The record · The record

What was believed, and when.

What was published, when it was published, and what it replaced. Every earlier version stays on the record.

Refreshed every six hours Every figure links to its publisher Where it comes from →

The record, at a date you choose

Move the slider. Every figure below reverts to what was actually on the record that day, not to what is known now.

When a publisher restates a figure, this record keeps the earlier value beside the new one, each with the date it was issued and a link to where it was said. Nothing is overwritten, so the distance between a forecast and the one before it can be read off the page.

What changed in the record

A factual difference between two dates. No entry here describes a movement; each one states who said what, when, and what it replaced.

Between and

Take it with you

The site publishes a feed of its own changes. It reads 45 feeds from other people; publishing one is how it becomes an input to their reading rather than another destination.

Every entry above is also an Atom feed entry, and the same list generates a plain-text digest that can be sent as it stands. Nothing in either is selected for importance or described as good or bad. The list is everything that entered the record in the window, in date order, which is the only editorial position this project is able to take.

Generated by tools/gen_digest.py. The feed is the distribution argument for the whole idea: a competitor can copy the numbers, because they are public. A competitor cannot publish what changed, because they did not keep the earlier version.

Why this is the hard part

The constraint that makes the record publishable is the same one that makes it valuable.

A vendor that published this chart would be publishing its own error rate. Every incumbent sells the current forecast, so no incumbent can hold forecasters to account. Only a party with no forecast to defend can do this, which is why the rule here is absolute: publish the sequence, never the verdict. "Issued 3 December, restated 2 June, source" is a record. Adding a word about whether that was good is a rating, and forfeits the whole position.

Published by YINDOR. Every figure links to its publisher. Not investment advice, and no view is stated on any company or security anywhere on this site. This build is a working prototype. The exchange-rate file covers 2014-01-02 to 2026-09-29 and was last collected 2026-09-29; the vintage file was last collected 2026-09-16. Both dates are read from the files themselves, so a footer that has gone stale says so rather than repeating what was true at the time it was written.