How do you verify a trading track record?

The short answer

A verified quant track record is one a third party has recomputed, not one a platform has badged. Apply the recompute test: a track record is verifiable exactly when a third party could recompute it — which requires the trade list it was computed from, the accounting rules used to compute it, and enough provenance to pin both to a real account. That is the whole standard. A record that ships those three things can be checked in a spreadsheet; a record that ships a chart and an assertion cannot be checked at all, and the polite word for an uncheckable claim is a story.

The practical corollary: ask for the three things. The reaction is itself a measurement. A seller with a real record can produce a trade list in minutes; the ways of not producing one are where most of these conversations honestly end.

WHAT THIS IS — AND WHAT IS NOT PUBLISHED. This article is method: what makes a record checkable and how far each form of evidence reaches. No seller, platform or record is measured, named or ranked here, and nothing should be inferred about any. Where a specific claimed record needs recomputing, that is a measurement — see the close of this piece. Status of any measured claim: NOT YET PUBLISHED.

Prerequisite Knowledge

You need whatever the seller has actually provided — a screenshot, a stats-page link, a statement export — and a clear note of the headline claim being made from it (the return figure, the drawdown figure, the win rate). The method below tells you what each form of evidence can support; the gap between that and the claim is the finding.

What counts as a verified quant track record

Two different things get called verified. A stats platform’s badge verifies a data connection: from its start date onward, the numbers shown came from an account the platform could read. A verified record in the sense this site uses the word is one that someone other than its owner has recomputed from the trade list, the accounting rules and the provenance behind it — and that reproduced, or diverged by an amount the report states. The first is a fact about plumbing; the second is a fact about the numbers. Whether the record is a quant’s, a signal service’s or a prop-firm evaluee’s, the standard is the same, because the test does not depend on what produced the trades.

The ladder of evidence

Each rung establishes strictly more than the one below it — and the top rung is the only one that is verification rather than trust.

1. A screenshot

Establishes nothing. A screenshot is a picture of a screen, and screens render whatever they are asked to render — including demo accounts, edited pages and image files. Treat it as an invitation to climb, never as evidence.

2. A stats-platform page

Establishes a data connection, with the limits the first debunked claim above sets out: real link, unknown completeness. Useful checks while you are there: the account’s start date against the story told about it, gaps in the history, and whether the platform distinguishes real from demo accounts at all.

3. A statement export

Establishes what the record claims in detail — which finally makes it checkable for internal consistency, and editable by anyone with a text editor. Paradoxically this is progress: a detailed fabrication has many more places to be wrong than a chart. Balance arithmetic that does not tie from row to row, position sizes impossible at the stated leverage, timestamps out of order or dense inside market closures — a real export survives these checks without effort; an edited one frequently does not.

4. Read-only access, or a statement you pull yourself

Establishes the account’s history as the venue reports it — the first rung where a third party, rather than the seller, is the source. What it still cannot establish: whether this is the only account, and whether the record’s shape is selection rather than skill.

5. Recomputation from inputs

The top rung, and the only one that is verification in the strict sense: take the trade list, apply the stated accounting rules, and reproduce the headline numbers independently. Three outcomes are possible — it reproduces, it diverges by a quantifiable amount, or it cannot be recomputed from what was provided. All three are findings. The third is the most common and the most decisive, because a claim that cannot be recomputed from its own materials was never a measurement to begin with.

What even a real record cannot tell you

Suppose everything checks: the connection is live, the statement ties, the numbers reproduce. Two questions remain that no single record can answer. The denominator: how many accounts or strategy versions were started for this one to be the survivor shown — the same selection problem that inflates backtests, operating on live records. And the sizing shape: whether the calm was produced by a style whose losses are rare, large and still ahead. The risk-of-ruin mathematics and the survivorship-bias entry carry the details; pre-registration is the discipline that answers the denominator question by construction, which is why this site’s own measurements are run under it.

The Observable Mechanism

Every check in the ladder up to internal consistency is computable by you, today, from the materials a seller either provides or refuses to — no platform access, no permission, one spreadsheet. The refusal itself is observable. Full recomputation needs only what the record was allegedly computed from, which is precisely why asking for it is such an efficient filter.

What This Does Not Establish (The Limits)

The ladder above establishes what each form of evidence can support — not that any particular record is false, and not that a record which passes every rung predicts anything. A fully verified history is a fact about the past; the leap from verified past to expected future runs through the selection and sizing questions above, and no verification removes them. It also does not establish intent: a record that fails to recompute was not necessarily faked — sloppy accounting produces the same finding, and the report that says “cannot be recomputed” says exactly that, no more.

Where this leads

The recompute test is machinery this site runs as its own discipline: the Reproducible Verdict Kernel exists so that every measurement published here can be recomputed by a competent third party from hashed inputs and a stated method. And when the record in question is one someone is selling you — a signal service, a managed account, a strategy — the Verdict Recompute assay runs the test as a one-off: you send the claimed result, its supposed inputs and the stated method, and the report returns one of the three verdicts with the working shown. The third verdict is the one vendors would least like to exist — and the one a claim that cannot produce its own inputs earns.