Empirical evidence · controlled claim boundary

Validation Guide

Physical validation asks whether the mathematical model adequately represents measured reality for an intended use. It is not established for physicsbase today.

NOT_ESTABLISHEDphysical validation

0eligible experiments

0rejected records

0implied passes

This is deliberately not a PASS. Three comparisons with published measurements are documented as experimental benchmarks, but none supplies the complete quantified measurement/model uncertainty and independently preregistered protocol required here. Analytical solutions, NAFEMS targets, solver residuals, and agreement with Abaqus are solution-benchmark evidence; empirical screening without the admission controls is not promoted to physical validation.

Evidence admission gate

Traceable experiment

Permanent citation, versioned raw data, license, local SHA-256 match, specimen, facility, instrumentation, conditions, and data-reduction method are mandatory.

Uncertainty

Every observation must carry a numerical experimental uncertainty with basis and coverage. Every prediction must carry model-form/input/numerical uncertainty with a stated basis.

No data leakage

Calibration and validation observation IDs must be disjoint. Their freeze record and the acceptance metric, threshold, uncertainty combination, and rationale must predate model execution.

Adopted comparison metric

For an observation y and prediction m, the executable gate uses |m-y| / sqrt(u_exp² + u_model² - 2 rho u_exp u_model), where both u values are harmonized standard uncertainties and rho is a predeclared correlation coefficient with a written rationale. The study owns the threshold and applicability domain. This metric does not erase bias, coverage mismatch, or sparse-data limitations; those remain study responsibilities.

Current coverage

Domain	Status	Evidence required before a claim
linear static structures	NOT_ESTABLISHED	independent coupon/component tests with force, displacement, strain, uncertainty, and held-out predictions
modal dynamics	NOT_ESTABLISHED	measured mode frequencies and shapes with sensor, fixture, mass-loading, and repeatability uncertainty
transient dynamics	NOT_ESTABLISHED	time-synchronized load/response histories with bandwidth and timing uncertainty
heat transfer	NOT_ESTABLISHED	calibrated temperature/heat-flux measurements with thermal contact and boundary-condition uncertainty
nonlinear material response	NOT_ESTABLISHED	separate calibration and component-level validation datasets with batch/material variability
contact and friction	NOT_ESTABLISHED	measured contact force, slip, pressure or displacement with independently characterized friction
CAD-to-mesh-to-solve workflow	NOT_ESTABLISHED	manufactured-part measurements and mesh/process repeatability across supported CAD classes

What the executable gate rejects

Analytical, manufactured-solution, or cross-code results labeled as physical validation.
Experimental numbers without positive numerical uncertainty and a stated basis/coverage.
Predictions evaluated on calibration data or missing held-out predictions.
Acceptance thresholds chosen after seeing the validation outcome.
Broad claims without an intended use, validated range, and exclusions.

Path to established validation

Select a narrow intended use and quantity of interest.
Acquire licensed raw experimental data and full test metadata from an independent source or commissioned laboratory.
Pre-register calibration inputs, held-out validation IDs, metric, uncertainty model, and threshold.
Freeze the solver version and model before evaluating held-out data.
Publish PASS or FAIL with residual plots, uncertainty intervals, sensitivity, and applicability limits. Retain failures.

Methodology sources

ASME V&V 10-2019, Standard for Verification and Validation in Computational Solid Mechanics
NASA-STD-7009B, Standard for Models and Simulations

These inform the evidence separation; physicsbase does not claim certification by either organization.

Reproduce the status

python -m validation.suite

The schema and evaluator live in validation/schema.py. Framework unit tests receive no qualification credit.