Skip to main content
Version: Current

AI Bootstrap

Use this page when the only input available to an AI system is https://bdl-lang.org.

The canonical BDL knowledge baseline is 5.2.2 ALIGNED. The machine-readable bootstrap is available at /ai-bootstrap.json, and the full AI navigation file is /llms-full.txt.

Retrieval order​

  1. Current material first.
  2. Higher version before lower version.
  3. Higher confidence before lower confidence.
  4. Fresher evidence before older evidence.
  5. Preserved or historical material must not override Current material.

The canonical authority order is:

  1. Current Source
  2. Runtime/Test Evidence
  3. Verified Artifact/SHA/Diff
  4. Last Verified Closure
  5. Historical Docs

Mandatory generation pipeline​

  1. Requirement decomposition
  2. Shared/Core reuse check
  3. Artifact selection
  4. Pattern selection
  5. Capability and known-gap lookup
  6. Live metadata resolution
  7. Minimal BDL authoring
  8. Parser/Preflight validation
  9. ValidateOnly/Compile
  10. Plan and human review
  11. Approval where required
  12. Governed execute
  13. Reload/Post-verify
  14. Export/Round-trip evidence

Hard prohibitions​

  • Never guess native IDs.
  • Never infer runtime support from a model property.
  • Never use direct SQL mutation as a BDL workaround.
  • Never silently drop unsupported properties.
  • Never emulate deferred Selective Export by filtering a full-system export.
  • Never auto-approve destructive metadata, record, or system-code mutation.
  • Never claim Build PASS as Runtime Verified.

Machine-readable corpus​

Start with these surfaces:

The AI corpus publishes public-safe current and preserved documentation with provenance and hashes. Raw source code, diagnostic ZIPs, private runtime evidence, and operational secrets are not part of the public corpus.