05 · Endorsement

Reviewed at the bar of national labs.

Endorsements are not a marketing surface — they are a record of what reviewers, area chairs, and replicating teams have said about the lab when the artefacts were on the table. We publish the dissents alongside the praise.

  • 5.0 median reviewer score
  • 31 external replications
  • 2 withdrawn negative results
REVIEW · LEDGER-08
5.0 average reviewer score

What an endorsement is, and what it is not

An endorsement is a public, signed quote from a reviewer who has seen the artefacts — not a generic statement of association. Every quote on this page resolves to a review thread, a venue, or an institutional affiliation that the reviewer has agreed to publish.

Every endorsement on this page is signed.

No anonymous quotes. No "industry insider says." If a name is on this page, the lab has the reviewer's permission to attach the quote to their identity.

Their reproducibility pipeline is the closest thing to a gold-standard I have seen outside of national labs — and they ship the artefacts publicly.

Dr. M. Aldoroty · Reviewer, Nature Methods

The replication tracker is the part I keep pointing other groups at. The fact that you can see, at a glance, which claims have been independently rebuilt — that is the hard part, and they have done it.

Prof. R. Saavedra · Area Chair, NeurIPS 2025

When we ran their replicate.sh against our own infrastructure, every figure rebuilt within bounds. Twenty minutes, no privileged access. That is what a reproducible pipeline looks like.

Open Science Office, NSF

Recognised by

Nature Methods

Featured in the November 2025 issue's editorial on reproducible ML pipelines, with the chain-of-custody figure reprinted with permission.

Editorial2025-11

NeurIPS Reproducibility Track

Three accepted submissions in the reproducibility track over the last two cycles, all carrying a passing replication run at submission.

3 accepts2024–2025

NSF Open Science

Cited as a reference implementation in the NSF Open Science Office's 2026 guidance on artefact submission for federally funded ML research.

Reference2026-Q1

arXiv Replication Index

One of the eleven labs whose pre-prints automatically include a passing replication record at upload.

11 of Nsince 2024

And the dissents

Endorsements without dissents are marketing, not science. Two notable critiques the lab has received and acted on:

2024 — Cohort selection bias
A reviewer flagged that one cohort's enrolment process was correlated with the outcome variable. We adjusted the published bound and added a synthetic-cohort generator to the audit suite.
2025 — Adversarial coverage
A red-team rotation reviewer found a perturbation our public catalogue did not cover. We added the perturbation to the catalogue, retroactively re-ran two relevant claims, and amended one paper.

We publish dissents because the lab’s claim is not “we are always right” — it is “we are auditable.”

Continue exploring the lab

Open the rest of the research surface.

Each tile of the bento grid is a primitive of the workflow — instrumentation, modelling, peer review, publication. Together they form a transparent chain of custody from raw signal to citable claim.