← Back to Articles & Artefacts
artefactseast

Research Map — Wave 1

IAIP Research
deep-search--karpathy--llm-wiki--and--rise-framework--qmd-episodic--implication

Research Map — Wave 1

Theme clusters

ThemeWhy it mattersLocal anchorsNext-wave literature cluster
Persistent markdown wikiKarpathy's core move is compiled knowledge that accumulates instead of being re-derived each time.sources/karpathy-llm-wiki-gist/llm-wiki.md:11-13,31-53Personal knowledge bases, wiki maintenance, repository knowledge graphs, living documentation
Schema, governance, and kinshipA persistent wiki only stays coherent if page structure, relations, and update rules are explicit./workspace/repos/jgwill/llms-txt/llms-kinship-hub-system.md:33-69,75-100; /workspace/repos/miadisabelle/mia-awesome-copilot/AGENTS.md:58-107documentation schemas, knowledge governance, provenance, graph/ontology maintenance
Incremental ingest / query / lintKarpathy's ingest/query/lint loop already rhymes with PDE + subagent orchestration + verification.sources/karpathy-llm-wiki-gist/llm-wiki.md:35-53; /src/Miadi/rispecs/skills/rise-pde-session-multi-agents/SKILL.md:11-16,46-50,126-140,188-220; /workspace/repos/miadisabelle/mia-awesome-copilot/plugins/doublecheck/README.md:1-25multi-agent orchestration, documentation quality assurance, human-in-the-loop review
QMD as search/enrichment substrateCurrent local evidence shows QMD already enriches PDE artifacts, but should not yet be treated as the only routing ontology./workspace/repos/jgwill/llms-txt/llms-pde.txt:120-152,195-198; /src/Miadi/rispecs/miadi-code/pde/pde-four-questions-pipeline.rispec.md:17-23,42-64,104-128semantic retrieval, repo-local search, hybrid BM25/vector wiki search, retrieval evaluation
Inquiry-routing and episodic continuityKarpathy's pattern becomes more powerful when question lineage and answer promotion are preserved across sessions./workspace/repos/jgwill/llms-txt/llms-inquiry-routing.md:14-22,44-47,71-79,131-147; /src/Miadi/miadi-code/rispecs/pde-inquiry-routing.spec.md:23-41,67-76; /src/Miadi/rispecs/skills/mcp/04-ceremonial-memory.spec.md:33-41,66-81,89-119episodic memory for agents, session continuity, question routing, memory promotion pipelines
Kit vs suite packagingThe research needs to land as concrete agent/plugin/skill/package surfaces, not just design language./src/Miadi/miadi-code/references/claude/plugins.md:11-17,25-35,173-187,199-212; /src/Miadi/miadi-code/references/claude/sub-agents.md:143-171,181-208; /workspace/repos/miadisabelle/mia-awesome-copilot/plugins/context-engineering/README.md:1-27; /workspace/repos/miadisabelle/mia-awesome-copilot/plugins/structured-autonomy/README.md:1-20plugin ecosystems, subagent packaging, developer tool marketplaces, runtime composition

Search questions derived from the article

  1. Compiled knowledge — What architectures let an agent maintain a markdown wiki that accumulates synthesis, contradictions, and cross-links over time instead of re-running RAG every query?
  2. Spec-to-wiki promotion — How should RISE rispecs, PDE artifacts, and QMD enrichment outputs promote into stable wiki pages without losing provenance or human review checkpoints?
  3. Routing substrate — When should QMD act as search backend, when as routing signal, and when should a separate inquiry-routing layer remain authoritative?
  4. Episodic continuity — What memory models best connect session traces, inquiry batches, and promoted wiki pages so later agents inherit context without collapsing all nuance into summaries?
  5. Governance — What merge/lint/verification patterns keep a wiki trustworthy when multiple agents update it incrementally?
  6. Packaging — Which responsibilities belong in repo-local Copilot plugins/skills/subagents, and which belong in shared orchestration-suite infrastructure?

Wave-2 source clusters to pull

  • Agent memory / episodic memory: agent continuity, episodic memory stores, persistent task memory
  • Documentation knowledge systems: software documentation graphs, knowledge base maintenance, repository intelligence
  • Retrieval and routing: hybrid search, semantic query routing, query decomposition for code/document corpora
  • Multi-agent orchestration: manager/worker, verification layers, critique/revision loops
  • Human-in-the-loop wiki maintenance: provenance-preserving doc updates, living documentation workflows
  • Indigenous / relational governance overlays: accountability, consent/protocol, non-destructive merge/update patterns