Neutrino Atlas
EN
  1. Terms
  2. Analysis & measurement
  3. Statistics & inference
  4. Sensitivity

Sensitivity

ConceptLevel 3 of 4Also called: median sensitivity, discovery potential

The median result an experiment expects in the absence (or presence) of a signal, used for design and comparison.

Sensitivity is computed from many simulated pseudo-experiments. Comparing it with the actual result shows whether an observation fluctuated up or down. For example, LEGEND-200's median sensitivity was 1.0 × 10²⁶ yr while its observed limit was 0.5 × 10²⁶ yr.

In plain words

Before an experiment is built, and again before its data are analysed, physicists ask: what result should we expect if nature is as we assume? Sensitivity is the answer. It is not a measurement but a forecast, the typical outcome of the experiment. Comparing the real result with that forecast shows whether the data were lucky, unlucky or simply average.

Two kinds

  • Exclusion sensitivity: the median upper limit the experiment would set if there were no signal. For double beta decay it is a half-life; for KATRIN a neutrino mass.
  • Discovery potential: the size of signal that would be detected at a given significance, often 3σ or 5σ, in half of all repetitions.

How it is computed

Analysts generate many pseudo-experiments from Monte Carlo simulation, each a possible outcome with random fluctuations, and run the full analysis on each. The median result is the sensitivity, and the spread gives the familiar bands at ±1σ and ±2σ, sometimes called the 'Brazil band' after its green and yellow colours. A shortcut, the 'Asimov' data set described by Cowan, Cranmer, Gross and Vitells in 2011, replaces the ensemble with a single data set equal to the expectation.

Examples

  • LEGEND-200 reported an observed limit of T½ > 0.5 × 10²⁶ years against a median sensitivity of 1.0 × 10²⁶ years: a weaker limit than expected, consistent with an upward background fluctuation. The collaboration aims at a final 3σ discovery sensitivity beyond 10²⁷ years.
  • JUNO was designed to reach about 3σ sensitivity to the mass ordering with roughly six years of data.
  • DUNE and Hyper-Kamiokande present their CP-violation reach as the fraction of possible δCP values that could be established at 3σ or 5σ.

Why it matters

Sensitivity is how experiments are compared fairly and how funding agencies judge proposals. It also guards against over-reading results: a limit much stronger than the sensitivity usually reflects a downward background fluctuation, not a superior detector, which is why the Feldman–Cousins authors recommended quoting both.

Frequently asked questions

Can an experiment do better than its sensitivity?

Its observed limit can be stronger if background happens to fluctuate down, but that is luck. On average, an experiment achieves exactly its median sensitivity.

Why use the median rather than the mean?

Limits from pseudo-experiments often have skewed distributions. The median is robust: half of experiments do better, half worse.

Related terms

Mentioned in 6 pages

Last updated: · Editorial standard

Cite this page

Text

Neutrino Atlas (2026). “Sensitivity”. Retrieved 22 November 2026, from https://neutrino-atlas.com/terms/sensitivity/

BibTeX

@misc{neutrinoatlas2026termssensitivity,
  title        = {Sensitivity},
  author       = {{Neutrino Atlas}},
  year         = {2026},
  url          = {https://neutrino-atlas.com/terms/sensitivity/},
  note         = {Accessed 2026-11-22},
  language     = {english}
}