Project Memory Structure

Boundline uses two repo-visible document roots to keep reusable inputs and durable outputs separate from runtime state:

• docs/project/ holds stable project memory that planning and governed delivery can reuse.
• docs/evidence/ holds consolidated feature outputs and evidence bundles that should remain visible after a delivery cycle completes.

These roots are not the same as runtime storage:

• .boundline/ keeps session state, traces, checkpoints, and transient governance artifacts.
• .boundline/context-intelligence/ keeps the derived retrieval DB, manifest.json, and SQLite WAL/SHM sidecars used by local semantic retrieval.
• The large-codebase substrate extends that same derived state with repository-map readiness, digest-backed context refs, omission findings, and freshness-bound snapshot-cache metadata. None of those derived artifacts are reviewed memory or authoritative planning truth.
• .canon/ keeps raw Canon run packets and Canon-owned runtime payloads.

Default Layout

docs/
  project/
    README.md
    architecture.md
    domain-language.md
  evidence/
    README.md
    feature-slug/
      summary.md
      validation.md

The exact file names can vary. The stable rule is ownership:

• curated reusable inputs belong in docs/project/
• durable delivery outputs belong in docs/evidence/
• derived semantic index state belongs under .boundline/
• transient runtime artifacts do not belong in either folder
• large-codebase snapshot cache state must remain disposable and rebuildable

Bootstrap Behavior

boundline init creates both roots and seeds a README in each one. boundline update --apply recreates them when they are missing.

Related Pages

• Getting Started
• Configuration Reference
• Canon Integration
• Architecture And Decisions