Scope and Non-Goals

bijux-canon-reason owns inspectable reasoning records. It turns a declared problem and evidence into a content-addressed plan, typed events, supported claims, verification findings, and a manifested run that can be reviewed and replayed from retained inputs.

flowchart LR
    spec["problem + constraints"]
    evidence["pinned or retrieved evidence"]
    reason["plan, execute, claim, verify"]
    run["manifested reasoning run"]
    orchestration["agent and runtime policy"]

    spec --> reason
    evidence --> reason --> run --> orchestration

In scope

• immutable problem, plan, claim, trace, evidence, support, and verification models with content-derived identity;
• deterministic planning over understand, gather, derive, verify, and finalize actions;
• runtime and tool protocols, ordered event recording, fail-fast execution, and explicit insufficient-evidence outcomes;
• local pinned-corpus BM25 retrieval and extractive reasoning as a reproducible reference path;
• exact support spans, snippet hashes, evidence provenance, derived grounding, structural verification, and registered check results;
• canonical JSON/JSONL, trace fingerprints, invariant checksums, manifests, frozen replay, and structural diffs;
• CLI, Python, HTTP run, verification, replay, evaluation, and artifact inspection surfaces.

Non-goals

Not owned here	Owning boundary
Preparing source text or defining chunk representation	bijux-canon-ingest
Governing vector backend selection, ANN bounds, or retrieval replay	bijux-canon-index
Choosing which reasoning role runs next or when a workflow converges	bijux-canon-agent
Accepting a complete tenant flow or governing end-to-end replay	bijux-canon-runtime
Proving a source is authoritative, current, complete, or true	domain review and source governance
Automatically calibrating claim confidence	caller or domain-specific calibration process
Distributed retrieval, scheduling, sandboxing, quotas, or secret management	hosting system

Verification boundary

Verification proves that registered structural, linkage, provenance, hash, support, tool, and replay rules passed over the retained record. It cannot find all omitted evidence, expose every unstated assumption, or turn faithfully hashed false content into truth.

The bundled extractive reasoner and BM25 retriever are reference components. They demonstrate the artifact and replay contract; they are not claims of general reasoning ability or state-of-the-art retrieval.

Scope test

A change belongs here when it changes the inspectable relationship among a problem, plan, evidence, tool result, claim, finding, or reasoning-run digest. If it changes retrieval ranking rather than support meaning, workflow order rather than plan semantics, or run acceptance rather than verification facts, it belongs in an adjacent package.

See the capability map and known limitations for the precise implemented and epistemic boundaries.