LOGOS

NEXUS knowledge repo — curated docs, immutable event store, tooling, external snapshots, and per-project working notes, all in one place.

LOGOS records what's been learned across the ecosystem as a mix of file-authored conventions (CORPUS/) and immutable TOML events (CHRONICON/). It is the source of truth for the NEXUS-level AGENTS.md that every repo's CLAUDE.md / AGENTS.md is regenerated from.

Structure

• AGENTS.source.md -- master source for every NEXUS repo's AGENTS.md / CLAUDE.md (regen pipeline output).
• CORPUS/ -- curated, file-authored content.
  • conventions/ -- cross-cutting standards (the NEXUS rulebook).
  • methods/ -- methodologies / patterns / approaches.
  • audits/ -- narrative bodies for Audit events (referenced from each Audit event via body_uri).
• ARMAMENTARIUM/ -- build / regen / git-hook / API-inspection scripts. See ARMAMENTARIUM/README.md.
• ARKHEION/ -- external snapshots (Unicode data, upstream reference files, historical CLAUDE.md captures, etc.).
• CHRONICON/ -- immutable TOML events at top level (flat layout). The type = field on each event carries its kind: Decision, Observation, Question, Finding, Reference, Correction, Audit, Experiment. Filename is <uuidv7>.toml; no slug, no prefix, no per-type subdir.
• HYPOMNEMATA/ -- project working notes, organized as <project>/<note>.md. The shared field-notebook for in-flight thinking that hasn't (yet) become a formalized event.
• CHRONOS/ -- temporal indexes (lazy; populated as the corpus grows).
• CODECS/ -- serializers / readers / schema migrations.

Single-repo aggregation

Per-project <project>/LOGOS/ directories are retired. All events, notes, and curated content live in this one repo. Events capture their origin in TOML provenance (origin_project, origin_machine, origin_agent, recorded_at, git_sha) so the trail isn't lost.

A specialized machine can clone just this repo and have the full knowledge corpus without needing any project source code.

Event schema

Each event TOML uses a three-section structure:

• [provenance] -- where / who / when (origin project, machine, agent, timestamp, git context).
• [meta] -- how to classify and relate (tags, references, supersedes, confidence).
• [content] -- the event body (inline markdown via body = '''...''' or body_uri to a colocated file -- the Audit pattern uses body_uri pointing into CORPUS/audits/).

See /NEXUS/LOGOS/CORPUS/methods/event-sourcing-as-knowledge-store.md for the full methodology.

Querying

Knowledge about a topic (fsharp, tomlyn, fun-blazor, ...) is read by scanning CHRONICON/*.toml for events whose tags include that topic and synthesizing the contents in-line. There is no projected view layer right now -- a simpler query surface (potentially DuckDB-backed) is on the roadmap; until then, the read path is manual TOML scan.