Skip to content

OpenDeco research programme

Global Decompression Benchmark

A reproducible comparison of documented decompression models, procedures and configurations against versioned exposure and outcome datasets.

Research use only. Benchmark outputs are scientific comparisons, not dive plans or operational decompression guidance.

01

Evaluation

Evaluation is performed only where the available dataset and outcome definition support the question being asked.

IdentityFamily · version · implementation · configuration
EvidenceDataset · outcome · lineage · access
AnalysisMetric · uncertainty · support · failure region

02

Calibration and external validation

A model can perform well on data used to develop or tune it. OpenDeco records known model–dataset relationships and separates calibration-related results from independent external evaluation. Unknown lineage remains labelled as unknown.

03

Results

Results are organised by scientific question. OpenDeco does not publish a universal safety score or manufacture a single overall winner.

04

How to read the results

Performance belongs to a defined dataset, outcome, configuration and protocol. It is not a property of a model name in isolation.

  1. 01

    Model identity

    Check the exact version, implementation and configuration.

  2. 02

    Evidence

    Check the dataset, outcome definition, event count and relationship to model development.

  3. 03

    Uncertainty

    Read point estimates with their intervals and support counts where the analysis requires them.

  4. 04

    Failure regions

    Treat exposure regions with little or no supporting data as unsupported, not as evidence of equivalence.

05

Reproducibility

Public releases include dataset versions, model identities, code commit, run manifest, protocol version and artifact checksums. Restrictions that prevent exact redistribution or reproduction are stated explicitly.