Normal Range on the Enter Measurement Page

On the Measurements view of an experiment in Uncountable, we automatically infer a “normal range” for each numeric output. This range helps identify outliers. Any value that falls outside the inferred range will be highlighted in yellow with a warning message.

What Is the Normal Range?

The normal range provides a statistical estimate of what output values are considered typical across your dataset. It is used for numeric outputs only—non-numeric outputs are not evaluated in this way.

Each output has a single normal range, regardless of its condition values (e.g., temperature, pH, etc.). This means we do not calculate separate ranges for each condition parameter set. The same range applies across all conditions for a given output.

The normal range is calculated based on a few statistics derived from the output of interest (across all material families)*:

• 5th percentile (‘5pct’)
• 95th percentile (‘95pct’)
• Standard deviation (‘std’)

How Is the Normal Range Calculated?

The range is inferred using statistics computed from all entities that share the same output, across all material families:

• 5th percentile (5pct)
• 95th percentile (95pct)
• Standard deviation (std)

We define the lower bound as the minimum of:

• 5pct – 2 × std
• 5pct / 5

We define the upper bound as the maximum of:

• 95pct + 2 × std
• 95pct × 5

So, the inferred normal range is:

Special Cases and Snapping

In some cases, the calculated lower bound may be negative, even though all observed values are non-negative. In these cases, we snap the lower bound to 0.

For example, if all values of an output are greater than 0, but the calculated lower bound is –2, we will adjust the range to start at 0 instead.

Note that there are additional rules that apply under certain edge cases. We “snap” the values to 0 depending on the distribution. For example, if an output is always greater than 0, we will snap the lower bound to 0, even if the calculated lower bound is below 0.

Working Example

Suppose Output A has the following statistics across your data:

• 5th percentile = 0.5
• 95th percentile = 10
• Standard deviation = 10
• No recorded values are negative

Then:

• Lower bound = min(0.5 – 2×10, 0.5 / 5) → min(–19.5, 0.1) → –19.5, snapped to 0
• Upper bound = max(10 + 2×10, 10 × 5) → max(30, 50) → 50

Inferred normal range = [0, 50]

*Detailed explanations of percentile and standard deviation can be found on Wikipedia.