Working notes, specifications and whitepapers behind the systems. Written because the design decision was hard enough to be worth arguing on paper before it was worth writing in Go.
AI agent frameworks treat personality and reasoning style as prompt engineering problems. The agent is a system prompt, a set of tools, and a loop. This paper argues that is a dead end for production systems and presents The Human Model, a compositional architecture that decomposes agents into seven structured primitives drawn from cognitive science and personality psychology: Skills, Traits, Behaviors, Cognitive Styles, Communication Styles, Perception, and Personas. We describe Cortex, an open-source Go implementation with multi-tenancy, execution tracking, and human-in-the-loop checkpoints. A survey of cognitive architectures (SOAR, ACT-R, BDI), personality modeling research, and contemporary LLM frameworks shows that no existing system covers all seven dimensions. The Human Model fills that gap by applying validated psychological constructs as engineering primitives rather than prompt decorations.
CortexAppend-only hash-linked logs and a legal right to erasure are incompatible in their naive forms. Per-subject encryption resolves the conflict, and the resulting guarantee is bounded by key destruction, surviving metadata and hash algorithm lifetime, each of which is stated explicitly.
ChronicleVaultKeysmithPortable object-relational mappers express queries in the intersection of what their backends support, and applications escape that subset at the first real query. Measures the cost of eliminating query-time reflection on a single insert microbenchmark, and argues the module boundary belongs below query construction.
GroveFabriqA subscription over a filtered, sorted, limited query can return exact enter, leave, move and update deltas if the in-engine window is kept an exact prefix of the source-ordered result. Gives the cushion and keyset refill construction, its correctness argument, and the sharded case it does not yet cover.
FabriqGroveArgues that long-running agents degrade through attention dilution rather than capacity exhaustion, which would explain why a larger context window does not help. Separates working memory, summarised history and graphed claims, and sets out the measurement that would settle the claim.
CortexWeaveAn ordering argument for content safety: detectors sorted by cost and precision, with an admissibility condition governing early termination. Argues the primary benefit is single-cause attribution rather than latency, and reports the replicated finding that deny lists outperform classifiers.
ShieldSentinelCortexA construction for compressing an agent episode into typed claims that retain the observations they rest on, written as graph edges a later contradiction can locate and revise. Bounds revision cost by the transitive support closure and is explicit about where the independence assumption fails.
CortexHow a service declares its own API surface across seven schema formats, with health and capability hints, so that a gateway derives its routing table as a pure function of the manifests it can see. Includes conformance requirements and the costs of putting discovery on the routing path. Two implementations consume it, though both share one author.
FARPOctopusBastionState, a versioned event and an outbox row committed in one transaction, then relayed to derived projections. States precisely what the construction guarantees and what it does not, and compares bounded divergence detection against the unbounded exposure of handler-side dual writes, analytically rather than by measurement.
FabriqGroveSupervised failure prediction assumes a labelled dataset that maintained industrial assets do not produce. Characterises four distinct sources of label corruption, and sets out three practical responses and the validation evidence each one can actually support, ordered by experience rather than by measurement.
FabriqSentinelEquipment hierarchies are trees and plants are not. Shared headers, site utilities and shared spares each break the tree, and the standard responses either duplicate nodes or misrepresent connectivity. Proposes retaining the hierarchy as one edge type inside a multi-relational graph.
FabriqDeriving node identity from a normalised natural key rather than allocating it makes multi-source ingestion idempotent, order independent and convergent. Analyses the two places it fails, and argues the error asymmetry requires strictness by default.
FabriqWhat it costs to keep an industrial digital twin honest when observations arrive late, duplicated and out of order. Derives a per-channel reorder window from measured lateness, argues that staleness must be a value in the type rather than an absence, and states the cases where interpolation is a defect. The lateness measurements themselves are not reproduced.
FabriqGrovePolyglot storage abstractions either reduce to what every backend supports or fail at runtime when a driver cannot honour a call. Opt-in capability interfaces let calling code degrade deliberately, and the same argument applies to ORMs and object stores.
TroveGroveTwo abandoned Rust frameworks, examined for the same failure: an abstraction that re-exports its decisions as type parameters has not hidden them. Proposes counting empty interface implementations as a measurable design signal.
Ultimate (Rust)NorthForgeCompositing a game engine into a reactive widget tree involves three independent lifecycles that no party coordinates. A survey of the failure modes, the ownership model that resolves them, and the cost of a second renderer measured on one class of mobile hardware.
Flutter Unity WidgetGameFramework