Reproduction
Re-execute the reported work using the supplied artifact or a reconstructed equivalent, recording material deviations.
How verification works
See how VRL turns a research claim and its artifacts into a traceable, bounded conclusion.
The claim, decision, test conditions, and tolerances are fixed before we examine the result.
Replication
Test the same claim with an independently constructed implementation, dataset, or experimental design.
Claim and conditionsWhat is being tested?
Materials and environmentWhat can be reconstructed?
Independent testsWhat happens when the work is rerun or varied?
Conclusion and limitsWhat does the evidence establish?
What you provide
The paper or claim, available artifacts, expected results, access constraints, and the decision or publication context.
Where scope is fixed
Before execution, the test plan records claims, conditions, tolerances, exclusions, publication terms, and how assistance will be logged.
What happens before publication
The subject may correct factual errors and submit a response. The method and conclusion remain under independent control.
Evaluation stages
Open a stage to see what it means for the client.
01Define the claimRecord the claim, the decision it informs, tolerances, conditions, and exclusions.
You identify the result or decision that matters. We turn it into a testable statement and agree on what would count as support, failure, or an inconclusive result.
02Preserve the evidencePreserve code, data, versions, provenance, access conditions, and known gaps.
You provide the available paper, repository, data, environment, and access constraints. We record what was received and preserve the versions used.
03Reconstruct and reproduceRebuild the environment, record deviations, and compare outputs with declared tolerances.
We first work from the supplied instructions. Requests for assistance and material deviations are logged so the record shows what an independent evaluator could and could not do.
04Audit the methodInspect baselines, workloads, measurement, statistics, exclusions, and validity threats.
The review follows the evidence behind the claim, including comparison choices that may make a reported advantage appear stronger or weaker.
05Test the boundariesVary reasonable parameters, workloads, datasets, environments, and cost assumptions.
Sensitivity tests are selected for their ability to change the conclusion, not to produce a larger volume of experiments.
06State the conclusion and limitsReport the result, uncertainty, limits, review status, and revalidation trigger.
You receive a conclusion tied to the tested conditions, with unresolved questions and the circumstances that would require revalidation.
Evidence labels
Labels describe completed work at claim level. They do not grade a whole paper or replace the report's stated limits.
ArtifactWhat materials were found and whether they could be used.
- Identified
- Available
- Functional
ExaminationWhich forms of independent testing were completed.
- Reproduced
- Method audited
- Robustness tested
- Independently corroborated
ConclusionHow far the observed evidence supports the claim.
- Supported within tested conditions
- Not reproduced
- Inconclusive
Have a specific claim?