How this works
Strata is a publication experiment with one bet: that a blog post should be a document that keeps changing, and that the reader should be able to see the changing. Everything below follows from that.
Posts carry their revision history
Publishing here is not a one-time event. When a post is corrected, a new revision is recorded with a required description of what changed, and readers can read any earlier version in place with the edits shown against the previous one. A correction is not an embarrassment to be quietly patched; it is the most interesting thing a long-lived post can contain.
If you have read a post before, you will be told what changed and offered the diff. You can also subscribe to a post and get an email only when a substantial revision lands — a notification that fires because something genuinely changed, not because a feed wanted engagement.
You set the depth you read at
Most technical writing picks one altitude and alienates half its audience: explain the prerequisites and the experts bail, skip them and the beginners drown. Here, blocks are tagged in layers, and you choose.
- Skim
- The argument only. Usually about ninety seconds.
- Read
- The full argument, with prerequisites and context surfaced inline the first time you need them.
- Study
- Everything. Footnotes, source citations, appendices, and the author’s private margin notes.
Your choice persists across every post and both themes. The page is fully rendered and readable with JavaScript switched off — the dial only ever changes what is already there.
Argument happens in the margin
Select any sentence and you can attach a note to it, typed as a comment, a correction, an extension, a disagreement, a worked example, or an update. Notes live beside the sentence they are about rather than three thousand pixels below it, where they rot.
Two details make this durable. A note stores the quoted text with its surrounding context, so it can find its sentence again after an edit rather than silently reattaching to different prose. And every note is pinned to the revision it was written against — if the sentence changes, the note says so and links to the version it was arguing with.
Authors can leave notes of their own — the reasoning behind a decision, including the parts they got wrong. When an author accepts a correction, it is folded into the post’s history as an attributed amendment, and the person who wrote it becomes a co-author of the record.
Writing here is not one event
Posts have a lifecycle, and it is shown rather than hidden, so you can weigh a claim appropriately.
- Seedling — an early thought, published before it is finished, on purpose.
- Budding — an argument forming, with references attached to its claims.
- Evergreen — actively maintained, with a named maintainer and a review cadence.
Right now: 3 evergreen, 2 budding, 1 growing. A post that has not been reviewed in a long time says so. Staleness is a nudge to the author, not a badge of shame.
What we refuse to do
Ads. Likes. Follower counts. Infinite scroll. Trending. Engagement bait. Auto-generated filler. A ranking you cannot inspect. Dark patterns and fake urgency. There are no notifications that fire for engagement’s sake, and nothing on this site exists to make you stay longer than you want to.
If we cannot explain why a piece of content surfaced, it does not surface. And the proof of craft we ask of writers is not a follower count — it is a visible record of having been wrong and corrected in public.
Who writes here
Mira Okonkwo
SRE. Maintains more services than she would like. Interested in what pages look like at 3am.
Sam Reyes
Building eval harnesses. Suspicious of dashboards that only go up and to the right.
Vedant
Works on retrieval infrastructure. Writes about the parts of ML that turn out to be distributed systems problems.