Environment reconstruction
Record operating system, language and dependency versions, hardware assumptions, seeds, configuration and the exact commands required to run the workflow.
Computational reproducibility audit
Find out whether the declared result can be rebuilt.
Haivian audits one frozen computational claim against its code, data, environment and recorded outputs. The goal is an inspectable account of what reran, what matched, what failed and what could not be tested.
What the audit can examine
Code, environment, data and claim in one scope.
The accepted plan reflects the artifact actually supplied. Reproducibility is not inferred from a repository link alone.
Data and provenance
Identify input datasets, preprocessing, permissions, checksums and substitutions. Private or unavailable inputs are stated as limits rather than silently replaced.
Output comparison
Compare produced artifacts, logs, metrics and tolerances with the declared result. Distinguish exact agreement from statistical or qualitative similarity.
The reproducibility record
Evidence that another reviewer can inspect.
The report preserves the difference between a successful rerun and a broader scientific validation.
Freeze
Identify the commit or archive, data version, environment, commands, expected outputs, claim and permitted access boundary.
Reconstruct
Build the declared environment without weakening secrets, licences or customer controls. Record deviations required to make the workflow executable.
Execute
Run the agreed workflow and preserve return codes, logs, hashes, timings and resource observations that are meaningful to the claim.
Interpret
State whether the exact result reproduced and whether that rerun supports the wider claim. Report missing evidence and alternative explanations.
What this service does not promise
A rerun is evidence, not blanket validation.
A successful execution does not by itself establish novelty, causal validity, generalization, security or production performance. A failed execution may identify missing instructions or environmental drift rather than refute the scientific claim.
Private repositories and restricted data require an agreed access path. Haivian does not ask customers to publish secrets or waive third-party licences. Any sanitized public artifact release remains optional and separately approved.
Inputs and fit
What makes a computational result auditable.
A reproducibility audit works best when the claim, executable path and expected evidence can be tied to one frozen release.
Supply a stable package
Provide an archive or commit, environment specification, licences, data identities, commands and expected outputs. Identify credentials or restricted inputs without putting secrets into public files or ordinary messages.
Define comparison rules
State whether outputs must match exactly, within a numerical tolerance or by a declared statistical criterion. If randomness matters, specify seeds, repetition counts and the evidence needed to assess variation.
Separate execution from inference
Identify the scientific conclusion that the computation is meant to support. Haivian reports both whether the workflow reran and whether that execution is sufficient for the stated claim.
Existing public examples. The Validation Journal includes a record-based protocol audit that distinguishes archived digest agreement from missing historical wire frames, and a Taylor–Green vortex validation that separates exact analytical checks from solver or experimental claims. Each example is version-bounded and states what was not reproduced.