CIHigh is the upper bound of that interval. Wilson rather than the normal approximation because the normal one produces bounds past 100 exactly where benchmark scores live — at 194/198 that is the top of the board, not a corner case.
CILow and CIHigh are the 95% Wilson interval on Measured, in percent. They are what makes the score comparable: at n=198 a 98% carries roughly ±2 points, so most differences at the top of a board are not distinguishable and a bare number implies a precision it does not have. Absent when there is no measurement.
Mean is the unweighted average of every claim, which answers a different question from Published: what the field says on average, rather than what the vendor says about itself. With one claim the two are equal.
Run names the measurement Measured came from, and MeasuredAt is when it ran. A score with no date is not a fact about a model, it is a fact about a model on a day — and models change, so the date is what makes the number checkable rather than merely quoted.
Spread is the distance between the highest and lowest of them, nil when there is only one. It is the disagreement AMONG sources, which a single Published number cannot show — signal in the same way the published-minus-measured gap is.