Validation

What the evidence supports.

Checking software, evaluating a scientific method, and validating a tool in operational use answer different questions. Each needs its own evidence.

Software verification

Repeatable results are a starting point

PTEM Labs retains input and software versions for replay and uses schema, regression, and integration tests to check inputs, software changes, and releases. The public research package includes code, checksums, and tests for reproducing Temporal V2.

Passing tests does not establish scientific validity, decision accuracy, or operator benefit. A repeatable calculation can still be unsuitable for a particular question.

Scientific evaluation

Historical Atlantic intensity records

The EarthArXiv preprint evaluates Temporal V2, a diagnostic derived from maximum sustained wind in Atlantic HURDAT2 records: 1,991 storms from 1851–2024. It examines persistence in intensity history and its descriptive relationship to rapid-intensification occurrence. The preprint has not been peer reviewed.

The tested simple persistence comparator had a better model fit than Temporal V2 in the reported duration-controlled comparison. Temporal V2 activation was not generally an advance marker of rapid intensification. These findings do not establish forecasting skill, causality, or a physical storm-structure measurement.

Zenodo v4.0 provides the data and code for reproducing the temporal analysis and comparisons. The exact environmental input matrix is not included; those analyses require additional inputs.

The archive’s date range describes the research dataset. The free Observatory currently offers six selected storm exhibits. These findings concern Temporal V2; they do not establish the performance of independent methods or assess PTEM Labs’ broader engineering capabilities.

See Research for the public artifacts and full scope of the reported work.

Operational validation

Application-specific evidence is still required

The public research does not establish operational validation. The navigation prototype uses three simulated source families and has not been validated for operational use or measured operator benefit. Hurricane findings do not validate navigation or other applications.

Customer-specific integration, deployment, and cybersecurity readiness require separate qualification. Plans describe intended work. Completed technical work can be supported by implementation records, tests, and evaluation results, whether public or private. Discussing an evaluation is a way to define the evidence needed for a particular use case.