Joe Nicora

SvelteKit and Skeleton on the frontend, Fastify on the API, MongoDB 7 underneath, all running under Docker Compose. Auth, RBAC, in-app messaging, audit logging, and admin settings ship by default. Dev overlay adds an optional Ollama-backed chat assistant.

A build-time module system clones the scaffold per project. arch.js create copies the template, then merges nav entries, route trees, permissions, dependencies, and env vars from each module manifest. Collisions are checked before any write. New projects start with the modules they need.

Permissions live in MongoDB as resource-action pairs. Both the Fastify routes and the Svelte frontend call hasPermission with the same shape. Manage Users and Manage Roles admin pages ship by default. Adding a resource means appending to a JSON file, not editing middleware.

Designed for Windows-first development with Docker Desktop on the WSL 2 backend. Bind mounts skip filesystem events, so API rebuilds stay explicit rather than implicit. Production runs the same compose file with the dev overlay stripped. Sessions, password reset email, and audit log work out of the box.

A markdown pipeline for analyzing human-AI conversations after the fact. Each turn gets a log entry recording concepts, terms, assertions, and seven categories of drift marker. The format is mechanical enough that two analysts working independently from the same transcript reach similar conclusions about what changed.

Drift markers cover vocabulary substitution, premature resolution, framework introduction, connective capture, confidence shift, scope addition, and elaborative expansion. Each has explicit detection criteria. Elaborative expansion specifically requires classifying solicitation status and reconstructing whether the AI presented decisions as choices the user could evaluate.

A conversation state log tracks the macro condition turn by turn: dominant vocabulary, dominant framework, confidence level, structural resolution, concept ownership. Transitions classify as stable, shift, or replacement. The state history surfaces direction changes that no individual message in the conversation explicitly announced.

A cross-session concept registry persists ideas across conversations. Each entry records origin, current form, sessions seen, adoption trajectory, and whether the concept was ever explicitly evaluated. AI-introduced concepts in active use that the user has never reviewed get flagged automatically before the next session begins.

Terheia resolves through a fixed pipeline. Cosmology constrains world. World constrains society. Society constrains infrastructure. Each layer filters what the next can produce. The system is generative rather than descriptive — instead of documenting the world, the canon defines the rules that produce it.

Five moons govern emotional baseline, ritual stabilization, and magical efficiency. Five recognized magic schools route through living substrate channels. A sixth — Psionic — exists in practice but not in the Accord taxonomy by institutional policy. Detection of its residue requires Hexe Bound attunement at the death boundary.

Three institutions divide knowledge: Accord controls research and classification, Crown enforces narrative and law, Reverie distributes culture and ecology. Eleven factions cut across them with overlapping moon orientations. Institutional failure proceeds through incomplete understanding rather than malice. Costs accumulate across narrative and do not reverse.

Objects pass through nine cascading filters: moon, biome, state, culture, faction, institution, wealth, ethnicity, material. A second pass routes through the Sensory Resonance Matrix — texture, sound, scent, emotional bias. The output is a chipped lacquer toy soldier with paint worn smooth by small hands.