replication review

Does the code back the claim?

Drop a whole replication package. Dodo Review traces every claim in the paper to the artifacts that should support it — and tells you where they don't.

Start a review → See pricing

Dodo Review verdict cards: claims marked true, false, uncertain, or unverifiable, with justification and evidence pointers into the code

how it works

step 01

Upload folder

Bring the replication package as-is — paper, scripts, and data files, in the layout you already have.

step 02

Trace claims to code

Dodo Review reads the paper's claims and follows each one into the scripts and outputs that should support it.

step 03

Verdict cards

Each claim gets a verdict — true, false, uncertain, or unverifiable — with justification and evidence pointers into the code.

why dodo review

Claims, not checklists

The unit of review is a claim from the paper — "standard errors are clustered by industry" — not a generic linting rule. Every verdict is about whether the code does what the text says.

Evidence you can follow

Verdicts come with pointers like analysis/table3.do#L47 — the exact line that supports or contradicts the claim, so you can check the checker.

Honest about limits

When the package doesn't contain enough to decide, the verdict says uncertain or unverifiable — no false confidence.

benchmarks soonBuilt for real research code

General-purpose LLM tools are tuned on clean software-engineering benchmarks. Dodo Review is built for the code scientists actually write — messier, less tested, and more error-prone. Benchmark numbers will land right here.