🌀
← Back to Articles & Artefacts
artefactswest

RISE / QMD Implications — Wave 1

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

RISE / QMD Implications — Wave 1

Observed state

  • llms-txt already has the right upgrade logic, but not yet the actual rispecs/ corpus: rispecs/README.md is still a staging note rather than a living compiled wiki (/workspace/repos/jgwill/llms-txt/rispecs/README.md:1-4).
  • The RISE framework already describes continuous rispec improvement, commit mining, staleness detection, and future helper tools such as rise_detect_spec_staleness, rise_suggest_upgrades, and rise_validate_spec (/workspace/repos/jgwill/llms-txt/llms-rise-framework.txt:594-605,608-676,727-835,966-981).
  • PDE already stores durable paired markdown+JSON artifacts and companion outputs like pde-inquiry-enrichment.md and pde-four-questions.md (/workspace/repos/jgwill/llms-txt/llms-pde.txt:103-122,125-152).
  • Inquiry routing already names QMD-local as a route target, but the docs still frame routing as a broader layer than QMD alone (/workspace/repos/jgwill/llms-txt/llms-inquiry-routing.md:69-79,141-147).
  • Kinship rules explicitly reject destructive regeneration and require merge-preserving updates (/workspace/repos/jgwill/llms-txt/llms-kinship-hub-system.md:75-100).

What Karpathy's article should change

  1. Turn rispecs/ into a compiled markdown wiki layer.
    The new responsibility is not just "write specs when needed" but "maintain an interlinked rispec wiki that accumulates synthesized understanding over time," matching Karpathy's persistent wiki pattern (sources/karpathy-llm-wiki-gist/llm-wiki.md:11-13,31-41).

  2. Extend RISE Phase 4 from spec export to wiki maintenance.
    The framework already knows how to detect staleness and suggest rispec upgrades. The next step is to let those upgrades target wiki pages, summaries, and relationship maps — not only standalone spec files (/workspace/repos/jgwill/llms-txt/llms-rise-framework.txt:594-605,966-981).

  3. Keep automation merge-based, not overwrite-based.
    KINSHIP.md is explicitly a living charter that should not be regenerated from scratch. The same rule should govern rispec/wiki maintenance: page updates must preserve prior commitments, lineage, and unresolved tensions (/workspace/repos/jgwill/llms-txt/llms-kinship-hub-system.md:95-100).

  4. Add explicit wiki metadata.
    Wave 2 should define frontmatter for rispec/wiki pages: doc_id, lineage, source_commits, related_pages, freshness, validation_status, query_provenance, and promotion_status. This is the bridge between Karpathy's wiki pages and RISE/QMD automation.

  5. Make lint/health-check passes first-class.
    Karpathy's lint pass for contradictions/orphans/stale claims should become a RISE-compatible maintenance phase backed by the existing compliance mindset in llms-txt (sources/karpathy-llm-wiki-gist/llm-wiki.md:41-53; /workspace/repos/jgwill/llms-txt/llms-txt-compliance-checklist.md:31-44,102-109).

QMD-specific implications

  • QMD should stay an optional search/enrichment substrate, not the only ontology.
    Current PDE docs describe miaco qmd-inquiry-decompose as enrichment over decomposed facets, not as the sole authority for routing or page structure (/workspace/repos/jgwill/llms-txt/llms-pde.txt:120-152,195-198).
  • QMD needs a promotion contract.
    Today it enriches PDE artifacts; next it should help propose or update rispec/wiki pages, with human review and kinship-aware merge rules before promotion.
  • QMD results need provenance-bearing citations.
    qmd://... hits in enrichment artifacts are already promising, but they need promotion rules into stable markdown pages rather than remaining isolated per-session sidecars.

Episodic / inquiry-routing implications

  • Inquiry routing already carries the right shape: relational context, accountability, ceremonial intent, routing targets, and future spiral tracking (/workspace/repos/jgwill/llms-txt/llms-inquiry-routing.md:14-22,50-60,69-79,141-147).
  • The missing piece is a formal episode → PDE → inquiry → enrichment → promoted page chain so later sessions inherit the journey, not just the terminal result.
  • Wave 2 should define at least four promotion targets:
    1. wiki-page-draft
    2. rispec-upgrade
    3. episodic-memory-store
    4. kinship-update-review

Proposed automation loop

  1. Detect changed sources, commits, PDE artifacts, and inquiry outputs.
  2. Run QMD-backed enrichment over decomposed facets.
  3. Draft or merge rispec/wiki page updates with lineage metadata.
  4. Run kinship-aware lint: stale claims, contradictions, orphan pages, missing relations.
  5. Write index.md + append-only log.md changes.
  6. Mark promoted outputs back into episodic memory / inquiry lineage.