The most consequential rule on this site is about the numbers that are not there. Handle the gaps wrong and every score becomes a measure of our own diligence rather than the provider’s conduct.
Two rules govern absence. The first: no evidence, no score. A fact reaches a scoring rule only if it carries a citation to a primary source, so a fact with an empty citation list cannot be built in the first place, there is no way to smuggle an unsupported claim into the arithmetic. The second follows from it: a dimension we hold no evidence for is excluded from the weighted mean, not entered as zero.
Why absent cannot mean zero
Scoring a gap as zero would mean penalising a provider for work we have not done. Worse, it would systematically reward whichever providers we happened to research least, the thinner our file on a provider, the more zeros it would dodge simply by our not having looked. That is the precise failure the whole model exists to avoid, so a dimension with no facts at all scores neither 0 nor 10 but nothing, and drops out of the mean.
The price of this rule is that scores are not automatically comparable, and we pay it in the open. Every dimension carries a coverage fraction, how many of its rules had evidence to act on, and every headline carries a scored-weight. An 8.4 scored on 72% of the methodology means 28% had no data, and the page says so beside the number. The interval’s floor makes the same admission structurally: it counts everything unmeasured as zero, so it is the number that does treat absence as nothing, as a floor to rank by, never as a penalty printed on the headline.
The gap we found, versus the gap they left
Two kinds of absence look alike and are not. notAssessed is our gap: we have not looked. notDisclosed is a cited negative: we looked, and the provider does not disclose the thing, a scored finding, cited to the page that fails to mention it. The absent-data invariant protects the first. It must never quietly excuse the second, which is a real negative the provider owns.
The honest limitation
There is a hole in this, and we would rather name it than let you find it. Invariant two, “absent data never lowers a score”, becomes, for a normalized dimension, “absent data raises a ratio.” A normalized score is points earned over points available, counted only across the rules that had evidence. Delete a single citation inside such a dimension and the denominator shrinks with the numerator, and the ratio can rise. The interval and the dominance rule close the renormalisation channel; they leave this citation channel open.
The through-line is one sentence: on this site, “no evidence” is a statement about the limits of our work, never a charge against the provider, and so it can hold a rating down from certainty, but it can never push one down as a penalty.