Calibration overlap
Known relationships between evaluation data and model development are recorded with the result. Overlapping evidence is not presented as independent validation.
Research practice
Methods
How OpenDeco handles source data, model identity, implementation verification, benchmarking and reproducibility.
Method guides
From source to release
Data
Source material is preserved separately from normalized research records. Transformations, exclusions and derived fields are versioned.
Acquisition · provenance · normalization · quality controlRead methodModels
Model records distinguish published theory, exact parameter sets, implementations, procedures and configurations.
Identity · primary sources · verification · configurationRead methodBenchmark
Benchmark analyses are defined against explicit datasets, outcomes, metrics, exclusions and model configurations.
Eligibility · overlap · metrics · sensitivityRead methodReproducibility
Released analyses retain the inputs, code identity, run metadata and artifacts needed to reconstruct the result as far as source rights permit.
Frozen inputs · manifests · checksums · citationRead method
Across every analysis
Research controls
The record around a result is part of the result. These controls apply across data, model and benchmark work.
Uncertainty and sensitivity
Results state the assumptions that materially affect interpretation and test whether conclusions survive reasonable analytical choices.
Versioned record
Corrections create a new version. Published inputs, methods and limitations remain identifiable. They do not change silently.
OpenDeco methods support research and interpretation. They are not operational procedures for planning or conducting a pressure exposure.