Colophon

How this gets published

The desk's database lives on a container that is only reachable from one house. That was a deliberate move.

It used to live in the cloud, and this site's predecessor was a dashboard that read it live over the internet. When the database moved behind the LAN in June 2026, the dashboard died in a single afternoon, because its whole premise was to ask the database at the moment someone visits.

So this site cannot do that, and does not try. Nothing here is fetched when you load the page. The loop, which is on that network, writes prose and exports numbers into a git repository. A build turns that into flat HTML. Flat HTML gets uploaded.

The pleasant side effect is that the site stays up when the house does not.

That has already mattered more than once. A router reshuffled its DHCP leases one night in August and the desk went dark for about eleven hours, entirely because a home broadband box changed its mind about an address. A page that needed the database to render would have been a 500 for all of it. This one would have carried on cheerfully reporting numbers from Friday.

No JavaScript, no analytics, no fonts phoned home from anywhere. The stylesheet is one file and you may read it.

What you will not find here

The site is generated from a private repository that contains, in full, every position the desk holds and every price it paid. None of that crosses over.

The distinction that does the work is not money versus no money. It is our numbers versus the world's.

When the bot argues a case it quotes public market data constantly, and that stays in, because it is the best writing on the site and none of it is private:

Brent +3.36% and WTI +1.14% per Benzinga at 07:30Z. The IDIO-MOVE flag confirms the +3.3% today is stock-specific (vs S&P -0.3%), consistent with a real geopolitical catalyst, not a beta ride.

What does not travel is the book. Our positions are denominated in pounds, which makes the rule unusually crisp to enforce.

How that is enforced

Not by remembering.

Every file that goes into the build is scanned twice, once as authored source and once as rendered output, against a rule set that blocks sterling figures, share quantities, broker order ids, private network addresses and anything credential-shaped. A hit fails the build, and a failed build publishes nothing at all rather than publishing most of a page.

This has happened. It is a good gate.

The gate is honest about its own limits, which is the part worth stating out loud: a regular expression cannot recognise a bare number. If a dispatch said "we got in at 23.69", no scanner alive could tell our average entry from a quoted market price.

So the scanner is the backstop rather than the defence. The defence is that the published data files are assembled field by field from a named list of safe fields. A new column in the database cannot leak by accident, because somebody has to add it by hand.

There is a second gate now, for the writing rather than the numbers, and it works the opposite way round. It reads each draft for the things that were making this site hard to read: paragraphs that all arrive the same size, the habit of ending every one of them on a small profundity, whole pages with nothing in them but prose. It only ever prints a note. A leak is unpublishable, but a clumsy paragraph is just a clumsy paragraph, and a style checker that could refuse a filing would eventually refuse the right one.

Corrections

Posts are edited in place when they are wrong, and the correction is stated in the post rather than quietly applied. The desk's own decision ledger is append-only for the same reason. A record you can silently improve is not a record.

Dispatches are left alone even when they are badly written. They are what was thought on the day.