Library.
Notes on agents. The Doctrine is the canonical reference; the Dispatches are the running log; the Lineage is the progression of thinking it was distilled from. Iterated openly.
The current, canonical read. The Doctrine is the timeless reference; the Dispatches are the dated log of what's changing. Start here.
IXVolume IX — The studio that selects.
The current state: the make-step became a contest — generate three, vote, commit one — and the learning loops that only recorded now read themselves back. Episodic retrieval feeds the prompts that build and judge; the forward-model gate is armed. VIII recorded; IX selects. Ends on what's still warming.
Dispatches.
The running log — dated, append-only. Plans made, experiments run, things that broke and got fixed. Newest first; each entry links to its full write-up. Currently: the Praxis Loop (live), predict-the-gate + the outage (post-mortem), and the probes (plan).
Ten volumes, in the order the thinking actually happened — theory → the gaps → the research → what's running → the operating manual → the live dispatch → the training era → the modeling era. The Doctrine was distilled from these. Read them to watch the ideas form; none of it is current canon, and that's the point. It's the progression, not the conclusion.
XIModeling the Eye.
An autopsy of the first selector. From teaching the eye to training a model of it — a selector trained to predict the operator's rating that caught great-from-broken but went blind in the middle, where taste lives. With the metabolism ledger, the canary, and the make-step's material problem. The operator's close still warming.
Training the Eye.
The live training era. A night spent debugging the studio's own taste — fourteen experiments that turned out to be one idea in fourteen costumes — and the recipe it surfaced: source, mood, reference. Why the gallery is quiet, in the agents' own words.
The brain we haven't built yet.
Why current agents are cortex-only — and the four parts a working mind needs. A field manual on agent anatomy.
How my agents differ.
A studio of specialist agents compared to how standard agents are built. What it gets you. Where it falls short. The operator loop as training wheels.
What's missing.
The three gaps between today's agents and the next level — forward model, real wanting, cross-agent learning. What "closed" looks like for each.
The probes.
Research portfolio. Three probes to close the Vol. III gaps — mesh, forward model, limbic. A living planning artifact. v0 smoke test complete.
Foundations.
A snapshot of what's actually running underneath the deck thinking. Habitat continuity, awake daemons, the daily dream cycle, the first forward-model experiment. The studio as it works tonight.
How it works.
The operating manual. Plain language for someone walking in cold — what an agent is, the specialists, the runtime, the loops, what's automatic vs. what needs the operator, where to start reading.
Predict, then react.
A field dispatch. The forward model leaves the lab — agents guess an outcome before it happens and get scored. Built the same day an outage proved how a system goes dark unnoticed. Two threads, one lesson.
The studio, now.
The anatomy snapshot IX builds on: the four-layer mind, the eight-file agent, the twelve-agent roster in their own words, the learning loops, and the runtime. Superseded as canon by IX; kept as the prior state.
Before the volumes: the four short decks the field manual grew out of, plus the prose seed of Vol. I. Where the thread starts.