a difference of more than 200×, caused by an unsubdivided pull against Polymarket’s roughly 3,400-row-per-wallet feed cap. An interim floor this correction first published (“more than 130×”) itself understated the gap and was corrected once the full re-fetch completed. The other four counts on the table were not re-measured and are not restated.
• 2026-07-16 → 2026-08-03 — sample-size figure corrected twice. The note describing how small the 2026-07-11 pull was against swisstony’s lifetime book first borrowed a 2.9% figure that belonged to a different measurement (the scorecard’s 30-day fetch); corrected to this table’s own roughly 0.5%. That other 30-day completeness figure was itself later restated from 2.9% to 1.7% (2026-08-03), which the 2026-07-16 correction had anchored to.
• 2026-08-14 — “verified complete” wording corrected. This page previously said the swisstony re-fetch covered its “whole life … every window verified complete.” The re-fetch is the most complete lifetime /trades total obtained, paged to a clean under-cap finish; it was never independently checked against the separate /activity endpoint, which U1 records can carry fills /trades misses.
• 2026-07-16 — copy-trading citation corrected, Section 05. A stat drawn from the Yieldfund study (“~56% of even profitable leaders”) did not match the study’s own denominator; corrected to the study’s reported 43.61% of all leaders producing positive follower PnL.