Validation reports

These reports show the saved visual results of the torch-flash validation notebooks. Experimental markers, model curves, parity plots, and residual plots are retained so the tested systems and model behavior remain visible. The underlying research CSVs, numerical tables, and local-only notebook pairs are not published.

The reports distinguish three kinds of evidence:

• validation compares a selected model and parameter set with experimental observations;
• verification checks equations against a published calculation or independently generated software reference; and
• application demonstrates a computed property, sensitivity, or model surface without claiming experimental agreement.

Available reports

• GERG-2008 and EOS-CG — measured compressibility factors and the complete EOS-CG Table 8 verification set.
• Native EOS-CG-2021 — dense CO2/H2 numerical verification plus experimental MDEA density and speed of sound.
• Autodifferentiated thermal properties — CO2 and propane Joule-Thomson validation and homogeneous-state property profiles.
• Huron-Vidal mixing — phase diagrams, parity, and complete-temperature transfer for two alcohol/hydrocarbon binaries.
• Cubic volume translation — methane/n-decane density response for SRK and PR78 translations.
• Cross-co-volume interaction — before/after density fitting with a disjoint validation isotherm.
• Original UNIFAC — predictive alcohol/hydrocarbon phase diagrams and activity-coefficient behavior.
• Predictive Peng-Robinson 1978 — calibration-domain hydrocarbon VLE comparisons.

Publication and reproduction boundary

The figures are selected PNG outputs copied directly from top-to-bottom executed notebooks. A marker visibly represents an observation coordinate, but the source row, machine-readable value, CSV, and notebook output table are not included in the public documentation. This figure-only decision does not change the source dataset's license or establish a general right to redistribute it; see licensing and data provenance.

The local-only notebook pairs can be synchronized, executed, and republished in a checkout containing the lawful research inputs:

pixi run -e benchmarks notebooks-sync
pixi run -e benchmarks notebooks-run
pixi run -e notebooks validation-figures
pixi run -e default check-data-rights

The figure manifest records the exact executed-notebook checksum, source cell, evidence class, and data dependencies for every published plot.