Editorial Standards

Every post on this blog is written with AI. I hold editorial responsibility for all of it.

"AI-assisted, human-edited" is what everybody writes and it commits you to nothing. Here's what mine means.

The argument is mine. AI drafts, researches, and audits. It doesn't decide what I think. Before anything gets written, a post has to name one reader and one decision that changes for them, and it has to say something a person who already follows three other AI newsletters wouldn't know. Ideas that clear one test and not the other end up as a LinkedIn post, or as nothing.

Numbers get flagged while they're being written. If a figure, a date, or a quantity isn't coming from a cited source or from work I did myself, it gets marked the moment it goes on the page. Before publishing, each one gets a real source or it comes out. This is the rule I'd most want you to know about. A confident number with nothing behind it is how AI writing fails, and it is very hard to spot after the fact.

Sources get traced, not repeated. I go to the original: the announcement, the paper, the filing. Not the write-up of the write-up. Statistics get traced back to whoever generated the number, because the usual problem with a stat in this industry is that everyone is citing a summary of a summary and nobody checked what was actually measured. I fetch links before they go in. Dead ones appear as plain text instead of as a link that 404s on you.

Then a pass that didn't write it reads it. Every draft goes to a separate review that sees only the post and the rules, with no memory of having written it. Whatever wrote the copy is the worst judge of it. That review flags problems and isn't allowed to fix them, and a clean report isn't permission to publish. It just means my read is a spot check instead of a hunt. Whether the story is true, and whether the post is any good, stay with me.

The watermark

Under EU AI Act transparency rules, Anthropic started marking Claude's output with an invisible machine-readable watermark on 2 August 2026. Models that shipped before that date are being folded in over a transition period, so whether a given post here carries a mark depends on which model drafted it and when. Assume they will.

Worth knowing what that does and doesn't tell you. The mark survives copying and some editing, so it's a durable signal that a model touched the text. It is not a statement about authorship. The same mark lands on text a model translated or tightened for a human author, and Anthropic says so itself. It tells you a tool was involved. It doesn't tell you whose thinking you're reading. That's what this page is for.

Who's responsible

I am. Melissa Kolbe, Pallas Advisory. If something here is wrong, it's mine to fix. mkolbe@pallasadvisory.com

These standards took effect 7 July 2026 and apply to posts published from that date forward. Last revised 8 September 2026.