A set of orthogonal (mutually independent) vectors (or functions) that can be used to unambiguously identify exactly one region in an abstract space of interest.
When used in the context of technical requirements: the list of topics (some or all of which might be functions) which, when we are talking about any one of them, we need mention no other. The situation makes it easier to avoid accidentally contradicting ourselves between one requirement and another allocated to the same CI. Making this happen turns out to be difficult, but is no rat hole.
Change of Basis
We can often “change the basis”, meaning that we’ve altered the frame of reference without altering the thing measured therein. Some basis changes are exact (e.g., 1 foot converted to 12 inches), but others are not (4 inches converted to 1.333333…feet). The Fourier, Windowed Fourier, Discrete Cosine, and Wavelet Transforms are other commonly-encountered examples of basis change in which different kinds of precision can be lost, depending on the inputs being analyzed.
In all basis changes, the potential loss of precision due to the change must explicitly be considered. It cannot be statistically combined with “experimental errors” or other sources of uncertainty, but must be considered separately. Many basis changes are absolute and, therefore, not subject to stochastic variation. Because they are strongly algorithmic, basis changes usually don’t introduce bias. The only exception of which I am aware1 is where the basis change can be used to desensitize the reviewer to the magnitude of differences in value.
Footnotes
- And, unfortunately, that awareness is acute.[↩]