Uncertainty and Units

What it does

• Keeps units attached: enforces attaching units at input and stripping only at output with m_as("unit"), and catches offset-temperature arithmetic, logarithmic units that multiply on +, and bare .magnitude calls.
• Propagates uncertainty properly: runs the GUM law of propagation (JCGM 100) and Monte Carlo (JCGM 101) on the same model, then applies the clause 8 validation test to decide which result may be reported.
• Builds uncertainty budgets: converts calibration-certificate expanded uncertainties and rectangular/triangular limits with the correct divisor, then computes u_c, sensitivity coefficients, Welch-Satterthwaite effective degrees of freedom, k from the t-distribution, and U.
• Formats results: rounds the uncertainty first, then the value, emitting 12.346 ± 0.023 mm, 12.346(23) mm, scientific and LaTeX forms, plus the sentence that must accompany the number.
• Audits existing code statically: nine rules (UNIT001–CONST001) flag a missing absolute_sigma in curve_fit, np.std without ddof, ufloat rebuilt from nominal/std_dev (destroying correlations), and hard-coded CODATA literals. Exit code 1 makes it CI-ready.
• Sanity-checks magnitudes: 14 dimensionless groups (Reynolds, Peclet, Damkohler, Knudsen, Biot, Womersley...), characteristic scales like diffusion time and Debye length, and curated observed-magnitude bands — with dimensionality verified before any number is computed.

Who it's for

• Experimentalists and grad students who must report values with a defensible ±
• Metrology, calibration, and QA engineers working from certificates and data sheets
• Engineers/developers doing fluid, heat-transfer, or reaction calculations in Python
• Anyone reviewing someone else's analysis script for silent unit or uncertainty bugs

Examples

1. Density budget: feed mass 250.0 g (u=0.05), diameter 20.0 mm and height 40.0 mm (rectangular) to propagate_uncertainty.py and get u_c, sensitivity coefficients, per-component budget in percent, effective dof, k, U, both Monte Carlo coverage intervals, and a verdict on the linearization.
2. Legacy code review: audit_units.py --input analysis.py --fail-on medium surfaces a curve_fit call missing absolute_sigma=True, where parameter uncertainties differed by 31% on a synthetic fit.
3. Regime check: compute Reynolds for blood (1060 kg/m³, 0.5 mm/s, 8 µm, 3.5 mPa·s) to confirm laminar flow, or have "2 m eukaryotic cell" rejected as 4.3 decades outside the observed 5–100 µm band.

Install the skill I found on Claude Skill Mart.
Copy the skills/uncertainty-and-units folder from the GitHub repo K-Dense-AI/scientific-agent-skills into my ~/.claude/skills/uncertainty-and-units/.
When it's done, tell me in one line what this skill can do.