Blog

Here are the close readings. If the main article describes the system as a whole and in broad strokes, the blog goes down into detail: individual principles, the incidents that gave rise to them, and the working stances we use almost every day. The register of all posts is one: descriptive and calibrated — working stances marked as stances, hypotheses as hypotheses, each post with provenance and a section of honest boundaries.

This is not a showcase of finished tools. It is an account of principled conditions — of what we actually run into at work.

Posts

• The dyad: the unit that produces the result — why the result is produced by the "human + model" pairing rather than either side, and what follows for everything else.

• A tool that does what isn't its job: why a skill is an open system — part of a skill's correctness lives not in the body of the procedure but on its boundaries with the world; the account passed an internal adversarial check across nine of our skills.

• Memory that discards nothing — don't discard, load everything, forge the key at write time, the order of layers, injective provenance: why the reliability of the work rests first on the reliability of the data.

• Methodology is not a by-product — it is the product — in research the how outweighs the what: data without methodology is not reproducible. Plus the apparatus of self-knowledge.

• When it's all to spec, and it doesn't work — topological and functional integrity, the asymmetry between them, and the "tower of standards": where the most expensive, invisible failures live.

• Silicon psychology — why study the behavior of the "model + human" system as a whole. The most speculative post: more hypotheses than stances, and they are marked.

The blog is level L2 of our information architecture. Level L1 is the whole-system description; level L3 is formal publications.