Skip to content

validation.md

Latest commit

 

History

History
42 lines (33 loc) · 2.28 KB

validation.md

File metadata and controls

42 lines (33 loc) · 2.28 KB
Raw

Validation record

Hand and synthetic benchmarks

The exact benchmark generates M_n = (1-alpha) alpha^(n-1) at alpha 0.8. Regression over C3–C20 must return 0.8 within 1e-12; ftproductlab validate exposes this check without test framework dependencies.

Independent direct sums verify molecular and carbon distributions close to one and verify the analytical finite/tail cut formula. Other regression tests cover scale invariance, mass-to-mole conversion, category aggregation, input validation, fit failure for a positive slope, zero denominator, below-detection state, CSV schema, CLI/GUI agreement through the shared service, export, plotting, and import isolation.

Public-data validation

The 0.1.0 package was first exercised unchanged against three external cases. FTP-PUB-002 uses a raw, repository-hosted workbook; FTP-PUB-003 uses processed experimental tables in peer-reviewed supporting information; FTP-PUB-001 is a published-number and metadata benchmark because the combined per-carbon values needed to reproduce the reported fit are not tabulated.

Results include:

  • Cambridge Ru/TiO2 wax distributions, C30–C50 uniform fits: alpha = 0.88372625, 0.92999148, and 0.93840928 for feed H2/CO ratios 2, 1, and 0.5. Independent ordinary least-squares calculations on the workbook's logged values agree to floating-point precision.
  • Gao et al. industrial Fe/Mn Run A-2, uniform fits: alpha = 0.88235535 for C6–C20, 0.94460763 for C31–C60, and 0.92731071 for C6–C60. The fit-range dependence and structured residuals are evidence against silently adding a more complex model.
  • Partington et al. report alpha = 0.92 over C10–C40 and C5+ = 85.0% for a Co/TiO2 pilot-plant product. The article supports the explicit range/C1/C2 conventions, but the combined C10–C40 values needed for exact reproduction are figure-only.

The Cambridge case exposed one 0.1.0 plotting defect: observations used the measured-data total while the fitted overlay used its own predicted total. Numeric alpha, residuals, and exports were unaffected. A CC-BY-4.0 minimal fixture now reproduces that discrepancy and guards the corrected common-denominator overlay.

Complete provenance, checksums, scope, search record, and discrepancy classification are in public-data-sources.md.