How to read this site
A two-minute guide, then a glossary.
The idea
People with power say things about the future every day: inflation will fall, the deficit will shrink, this technology is safe, that one is dangerous. Usually nobody goes back to check. We write each statement down the day it is made, with a link to the official text, and go back when the date arrives.
Reading one entry
- The quote - the speaker's own words, short, with a link to the full official text.
- Can it be checked? Green means it names what, by when and how likely. Red means it is too vague to ever be proved right or wrong.
- The result - once the date has passed: came true, partly, or did not, with the official figure it was compared with.
- “Evidence & why” - opens the full record of that sentence: the sentences around it, every timestamp, the fingerprint of the text, an archived copy, and the reasons for each label.
Reading a report card
The headline is a plain count: how many forecasts have been checked and how many came true. Below it, each dot on the timeline is one forecast. Someone who speaks mostly in checkable terms and is mostly right earns trust; someone who speaks mostly in terms too vague to check cannot be wrong - and that is worth knowing too.
Reading the scoreboard
“Average miss: 0.9 points” means that when the institution said, for example, 2.0%, the result was on average 0.9 away - 1.1% or 2.9%. “Too low 70% of the time” means the misses lean one way.
Glossary
| Can be checked | The sentence names the event, a firm deadline, a likelihood and a measurable figure (4 of 4). Technical name: testable. |
|---|---|
| Partly checkable | It has 2 or 3 of those four elements. |
| Too vague to check | It has 0 or 1. Technical name: unresolvable as stated. |
| Forecast | A statement that something will happen, with a date, a likelihood or a figure. |
| Commitment | The speaker says what they or their institution will do (“we will…”). Technical name: strategic announcement. |
| Scenario | A conditional statement: “if X happens, then Y”. |
| Based on data / educated guess / stated on authority | How the speaker claims to know: by pointing to figures; by reasoning to a conjecture; or by simply asserting it. Technical names: empirical, abductive, oracular axis. |
| Ambiguity (0-100) | How many different readings the sentence allows. Hedging words (may, could, gradually) and missing elements raise it. |
| Likelihood | How sure the speaker sounded, as a number. “Will” counts as 85%, “likely” as 70%, “may” as 40% - the full table is in the methodology. |
| Due - being checked | The deadline has passed; the official figure is being looked up. The verdict is recorded automatically once the figure is published. |
| Brier score | A standard measure of forecast accuracy: 0 is perfect, 0.25 is what you get by always saying 50%. Shown after 10 checked forecasts. |
| Change of direction | The same speaker saying the opposite on the same subject within six months, found by a fixed rule. |
| Goalpost move | A forecast later reinterpreted by its author to fit what happened. |
| Seal | The daily fingerprint of the whole record, which makes later changes detectable. |