Where the thinking happens in paragraphs instead of products. Long-form essays on cognitive engineering, the Garden framework, and the lineage of Gardeners -- plus the serialized chapters of the Cognitive Engineering treatise as they ship.
The first hERG QSAR paper went public 2026-05-13. DOI 10.26434/chemrxiv.15003259/v1. Two-headed contribution: a within-family ground-truth dataset on iboga alkaloids, plus a portable architecture-failure protocol for QSAR audits.
Read →Catalog v2 was supposed to retract from 13 verified mono-monostatic bodies to 1. A second self-doubt pass surfaced the regime-structure finding: the verified set is an open regime in three k-families, not a single point.
Read →A physics interview surfaced Paul Schatz and the oloid. Two weeks later: four arXiv papers, a Stachel endorsement, and a 25.6x thermal advantage that rewrote its own headline mid-experiment.
Read →Chapter I -- “Prompts are software, and here's why that matters” -- is the inaugural long-form piece for the planned twelve-chapter treatise below. It will ship here first and serialize outward from there. Check back or see what I'm working on right now for updates.
Prompts are software. Why that framing unlocks everything that follows, and why the current state of "prompt engineering" is closer to superstition than to engineering.
Tokens, attention, RLHF, modes of failure. What the substrate you're engineering on top of actually is, mechanistically, so you can design for its real properties.
Prompt, brain, blueprint, circuit -- abstracted from Brainboot's branding. The taxonomy of composable cognitive work, and why each tier exists.
Why putting Zod schemas on the edges of a brain changes its failure shape. How type contracts make cognition testable in the way functions are testable.
Every brain has properties that must hold. Building them in as enforced runtime checks instead of hoping for them is the difference between software and wishful thinking.
Why a brain graph is not just a chained prompt. How to design sub-brain handoffs, budget propagation, and error surfaces across compositions.
The model cascade, prompt caching, batch economics. Why the pricing decision comes before the architecture decision, not after.
How to test a function whose output is non-deterministic but whose properties are not. The role of property-based tests in the cognitive engineering stack.
Why every commercial AI product must grow its user, and how that rule makes itself visible in the architecture when you honor it.
Open research. Honest limits. Where the framework runs out of traction and what would have to change upstream in the model layer for it to stop running out.
The honest comparison to agents, LangChain, prompt marketplaces, and every other existing thing the framework relates to. What this is, what it isn't, what it subsumes, what it doesn't.
The 3-year vision. The creator economy for brain authors, the Brainboot Index, and the question of what happens to software engineering when a non-trivial fraction of it is cognitive engineering.