The Archive of What We Turned Down Now Reads Like One

Our Lab Results archive has two doors. Last month the first one, the products that cleared verification, became a gallery. The second one, the products that did not, stayed what it had always been: a list of records, each opening with an internal experiment code and closing with the one interesting thing about it set in a grey monospace box.

The best writing on this site was being displayed as though it were machine output.

Why this half could not become a gallery

A product we turn down is never built, so it has no page, no photograph and no product record. We checked all of them on three separate fields and not one carries an image anywhere. There is nothing to show.

What there is instead is a sentence per record, written for that object and no other, saying what fell short. Those sentences are almost all different from each other, and they are now the widest and plainest thing on every row.

What the row says now

A headline, the reason in ordinary type, and a quiet line naming the maker, the category and the date. Beside it, a small stamp saying which of our gates the product failed, in words.

The score appears only where the score is why it failed. Most of our rejections carry a recorded score, but a good number of them were turned down on trust or on confidence rather than on the number, and printing a score beside those would tell a reader the product just missed when that is not what happened.

Gone from the row: the internal experiment code, a stamp repeating the heading, and a metadata line that stopped mid sentence on a third of the records.

How close they came

The page now opens with the distribution: how many of our rejections sat within a twentieth of a point of the bar, how many within a tenth, and so on out. The figures come from the same place The Bar draws its own wall from, so the two pages cannot disagree, and the strip states how many records it counts and how many it does not.

Two ways to narrow it

By category, and by which gate the product failed. There is also a sort that ranks by how close each one came. Records with no recorded score sort last and are never hidden: an archive that drops records when you change the sort is lying about its own size.

The list also pages at twenty four instead of ten, and it now tells machines which records it is showing, which it never did before.