Quantile Labs

Independence

Independence is not a value statement. It is a set of structures that make compromise difficult. This page sets out the entity separation, the publication rule, how conflicts are handled, and the work we refuse.

01 · Two entities

The research does not take money from the systems it evaluates.

The laboratory is two organisations, and the funding follows the function.

What each entity does, and who may pay it
Quantile LabsQuantile Assurance
Function Research. Publishes findings, notes, methods, and reviews. Commercial. Sells audits, training, and technical services.
Funded by Philanthropy, unrestricted grants, and research funding. Revenue, which cross-subsidises the research entity.
Constraint Accepts no money from any operator whose systems it has evaluated or intends to evaluate. Its clients are permanently ineligible as evaluation subjects. Not for a cooling-off period. Permanently.

Separate boards, separate accounts, and a documented firewall between them. Which entity an engagement belongs to is recorded at scoping and cannot be changed once work has started. Staff do not move between an Assurance engagement and an evaluation of the same organisation.

We take no equity, revenue share, or contingent fees in any form. No part of any fee depends on what an evaluation concludes.

The rule this exists to enforce

No operator funds its own evaluation. Ever. There is no tier of disclosure, no firewall and no publication commitment that makes subject-funded evaluation acceptable, so we do not do it, and the structure above is what stops it happening by accident.

02 · The publication rule

Findings are published whatever they conclude.

The rule appears in every engagement contract before signature. A party that will not accept it is not a client, and we would rather establish that in the first conversation than the last.

Confidentiality over commercial data, yes. Over findings, never. No funding or engagement agreement may include review rights over a finding: the clause does not exist in our contracts, rather than existing and being resisted.

The subject sees the complete draft, not a summary, and has a minimum of twenty-one days to reply, longer where a finding implicates patient or financial safety. Where a finding implies immediate safety risk we notify before the public timeline and allow a remediation window. The reply is published in full alongside the finding, unedited.

The right of reply is not a right of veto. A developer who fixes a failure we documented is the fastest route to reduced harm available to this laboratory, and we would rather invite that response than treat it as a defeat.

03 · Conflicts of interest

Declared before the study, published with the finding.

Before a study is accepted, every person who will work on it declares financial interests, prior employment, consulting relationships, and personal connections to the subject, its vendors, and its regulator. The declaration is repeated at publication, because relationships change during a study.

The conflicts register is public and updated per study at selection. Declared interests are published in the finding whether or not we judged them material, because a reader is better placed than we are to decide what colours a result.

Anyone with a material interest is removed from the study. Where a conflict cannot be contained by removing an individual, we decline the work and say so.

04 · What we will not do

Engagements we decline.

  • Any evaluation funded by the operator of the system under evaluation.
  • Any evaluation of a system operated by a Quantile Assurance client.
  • Work where the subject holds the right to prevent publication.
  • Work priced on a contingent basis, or paid in equity or revenue share.
  • Certification, accreditation, or any output presented as a mark of approval. We are not an accreditation body and hold no accreditation ourselves.
  • Evaluations scoped so narrowly that the result would be technically true and substantively misleading.
  • Work where the test set cannot be constructed to resolve the populations the harm model identifies, and the client will not fund the collection needed to build one.
  • Red-teaming or adversarial testing whose purpose is to produce a marketing claim.
Limits of this claim

We are working toward formal accreditation against an international standard for testing laboratories. We do not hold it, we are not certified, and nothing we publish should be read as carrying accredited status. When that changes, the standard and the accrediting body will be named here.