neqsim-benchmark-reference-data

name	neqsim-benchmark-reference-data
calculation_basis	data-retrieval
version	0.1.0
description	Turn the mandatory benchmark-validation step of a NeqSim task into a reproducible, source-traceable comparison. Supplies a registry of independent reference sources with authority tiers, validated ranges and stated uncertainties (IAPWS-95, IAPWS-IF97, Span-Wagner CO2, Setzmann-Wagner methane, Span nitrogen, Bucker-Wagner ethane, Lemmon propane, GERG-2008, CoolProp HEOS, NIST WebBook), an offline anchor table of published critical points, triple points, boiling points and one ambient liquid density that runs with no optional dependency and no network, an optional CoolProp backend for reference values at any state, and a comparison layer that grades PASS/WARN/FAIL, rejects a reference that does not outrank the model basis, records whether the deviation is inside the reference's own uncertainty, enforces the three-point minimum, and emits the exact benchmark_validation block that the task report generator and CI gate consume. USE WHEN: a task must validate NeqSim output against independent reference data, a benchmark notebook is being written, a benchmark_validation block must be produced for results.json, a reported deviation must be traced to a citable source, or an existing benchmark claim must be checked for independence and resolution before it is trusted.
last_verified	2026-08-08
requires	{"python_packages":[],"java_packages":[],"env":[],"network":[]}

Benchmark Reference Data

Every Standard and Comprehensive NeqSim task must compare its results against independent reference data before the numbers are reported. In practice that step gets rewritten in every benchmark notebook, with the reference values pasted in as bare literals, no citation, an invented tolerance, and no check that the reference is actually independent of the model being tested.

This skill supplies the parts that should not be rewritten: where the reference came from, whether it is allowed to be used as a benchmark here, how far the model may deviate before the comparison fails, and what the result block must look like so the report generator and the CI gate accept it.

It does not compute NeqSim results. It provides the other side of the comparison and the verdict.

When to Use

• A task is writing its XX_benchmark_validation.ipynb notebook.
• A benchmark_validation block must be produced for results.json.
• A property, critical constant, or saturation condition from a NeqSim fluid needs an independent, citable reference value.
• A number quoted in a report needs a traceable provenance chain from value to citation.
• An existing benchmark claim must be audited: is the reference independent, is the deviation larger than the reference's own uncertainty, are there enough points?

When Not to Use

• As a property engine. CoolProp and the anchor table are references, not the model — do not use them to produce the answer the task is asked for.
• For mixtures outside a reference formulation's validated range. The applicability gate will refuse, and it is right to refuse.
• As a substitute for measured data on the actual fluid. A reference EOS for a pure component does not validate a characterised reservoir fluid; it validates the pure-component limit of the model.
• To manufacture a PASS. Loosening the tolerance until a comparison passes is the failure mode this skill exists to make visible.

Inputs

Input	Meaning
model_value	the number the task computed (NeqSim, or any model under test)
reference	a ReferencePoint from the anchor table or the CoolProp backend
tolerance_pct	acceptance band in percent; defaults per property when omitted