| name | implement-knowledge-packet |
| description | Author or implement static expert KnowledgePacket support, including provenance, controls, token matching, leakage checks, and K-class assignment before dynamic RAG exists. |
Implement KnowledgePacket
Purpose
KnowledgePacket is a versioned static artifact used to test causal knowledge injection without introducing retrieval infrastructure.
Workflow
- Identify the knowledge type: generic principle, phase, opening structure, motif, example or current-position analysis.
- Record source, author/license, extraction method and semantic scope.
- Prohibit current best move, PV or transposition-near lookup unless the condition is explicitly K7.
- Produce correct, wrong/plausible, irrelevant token-matched and persona controls where the experiment requires them.
- Normalize token budget and presentation format.
- Compute content hash and assign K class.
- Validate against
knowledge-packet.schema.json.
- Add leakage audit and expected applicability criteria.
- Test manifest resolution and event provenance.
- Update experiment card, not just prompt text.
Evidence
A performance difference between baseline and any extra text is insufficient. Prefer contrasts such as:
correct skill > wrong plausible skill ≈ irrelevant token-matched control
Report susceptibility when wrong or adversarial knowledge changes behavior.