MEV - State & Graph Compiler for the Company Brain

MEV started as a Markdown validator and became the write half of the company brain. It reads a corpus of Markdown and MDX spread across nineteen repositories, validates it along five independent axes — OKF frontmatter structure, `[[wikilink]]` and file reference integrity, the `scope:doc_id` knowledge graph, each repo's planning-state schema, and bidirectional `index.md` coverage — and then compiles what it read into derived views that other tools consume. The compiler half is the interesting part. `emit-state` is a fixed-point compiler: it resolves cross-repo planning state — block statuses, dependency edges, epic membership — until the output stops changing, then splices generated tables back into the Markdown documents that describe them. That means a roadmap's sequence table is not hand-maintained prose that drifts; it is compiled output, regenerated from the same `state.json` files the boards read. `emit-graph` and `emit-block-graph` emit the knowledge graph and the corpus-wide block-dependency graph as JSON artifacts for downstream consumers. Because the output is machine-readable (`--json` on every validation path, with a non-zero exit on any error-severity diagnostic), MEV works as a gate rather than a report. The RAG indexer refuses to rebuild against a corpus that fails validation, and the push hook blocks on errors new since the last successful push — so a broken cross-reference is caught at the moment it is introduced rather than discovered later by a reader. The design constraint throughout is that a validator nobody trusts gets disabled. Every check is scoped so it reports only what it can prove, and the diagnostics name the file, the line, and the rule.