← All work

Genome

Research: can agent identity be encoded as inspectable substrate?

THE SUBSTRATE — DETERMINISTIC COMPOSITION THE WALL: SYSTEM CANNOT EVALUATE MEANING OUTSIDE THE SYSTEM Declared InputsHuman's declared actsfrozen · never query-derivedINSIDE THE SUBSTRATE GenesTyped declarative datadeterministic · no logicComposerLattice-meet compositionpure function · cannot read queryConsumerObserves external evidencerestricted signature · no queryExternalEvidenceretrieval · tool returnsSource-Discrimination ContractFour-check validator · runs in CI, not runtimeno part of the system derives behavior from the query's semanticsdeclared actsgene libraryeffective tierevidenceescalate / generateCI checks ↑

Key boundary: The substrate executes declared resolutions; it does not generate them. Source-discrimination is enforced at four syntactic sites that share one mechanism — the system can see what a value's expression draws from, but it cannot evaluate what a value means. The boundary is the same wall reached from three directions: query-derivation, sibling-internals, and gene-function. The system cannot evaluate meaning.

Why it matters: A substrate that pretended to resolve intent would be smuggling model-resolution into what claims to be declared structure — the category error the architecture was built to refuse. The boundary, named precisely, is what makes structural composability honest.

Notable decision: The boundary is not a caveat to the result — it is the result. The arc set out to find emergent synergy and found a precisely-bounded structural substrate instead. Honesty about where structural depth stops is the foundational commitment.

A completed research arc investigating whether an agent's identity and behavior can be encoded as composable, inspectable substrate rather than buried in opaque weights. It produced an honest split result — one hypothesis affirmed, one negative with a clear mechanism for why — and a methodological lesson I keep returning to: trace the measure, not just the mechanism. The negative result is the part I'm proudest of; it's the one that taught me something.

The central finding is the boundary: the substrate is load-bearing for conflict detection and the generic machinery of resolution; it is not load-bearing for resolution-of-intent, which lives in a human's declared act outside the system. The boundary is not a caveat to the result — it is the result. The arc set out to find emergent synergy and found a precisely-bounded structural substrate instead. Honesty about where structural depth stops is the foundational commitment.