Tools

The tools that give you one number

A single-number tool is the easiest to read and the hardest to learn anything from; what it keeps and what it throws away.

By Updated 4 min readTools

Guides on Tools: Every presentation choice on a result page, and what it does, A taxonomy of rating tools by what they output, A method for judging any rating tool before trusting it

A single-number rating tool gives you one figure, usually out of ten, and discards everything that produced it. That makes it the fastest kind of tool to read and share, and the hardest of the five kinds to learn from: it can say a score changed, never why.

What it is good for

A single number is fast to produce, fast to read, and fast to share, which is why it remains the most common shape a rating tool takes. It supports one comparison cleanly: is this submission higher or lower than that one, on this same tool, under roughly matched conditions. For a quick check - did this photo score better than the last one - a single number answers the question in the time it takes to glance at a screen, and nothing about that is a bad thing on its own.

What it discards

Everything that produced the number is invisible once it is compressed into one figure. If a resubmission moves from 6.4 to 7.1, a single-number tool cannot tell you whether that gain came from better light, a more flattering angle, or an entirely different judgement the tool happened to weight more heavily this time. A total compresses several unrelated judgements into one figure and discards which one moved - that sentence is the whole limitation of this type of tool, stated once and true of every result it will ever produce. Even published weights may not describe what drives a total: Paruolo, Saltelli and Saisana (2011) found that in several well-known composite indices, including the Human Development Index, the declared importance of components and their actual effect on the total were often very different.

It also discards any way to audit consistency component by component. With a breakdown, you can hold everything constant, change one variable, and check that the axis you expect to move actually moved and the others did not. With a single number, you can only check that the total moved - you can never check whether it moved for the reason you think.

Can it still be tested?

Yes, from the outside, even without a breakdown - a total-only tool can still be probed for consistency and sensitivity by submitting the same file twice, or a deliberately varied set, and watching how the single number responds. That workaround exists precisely because the tool itself will not tell you anything about its own internals; every insight has to be reconstructed from repeated inputs rather than read off a result page.

Where this type sits among the others

A single-number tool makes the weakest claim of the group that actually looks at an image: a value, with no stated components. A multi-axis tool makes a stronger claim by exposing the parts that built the total, at the cost of being harder to skim in one glance - Rate Cock's approach of showing six axes rather than a single figure is a direct answer to the exact limitation described above, trading a little speed for the ability to attribute a change to a cause. A pairwise tool makes a different, narrower claim again - order rather than value - and a single-number tool sits between that and a full breakdown in terms of how much it is willing to show its work.

When one number is enough

Not every use case needs a breakdown. A quick sanity check, a one-off curiosity, a submission with no intention of tracking change over time - a single number answers all of these adequately, because none of them requires attributing a result to a cause. The type becomes a liability specifically when someone tries to use it for something it cannot do: explaining why a score changed, or comparing it meaningfully against a different tool's number, where the absence of any visible structure leaves nothing to compare beyond the raw digit.

The same discarding happens outside AI scoring entirely. A tape measurement reduced to one figure still keeps its unit and its method alongside it if recorded properly, which is more than most single-number rating tools bother to show. A human review compressed to a single word or star rating loses its own version of the breakdown - tone, specificity, what was actually said - for the same reason: one number travels faster than the reasoning behind it, and travelling fast is usually the whole point of the design.

A single-number tool is not a worse tool for being simple. It is a tool that has made a specific trade, and the trade is worth knowing before reading too much into the one figure it hands back.

Read next

Full archive