Method

Principles

Resolution data over screenshots. Every audit re-derives the numbers from resolved market outcomes — settlement records, public trade feeds, order-book history — never from a dashboard image supplied by the person making the claim.

Executable prices only. A strategy is scored at the price you could actually have traded — the live ask you would have paid, the live bid you would have received — not at midpoints, not at the price after the move. Most published "edges" die at this step alone.

Fees and spread included. Taker fees (which peak near 50¢ and differ by market class) and spread-crossing costs are charged before any ROI is reported.

The luck test. A profitable record is only evidence of skill if it beats the prices it paid by more than chance allows. We compute the expected win count implied by entry prices and report how many standard errors the actual record sits above it. Under ~2σ, the honest verdict is that luck cannot be excluded — especially when the account was selected off a leaderboard of thousands.

One position per market. A signal that fires several times on the same market — a whale's order printing as several fills, or the same favorite re-qualifying every scan — counts once. A trader holding one position per market could not have collected the repeats, and counting them lets a few lucky markets pass as a pattern.

Selection and survivorship. Who chose this account to show you, and out of how many? A 1-in-40 record is remarkable in isolation and guaranteed in a pool of 9,000.

Conflict of interest. If the post funnels to a product — a platform, a copytrading button, a paid group — that is disclosed in the verdict.

Verdicts

Confirmed — the claim replicates in resolution data at executable prices. Plausible — consistent with the data; sample or price evidence still insufficient. Unproven — the record cannot be distinguished from luck or selection. Debunked — the claim fails when re-derived. Advertisement — the post is structured as a promotion; treat all numbers as marketing.

Whale files use their own scale: Sharp, Lucky so far, Market rate, Bleeding — same tests, applied to a wallet's resolved record.

Data

Audits draw on a continuously-collected database of resolved prediction-market outcomes: hundreds of thousands of large resolved trades, order-book snapshots at signal time, and per-market resolution records, cross-checked against public settlement feeds. Where a claim concerns a platform we do not index, we say so and score only what public data supports.

Revisions

Verdicts are not permanent. When later results contradict an entry, we change the numbers and the stamp, say what changed in the text, and show an "updated" date next to the original one. Lab entries in particular are revised as experiments reach their sample gates — and a result that looked strong early is often the one that changes.

How these pages are made

Drafts are written with an AI model (Anthropic's Claude), working from figures that our own scripts pull from the trading database described above; every number on a page comes from one of those queries, not from the model's memory. Nothing is published automatically: a person reviews each draft and publishes it by hand. No lawyer or licensed financial adviser reviews these pages, and nothing here is financial advice.