FILE 00 Wallet intelligence · Forensic division
Before you copy a prediction-market wallet, find out whether the edge is real. We reconstruct the trading history — completeness-tested, fill by fill — and separate repeatable directional edge from leaderboard artifacts, market-making, and reward farming — free. Only 6 of 25 top wallets passed.
Completeness-tested fetch Per-fill reconstruction 99.6% match to displayed PnL
Profit source
Give us the wallet. We’ll tell you — with evidence — whether the edge is real and copyable before you put a dollar behind it. No account, no fee.
See a real verdict: DEEDDIT · Real edge myzbsq · One-shot cnyek · Insufficient evidence
The verdict vocabulary is the product. Tools that rank wallets by PnL are everywhere — among those we inspected, none combines fill-by-fill reconciliation with a follower-fillable verdict while staying independent of the platforms. A score is not an explanation: our file names the mechanism and shows the evidence. A copyability rating is to a wallet what a credit rating is to a bond — what kind of result you’re looking at, not whether to buy.
Directional positions taken before the market moved, sized with intent, profitable after honest cost-basis reconstruction. The rare wallet worth studying.
Forensic tellEntries lead price; PnL survives per-fill reconstruction; no structural crutch.
Redemption arbitrage, cluster/wash behavior. The PnL is real — the edge is structural, and a retail copier following the same entries will lose money.
Forensic tellBuys near $0.99 pre-resolution; paired wallets; profit vanishes at copier timing.
PnL manufactured from liquidity rewards, rebates, and points — volume played for the subsidy, not the outcome. Copy the trades without the rewards and the returns evaporate.
Forensic tellTwo-sided volume at flat prices; PnL tracks incentive epochs, not events.
Two-sided quoting and inventory management. Displayed PnL reflects thousands of small spreads — a flow business, not a forecast you can follow.
Forensic tellBalanced buys and sells at adjacent prices; inventory mean-reverts; no directional lean survives netting.
Most of the profit traces to a single concentrated win. It proves variance, not an edge — there is no process to copy.
Forensic telln = 1: profit concentrated in one market and outcome; little history before or after.
Fills that won’t reconcile against the displayed profit — often because the history we can fetch is incomplete. Rather than issue a confident-but-wrong stamp, the file states exactly what’s missing.
Forensic tellFills don’t reconcile with the displayed figure — often where the fetch gate caps out. Refused rather than guessed.
PnL tells you who made money. MSR Decode tells you how, and whether you could have copied it. Polymarket’s 25 top wallets by 30-day profit, each run through the same forensic decode — and only 6 of 25 copyable. Three contrasting examples below; the full board is on the Scorecard. See the method → · the reconstruction workpaper →
The nearest published measurement — crypto copy-trading, not prediction markets — found 97% of “leader” wallets green on their own PnL while only ~44% produced positive PnL for the people copying them; we have not found a published Polymarket equivalent, which is a good part of why this site exists.
Source: Yieldfund, “Is Copy Trading Profitable? A 90-Day Multi-Exchange Study” (Vlad Hatze, 24 Nov 2025): 1,486 lead traders and 100,236 copier outcomes across Binance, Bybit and MEXC, collected over 90 days through 24 Oct 2025. Both figures cover all leaders in that sample. This is self-reported vendor research; its leader-level data is unpublished, so we cannot recompute it. It measures crypto copy-trading, not prediction markets — the closest measurement that exists, not ours.
We now test copyability forward, in public: every “Real edge” verdict gets a frozen, follower-accessible price snapshot and a checkpoint published at 24h, 7d and 30d — gains and losses alike. Day one already shows the gap: 4 of 6 July Real-edge wallets had no follower-accessible positions left the moment we froze them. See the Prospective Copyability Ledger → · Cite this data — or ask about dataset access →
See the full July Scorecard — 25 wallets, 6 passed.
Every wallet, its verdict, and the verbatim rationale — filterable, sourced, and refreshed monthly.
Get notified when a verdict flips.
A wallet can move in or out of “copyable” as more of its book becomes reconstructable. We email you when one does — nothing else, no paid tier. Privacy.
A verdict is only as good as the evidence chain behind it. Ours starts with the records we can verify — and tests what may be missing — before it ends in a named strategy class. No vibes, no silent sampling.
The six-way verdict classifies the trader; this separate label grades the fetch. A genuinely copyable wallet can still rest on an incomplete fetch. Collapsing both judgments into one word (“verified”) is dishonest.
We pull the wallet’s fill log from a capped public feed, forcing completeness window by window, never assuming it. Partial data produces confident, wrong verdicts: on 2026-07-16 — after publishing figures from it — we found one wallet whose fetch held 2.9% of its 30-day book (0.5% of its lifetime book) while every completeness check we own read green. If a short fetch still rebuilds profit, we stamp it and print the truncation beside the stamp; if not, we refuse.
On 2026-07-18 we tested the cycle’s other 24 wallets against 30-day re-fetch floors: 16 proved consistent with their published counts, 3 proved incomplete and had their counts rebuilt, and 5 remain inconclusive. Because our engine’s fetch is not time-scoped, a smaller 30-day floor proves nothing either way. See U2 in the unreconciled-items table — not papered over here.
We rebuild the wallet’s real PnL fill by fill and set it against the number the leaderboard displays. The gap between displayed and real is where artifacts hide.
Behavior is decoded into a labeled taxonomy — redemption arbitrage, reward-farming, cluster/wash, directional-real, market-making — and the verdict follows from the class, stated in plain English.
Two published decodes — one leaderboard darling taken apart, and a full Decode report exactly as the engine produced it.
Case file 01Artifact — not replicable
How a top wallet’s celebrated “edge” decomposed, fill by fill, into redemption arbitrage — profit a copier could never have captured.
Read the case →Case file 02One-shot — not an edge
A full Decode report. A $249K → $1.22M World Cup bet that looks like genius on the leaderboard — and decodes as a one-shot lottery ticket, not a repeatable edge.
Read the report →That the mechanism behind the profits — not the profits themselves — could realistically be executed by a follower without the original wallet’s structural advantages: timing, rebates, paired flow, or settlement position.
We fetch the wallet's fill history from the public feed, rebuild cost basis per fill, decompose PnL by market and mechanism, and classify the strategy. Where that fetch comes back short, we say so on the wallet's own row rather than treat what we got as the whole book. The rating states the class, the confidence, and the evidence behind both.
Public on-chain and exchange-published trade data only. No private access, no self-reported numbers.
Leaderboards net things a copier can’t ignore: rewards, rebates, paired entries, and settlement flows. The gap between the displayed figure and the rebuilt one is usually where the artifact hides.
The file says so — on the wallet’s own row, not in a footnote. If we can’t obtain enough history to reconcile the book against the displayed profit, we refuse to classify — and say so on the wallet’s own row. Where a fetch is truncated but the profit still reconstructs from the fills we do have, we stamp it and print the truncation beside the stamp. From the next cycle those are excluded from our accuracy set outright.
No. We publish research on the historical behavior of wallets. Ratings describe the replicability of past mechanisms; they are not individualized trading recommendations.