| OKF (Open Knowledge Format) | An open specification for storing knowledge as a directory of Markdown files + YAML frontmatter (v0.1, Google Cloud, 2026) |
| Bundle | A directory containing all knowledge files — the unit of distribution (in this project: wiki/) |
| Concept | One unit of knowledge = one .md file |
| Concept ID | The file's path within the bundle with .md stripped, e.g. tables/orders.md → tables/orders |
| Frontmatter | A YAML block at the top of a file, delimited by ---, storing metadata |
| Body | The Markdown content below the frontmatter |
| Link | A Markdown link between concepts = a relationship (untyped) |
| Citation | A link from a concept to an external source that supports a claim in the content |
| Reserved file | Reserved files: index.md (table of contents), log.md (change log) |
| Progressive disclosure | Showing the table of contents before opening the actual file — reduces context window overflow |
| Conformance | A bundle following the v0.1 rules (parseable frontmatter + type present + reserved files correctly structured) |
| type | The only required frontmatter field — specifies the kind of concept |
| Reference | A type used for synthesized knowledge (joins, metric definitions), typically under references/ |
| RAG | Retrieval-Augmented Generation — fetching raw document chunks at query time and stuffing them into context |
| LLM-wiki pattern | A concept (Karpathy) where AI synthesizes knowledge into a continuously maintained Markdown wiki, instead of re-fetching raw sources every time |
| Ingest | The process of taking raw sources and synthesizing them into wiki concepts (should be human-supervised) |
| Contradiction flag | A > **CONTRADICTION FLAG**: ... marker added when new information conflicts with existing content |
| BM25 | A keyword-based ranking algorithm that scores document relevance |
| Embedding | A vector representing the semantic meaning of text (generated by a model, e.g. via Ollama) |
| Semantic search | Search by semantic similarity (cosine distance between embeddings) |
| Hybrid search | Combining BM25 + semantic search |
| RRF (Reciprocal Rank Fusion) | A method for merging ranked results from multiple signals: Σ 1/(k + rank) |
| MCP (Model Context Protocol) | An open standard for AI agents to connect to external tools and data |
| MCP server | A service that exposes tools (search/get/propose) wrapping a bundle for agent connections |
| PR-gated | A write model using branch + Pull Request + CI + review |
| Lease/lock | A write model that reserves exclusive rights to a concept with a TTL, preventing write conflicts |
| Lease | A TTL-based reservation that expires automatically, verified by a token |
| Curator | (Model 3) A single agent that aggregates proposals and merges them into the wiki |
| CODEOWNERS | A file defining owners per subtree (used to split responsibility across teams in a monorepo) |
| Federated bundles | Multiple repos/bundles per domain; MCP servers mount multiple bundles, namespaced by bundle name |
| air-gap | An environment with no internet connection (closed network) |
| Gitea / GitLab CE | Self-hostable git servers used as the source of truth |
| AGENTS.md | A schema file that tells agents about the structure/rules/workflow (may be named CLAUDE.md/GEMINI.md) |