Crate corpora_engine 

The bus orchestrator and the IO edges: Sources (where records come from), Parser (bytes → records), GraphBuilder, Writer, and the git-backed [RevisionOracle]. All filesystem/VCS access lives here; everything in corpora-core/-rules/-report stays pure.

Re-exports

pub use engine::Engine;

pub use graph_builder::GraphBuilder;

pub use oracle::GitOracle;

pub use parser::FrontMatterFormat;

pub use parser::Parser;

pub use reporter::FailureReporter;

pub use reporter::Gate;

pub use toml_format::TomlFormat;

pub use source::FsWalk;

pub use source::Source;

pub use source::SourceDriver;

pub use source::VecSource;

pub use writer::Writer;

Modules

engine

Single-threaded, deterministic FIFO bus. Each event is dispatched to every interested subscriber in registration order; their emissions are collected then appended, so the output is reproducible — the property CI and report --check rely on.

graph_builder

Accumulates Parsed records; on Settled builds the immutable graph and emits GraphBuilt. On Changed it replaces one entry — that’s the whole incremental story (full rebuild for now; see plan §7).

oracle

Git-backed RevisionOracle — the trait lives in corpora-core; this is the IO impl. It shells out to git (no heavy VCS dependency); resolution is git rev-parse.

parser

Bytes → Record. The front-matter format is pluggable; v0 ships a TomlFormat stub so the wiring is real even though parsing lands in plan Phase 2.

reporter

Terminal-side subscribers: Gate prints diagnostics as they flow past (the running tally lives in the engine), and FailureReporter turns ParseFailed into a warning so unparsed files are visible rather than silently dropped.

source

Plugin seam #0 — where records come from. SourceDriver adapts a Source onto the bus: scan on RunStarted, re-discover on Changed.

toml_format

Real TOML front-matter + markdown body parsing (plan Phase 2).

writer

Persists Report artifacts to disk under out_dir. The only place report bytes touch the filesystem — adapters themselves stay pure.