Set Uncertainty

Shipping

set_uncertainty

Attach, override, or clear a quantity's standard uncertainty (±σ)

Signature

Inputs

valueScalar|Signalrequired— The quantity to receive the new uncertainty. Its numbers are untouched.
sigmaScalar|Signal— The uncertainty source, read only in **From σ input** mode. A Scalar applies uniformly; a Signal applies per-sample (length must match). If it carries a compatible unit it is converted into the value's unit first; incompatible units error. Negative values are taken as magnitudes.

Outputs

resultScalar|Signal— Same shape and numbers as `value`, carrying the new σ² (or exact when cleared / zero).

Parameters

Key	Type	Default	Notes
`mode`	enum	absolute	one of: absolute, relative, from_input, clear
`uncertainty`	float	0.0	σ in the value's unit (Absolute mode) or in percent of each value (Relative mode). 0 keeps the data exact. Active only in Absolute / Relative modes.

Description

Set Uncertainty is the entry point for measured error bounds. A calibration sheet says "±0.5 %", a datasheet says "±2 mV", a noise-floor estimator produces a per-sample σ signal — this node turns any of those into the engine's variance contract ( $σ^{2} = u n cer t ain t y^{2}$ , with None meaning exact), from which the uncertainty kernel propagates through the rest of the graph.

The value's numbers are never touched; only its σ² is set. Four modes:

Absolute (±σ) — a constant σ in the value's own unit, applied uniformly. $0 \Rightarrow$ exact.
Relative (±%) — $σ_{i} = ∣ v_{i} ∣ \cdot p /100$ per element, so the band scales with the signal. $0 \Rightarrow$ exact.
From σ input — σ comes from the sigma port, embodying the Captyse principle that parameters can be data. A Scalar σ applies uniformly; a Signal σ applies per-sample (lengths must match). If the σ carrier has its own compatible unit it is converted into the value's unit first ( $5 V$ value, $200 mV$ σ $\Rightarrow σ = 0.2 V$ ); incompatible units error. Only the scale ratio is used — an offset is meaningless for a spread (ADR-002). Negative σ values are taken as magnitudes.
Clear (exact) — drop uncertainty entirely, back to exact.

A per-sample σ Signal can only attach to a Signal value — feeding one to a Scalar value errors (use a Scalar σ instead). Any non-finite or non-positive σ collapses to exact.

Mathematics

Absolute: σ_{i} = ∣ a ∣ \Rightarrow σ_{i}^{2} = a^{2}

Relative: σ_{i} = ∣ v_{i} ∣ \cdot \frac{p}{100} \Rightarrow σ_{i}^{2} = (∣ v_{i} ∣ \frac{p}{100})^{2}

From input: σ_{i} = ∣ s_{i} \cdot f a c t or ∣, f a c t or = \frac{sc a l e _{σ -unit}}{sc a l e _{v a l u e -unit}}

Examples

Constant absolute σ

absolute mode, uncertainty = 0.5 on a $V$ signal [1, 2] yields uniform $σ = 0.5$ , i.e. $σ^{2} = [0.25, 0.25]$ .

Relative percentage band

relative mode, uncertainty = 10 on [1, -2] V gives $σ = [0.1, 0.2]$ ( $σ^{2} = [0.01, 0.04]$ ) — the band scales with each value's magnitude.

σ from a differently-unitted input

from_input mode: a $5 V$ value with a Scalar σ of $200 mV$ is unit-converted to $σ = 0.2 V$ , so $σ^{2} = 0.04$ . A $k g$ σ against a $V$ value errors (incompatible dimensions).

Per-sample noise floor

from_input with a σ Signal [0.1, 0.2] against value [1, 2] gives $σ^{2} = [0.01, 0.04]$ . Lengths must match, otherwise it errors.

Clearing to exact

clear mode drops any existing σ, returning the value as exact (variance None).

Applications

Stamping a datasheet or calibration tolerance (±σ or ±%) onto acquired data so uncertainty propagates through all downstream math.
Driving per-sample uncertainty from a measured noise-floor estimate wired into the σ port.
Converting a manufacturer's spec expressed in a different unit (mV) onto a value in another (V) without manual rescaling.
Clearing accumulated uncertainty to treat a signal as exact for a section of the graph.

Neat

Parameters can be data: the **From σ input** mode reads the uncertainty from a wire, so a live noise estimator can set the error band sample-by-sample.

The σ carrier's unit is reconciled by scale ratio only — offsets are deliberately ignored because a spread has no origin (ADR-002), so a 200 mV σ becomes 0.2 V cleanly.

Relative mode ties the band to |value|, which naturally produces a proportional confidence envelope on plots without any per-sample bookkeeping.

Known issues

A per-sample σ Signal cannot be attached to a Scalar value — feed a Scalar σ instead, or the node errors.

In From σ input mode the σ Signal length must exactly match the value Signal length; a mismatch is an error.

Any non-finite or non-positive resulting σ collapses to exact (variance cleared) rather than being stored.