Quantitative Verification & Audit

Quantitative work fails quietly. A result survives review not because it is correct but because the specific way it is wrong was never tested for. By the time capital or a decision is committed to it, the error is expensive and the reasoning is months cold.

We conduct independent audits of quantitative research and the pipelines feeding it. The engagement is adversarial by design: the objective is to find the reason a result should not be believed, and to say so clearly when no such reason exists.

What we examine

Look-ahead and leakage — revisions that were not available on the day, standardisation fitted across the evaluation boundary, and panels that cannot survive reconstruction from a truncated history.

Execution realism — whether assumed fills survive queue position, spread, latency and tick constraints, and whether a statistically real signal is still there after the cost of acting on it.

Specification errors — unit and time-step inconsistencies, transforms applied to the wrong quantity, moment conditions that do not match the assets they are computed over. In our experience these account for more implausible results than genuine methodological disagreement.

Test coverage — specifically whether the suite is capable of catching the defects that matter, or only exercises the paths that already work.

What you receive

A written report with findings graded by severity, each traced to a specific line or step, distinguishing errors that invalidate a headline result from those that change the number without changing the conclusion. Where we can identify the fix, we specify it.

We hold our own research to this standard and publish the audits, including the ones that invalidated our own findings.

Suited to firms allocating against internal research, and to teams that want a result stress-tested before it goes in front of a committee.