Uncertainty fundamentals

rtd_sensor.uncertainty provides numerical building blocks for measurement uncertainty. It does not decide which physical effects belong in your measurement model or which probability distribution is justified.

That separation is important: software can perform the arithmetic, but the measurement evidence must justify the assumptions.

Convert a symmetric bound to standard uncertainty

For a bound of ±a, you can explicitly choose a rectangular or triangular model:

from rtd_sensor import uncertainty

u_rectangular = uncertainty.standard_uncertainty_from_bound(
    0.35,
    distribution="rectangular",
)

u_triangular = uncertainty.standard_uncertainty_from_bound(
    0.35,
    distribution="triangular",
)

The numerical formulas are:

rectangular: a / sqrt(3)
triangular:  a / sqrt(6)

The choice of distribution is yours to justify.

Convert expanded uncertainty back to standard uncertainty

If an uncertainty source explicitly provides an expanded uncertainty and coverage factor:

u_standard = uncertainty.standard_uncertainty_from_expanded(
    0.20,
    coverage_factor=2.0,
)

Combine independent standard uncertainties

combined_u = uncertainty.combine_independent_standard_uncertainties(
    0.04,
    0.07,
    0.02,
)

This uses root-sum-square and assumes the supplied components are independent or uncorrelated for the purpose of this calculation.

Expanded uncertainty

expanded_u = uncertainty.expanded_uncertainty(
    combined_u,
    coverage_factor=2.0,
)

The function does not assign a confidence level to k=2. A probability interpretation requires justification from the complete uncertainty analysis.

Type A and Type B labels

TemperatureUncertaintyComponent can retain an optional evaluation_method of "A" or "B", plus source and note text. Those are provenance fields. They do not change the arithmetic and do not transform an invalid uncertainty quantity into a valid one.

Current limits

The simple combination helpers do not yet model covariance between correlated components, coefficient covariance, effective degrees of freedom, or Monte Carlo propagation.

For RTD-specific resistance propagation, continue to Resistance uncertainty propagation.