Why Horizon will never score a builder

The question comes up constantly, and it is a fair one. Horizon holds the whole public record on more than a hundred and twenty thousand projects. It would be trivial to reduce a builder to a letter grade, or a project to a number out of ten. Every competitor either does this or is asked why they don't.

The answer is that a score would be the least honest thing on the site.

A score launders an opinion into a fact

The inputs to any rating are filed facts. The weights are not. Deciding that a lapsed registration costs eight points and a missing escrow declaration costs twelve is a judgement someone made, and no amount of arithmetic downstream turns it back into a fact. What arrives on the page looks like a measurement and is actually an opinion wearing a number's clothes.

The person reading it cannot see the weights, cannot argue with them, and — this is the part that matters — will treat the output as more reliable than the prose it replaced, precisely because it is numeric.

The data is too uneven to rank across

There is a second, more concrete problem. The twenty-two registers do not publish the same fields. Some carry a declared cost; most do not. Four publish complaints; eighteen publish nothing. A rating built on what is present would systematically score projects in thin registers differently from identical projects in rich ones — not because the projects differ, but because the portals do. The disclosure study is the full measure of that unevenness, and it is severe.

What we do instead

Completeness, and the difference between silence and a fact. Every section on a record is either a filed fact with the authority named, or an explicit statement that the register holds nothing there. The two are never collapsed. A buyer who reads a Horizon page comes away knowing what is on the record, what is not, and who to ask about the gap — which is a better position than a grade would leave them in.

Where the whole corpus can say something a single record cannot, we count it and publish the method — and even there, nothing is ranked as better or worse, only ordered by size, with the page saying so in as many words.

The line, stated once

• Nothing guessed. Every fact traces to a filing. A question we cannot answer says so.
• No prediction. We describe what was filed, not what will happen.
• No verdict. We publish the record; we do not score the thing.

These are not marketing lines. They are the constraint that decides what gets built here, and they have killed features that would have tested well.