Lexicon
The terms the essays are built from. Each is defined here and linked to where it is actually argued.
English is the source for these terms. The Korean and Japanese essays gloss each one on first use.
Compilation thesis full page →
Compilation is the structural operation that turns higher-level specification into lower-level artifact. The translation is directional, lossy, and not reversible. Where the domain supplies an independent verifier that can check whether the artifact answers the specification, compilation can be scaled by a training loop that compounds with compute. Where no such verifier exists, compilation still happens through human work but does not compound, and AI capability gains stall regardless of model scale.
In Pixels → Protocols · (note to self) From Punch Cards to Python, From Python to Agents · The Taste Compiler · Compiled Away · The Architecture of Discovery · Why code compiles and law doesn't · The Substrate Inversion · Cognition went public · The coupled recursion
Evaluative vs generative judgment full page →
Recognizing that an output is right is cheaper than producing the right output, for most artifacts that matter. This is the asymmetry every reviewer relies on. The two judgments are not interchangeable. They develop through different practice, train different cognition, and stand on different sides of the workflows AI is collapsing. The asymmetry is the engineering target for the discovery system current LLMs do not yet form, and the structural reason senior judgment is staying scarce while output is becoming cheap.
In By Logic We Prove; By Intuition We Discover · The New Information Supply Chain · The Inference Lead in the LLM Era · The Missing Taste Model · The Taste Compiler · Compiled Away · The Architecture of Discovery
The linker
Constraint propagation across separately compiled AI artifacts. Each domain compiler drops different dimensions, and the missing dimensions are exactly what cross-system consistency requires. The linker is the system that reconciles the outputs after the fact, holds the cross-system constraints, and refuses artifacts that violate them. It is not orchestration, not RAG, and not a multi-agent framework. It is the constraint-propagation-and-rejection step downstream of all of them. Plausibly the most consequential missing piece in current AI infrastructure.
In Pixels → Protocols · Compiled Away · The Substrate Inversion
Ghost GDP
A term from Citrini Research for AI-driven economic output that inflates national accounts but never circulates through the consumer economy, because the machines producing it do not spend. The cognitive version, borrowed and redirected in Compiled Away, applies the same shape to thinking. Surface intelligence keeps showing up; the developmental pipeline underneath is drying up. The risk lives downstream of capability, not in capability itself. It is a metric problem and a pipeline problem, not an argument against AI production.
In The New Information Supply Chain · Compiled Away · The coupled recursion
Four engine model of discovery
Discovery runs on four cognitive engines: prediction (narrows the space of moves), taste (prunes without verbalization), search (rolls out consequences), and imagination (simulates in compressed representations that never pass through language). The proof system uses prediction and search and externalizes through language; current large language models are proof engines. The discovery system depends more heavily on taste and imagination, both of which operate in latent space. The frontier of what an AI can discover is exactly where taste runs out.
In By Logic We Prove; By Intuition We Discover · The Architecture of Discovery
Projection gap
The loss when high-dimensional representation projects into a narrower channel. The bottleneck is not the representation, which is wider than the channel. The bottleneck is the projection step. The gap operates inside a single model when latent thought emits as serial language, and between a human and an interface that has fewer degrees of freedom than cognition. A third site, between separately compiled AI artifacts whose missing dimensions were what cross-system consistency required, is the framing's extension into cross-artifact infrastructure. Structurally different from the representation-capacity framing it is often confused with.
In By Logic We Prove; By Intuition We Discover · Compiled Away
Spec gap
The distance between what you meant and what you managed to specify. Every output from an agent is a mixture of intent and inference, and the two are interleaved and unmarked. The gap is visible whenever an agent infers what you could have stated, infers wrong, and only you would have caught the wrongness. The gap has two layers. One shrinks each time externalization is forced. The other resists externalization as a structural feature of having lived a particular life in a particular body.
QAKD
Quality-Adjusted Knowledge Diffusion. For a given knowledge event such as a paper, model launch, benchmark update, or vulnerability disclosure, measure how much quality-weighted knowledge becomes available to each language community over time and identify the causal mechanisms behind the gaps.
Synthesizability boundary
The line between artifacts a verifier can certify from a specification and artifacts that resist any such certification. The line is task-level, not domain-level. Legal work has both sides. Medical work has both sides. The boundary moves over time, one verifier at a time, and the engineering moves that push it (decomposition, proxy construction, synthetic ground truth) are different from the engineering moves that train better generators.
Externalized taste
Critique is the institutional externalization, at scale, of the evaluative content Polanyi called tacit. Writing about specific artifacts prompts evaluative content that abstract introspection cannot produce. The accumulated reservoir is structurally different from solicited preference data, hand-codified principles, and model-generated rationales, and encodes evaluative judgment in a shape no other source supplies.
Not yet referenced in a published essay.