Cut-points read best-first. For a "higher is better" KPI they descend;
for "lower is better" they ascend. A value at or beyond the first scores 1, past the
fourth scores 5.
Weights are relative within a component and renormalise around whatever
is missing, so they need not total 100% — but they do today, and keeping them there makes
the numbers easier to reason about.
Weight 0 means "defined, not yet fed". Thirty of the KPIs Morgan asked
for have no data source yet. They carry a full definition and cut-points so that the moment
a feed exists, the value lands somewhere that already knows how to band it. Typing a weight
turns one on.
THE LETTER SCALE
These are Morgan's, from the Grade Scale sheet of the KPI
proposal he reviewed. Read as [from, to): a 2.49 is a B, a 2.50 is a C.
Worth knowing when you read the distribution. A blend of
91 KPIs does not sit on the same
scale as one KPI — averaging many indicators shrinks the spread, so the
book runs about 1.7 to 4.7 and most deals land in the middle letters, even
though the KPIs underneath discriminate hard (their band distribution is
bimodal). That is a finding about the ladder to put in front of Morgan, not
one to act on here: moving a cut-point changes what a B means.
They are fixed, not fitted. Cutting at the book's own
quintiles would force one deal in five to fail for ever, however good the
book got — a ranking rather than a standard — and a mean-based
ladder would move a deal's grade because other deals moved. These
stay put, so a grade belongs to the deal.