The Deep Dives Archive Now Shows the Research It Rests On

Our Deep Dives page is where the research behind our scoring lives. Until today it was a plain list of cards, and the cards were the problem: every one of them carried a tag reading Deep Dives, under a heading reading Deep Dives. Most carried a second tag reading Complete, which is the only value that field takes anywhere in the archive, so it separated nothing from nothing. The rest of each card was an internal experiment ID and a date.

The date was the part that misled. The list was ordered by the day we published each note, and it printed the day we last revised it, labelled Updated. Those two dates differ on every record in the archive, so a reader scrolling down a correctly ordered list saw dates jumping backwards and forwards and had no way to know why. Every visible date also carried that Updated label, so the page read as though nothing on it had ever been published and everything had changed last week.

What the page shows now. Most of these notes rest on one published, peer-reviewed study. Those records now carry that study’s citation on the card, in the ordinary academic short form, so anyone can go and read the paper without taking our word for anything. They are grouped by which of our four scoring dimensions the research supports, each group naming how many notes are in it and linking to the dimension it belongs to. The notes that rest on no single named study are grouped separately and say so. The notes themselves followed the same day: each one now closes with a citable record of its own, and where a single paper is behind it, that paper is named with its journal and year.

Two smaller things went with it. The date shown is now the date the list is ordered by, so the ordering is visible instead of hidden. That fix reached this page and stopped there: on the full lab list the same records went on printing their revision date, until the default was set the other way round across the lab. And the whole archive is on one page rather than split across seven, because an index is allowed to be long and seven pages behind a single Next button was never a good way to browse research. The old page addresses still work and send you to the archive.

The machine-readable half was empty and is not any more. The structured data for this page told search engines and language models that a collection lived here and then listed none of its members. It now lists every record on the page, in the order the page shows them, built from the same source the page itself renders from.