Skip to content

Methods

Every figure OpenDeco publishes has to be rebuildable from the methods on the Data, Models, Benchmark and Reproducibility pages. A correction is published as a new version, and the version it replaces stays available under its own number.

Source acquiredorigin, date, access termsNormalisedone exposure schema,pascals and secondsChecks passeddataset version frozenPrimary sourceequations, code, tablesImplementationrebuilt where noofficial code existsVerificationagainst published examplesBenchmark runeligibility and mode declaredManifest and rerunartifacts hashed,clean rerun from frozen inputsResult released
Source acquiredorigin, date, access termsNormalisedone exposure schema,pascals and secondsChecks passeddataset version frozenPrimary sourceequations, code, tablesImplementationrebuilt where noofficial code existsVerificationagainst published examplesBenchmark runeligibility and mode declaredManifest and rerunartifacts hashed,clean rerun from frozen inputsResult released
The dataset track and the model track meet at a benchmark run. A result can be published only after that run is rebuilt from its frozen inputs and produces the same artifacts.
  1. Data

    Sources keep an acquisition record and a hashed original. Pressure is stored in pascals, and depth in metres of sea water is derived from it.

  2. Models

    Implementations follow the primary equations, official source code or original tables. Where no official code exists, the implementation is rebuilt and labelled a recreation.

  3. Benchmark

    A release records which datasets and configurations were used, and names any overlap between the test data and the data a model was built on.

  4. Reproducibility

    The run manifest holds the code commit, the runtime image and the SHA-256 artifact digests. Before a release is published, the run is repeated and must reproduce the same artifacts.

OpenDeco methods are not operational procedures for planning or carrying out a dive or any other pressure exposure.