The totals this site will not calculate

Most of the engineering in a costing tool goes into producing numbers. A surprising amount of the trustworthy part goes into declining to.

A single headline figure is easy to build and easy to read, and that is exactly what makes it dangerous when the things underneath it are not the same kind of thing. Adding two currencies produces a number with no unit. Averaging a wholesale price against a retail one produces a price nobody pays. Both look completely ordinary on a dashboard.

What is kept apart, and by what

Two places the rule had slipped

Writing this up meant checking each rule against the code rather than against the intention, and two had come apart from it.

The reviews screen states that ratings stay separate by source even when every source uses a five-point scale, and the helper behind that screen honours it. The management reports screen, built later, averaged every rating it could find into one figure. One page asserted the rule and another printed the number that broke it.

The second was smaller and the same shape: a menu screen described itself as the comparison boundary for competitor pricing, while nothing anywhere joined a restaurant's own menu to a competitor's. Both have been corrected, and both now have a test that fails if the claim and the code part company again.

What this means when you read a figure here

• Several totals where you expected one usually means your data spans several currencies, and combining them is a decision for you rather than for the software.
• A short comparison list is the normal result. Strict matching discards more pairs than loose matching, and the pairs it keeps are the ones worth arguing from.
• An empty cell means the evidence for that figure does not exist yet, not that the value is zero.
• Any number that matters should name its own inputs. If it cannot, treat it as a claim rather than a measurement.