Topic Research Mining (TRM) Closed-Loop Research Workflow

This skill executes the end-to-end Topic Research Mining (TRM) research loop: mining live research notebooks in trm-vault, expanding and triaging discovered knowledge gaps against the local SQLite context cache, synthesizing Layer 2 wiki notes, and building thematic .nlm_pack files for NotebookLM grounding.

---

Architecture Overview

flowchart TD
    A["notebooklm-registry.json (12 Notebooks)"] --> B["1. Live Mining Sweep (schedule-task-wrapper-TRM-Notebooklm-Mine.ps1)"]
    B --> C["C:/Users/soren/trm-vault/trm/research-gaps/*.md"]
    C --> D["2. Multi-Model Cost Router & MHS Sweep (run-closed-loop-research-v2.mjs)"]
    D --> E["3. Cognitive Gap Triage (query-expander.mjs & trm-bot-runner.mjs)"]
    E --> F["wiki/research/rfc-gap-*.md Decision Notes"]
    D --> G["4. Layer 2 Wiki Synthesis (Tier 1 Mistral/DeepSeek or Tier 2 Frontier)"]
    D --> H["5. Thematic Knowledge Packs (.nlm_pack/pack_*.txt)"]
    F & G --> I["6. Wiki & Git Push (npm run wiki:publish)"]

---

Execution Procedures

1. Multi-Notebook Mining Sweep

To mine all registered notebooks in notebooklm-registry.json into C:\Users\soren\trm-vault\trm\research-gaps\:

pwsh.exe -NoProfile -ExecutionPolicy Bypass -File C:\dev\schedule-task-wrapper-TRM-Notebooklm-Mine.ps1

Outputs updated markdown tables for:

• cic-kb.md
• cic-daily-research.md
• willow-run-videos.md
• cast-iron-charlie-research-logs.md
• the-sorensen-photographic-archive-industrial-giants-at-willow-run.md
• castironcharlie-facebook.md
• cic-reddit.md
• cic-cuban-seizures-retired-assets.md
• cic-willow-run-aviation-engineering.md
• cic-post-war-willys-overland.md
• cic-ford-executive-dynamics-politics.md
• cic-research-deltas-living-matrix.md

---

2. Closed-Loop Orchestration & Wiki Synthesis

Run the live closed-loop synthesis orchestrator from the C:\dev repository root:

# Ingest mined gaps, synthesize Layer 2 wiki topics, and build .nlm_pack files
$env:TRM_SKIP_MINE='1'; node scripts/run-closed-loop-research-v2.mjs

Actions performed:

1. Step 0 — Cost-Routing Gateway & MHS Sweep: Reads host hardware profile (ModelHardwareStandard) and benchmark matrix (_integration/model_selection.json) to assign Tier 0 ($0 Ollama/FreeLLM), Tier 1 (Muscle Cloud), and Tier 2 (Frontier).
2. Step 1 — Gaps Ingestion: Writes consolidated snapshot to trm-research-gaps.md.
3. Step 2 — NotebookLM Grounding: Uploads gaps snapshot to the target daily notebook.
4. Step 3 — Topic Derivation: Extracts the 4 core research dimensions (open-contradictions, under-sourced, adjacent-topics, follow-up).
5. Step 4 — Layer 2 Wiki Synthesis: Synthesizes structured markdown pages in wiki/research/ via routed Muscle/Frontier inference.
6. Step 5 — Audit Log: Appends run telemetry, net cost savings, and SHA-256 hash chains to wiki/Log.md.
7. Step 6 — Knowledge Packs: Rebuilds thematic .nlm_pack/pack_*.txt files.

---

3. Automated Gap Triage & RFC Note Drafting

Evaluate pending gaps using cognitive query expansion against the SQLite context cache (.kb_cache/knowledge.db):

# Full daily wrapper with cache pre-sync and logging
npm run trm:triage:daily

# Direct CLI run with auto provider (Ollama -> OpenRouter -> Heuristic)
npm run trm:triage:auto

Actions performed:

• Expands each pending gap query using local or remote LLMs with SQLite LIMIT 0 dry-run validation.
• Queries SQLite FTS5 index for sub-millisecond lexical snippet matches.
• Generates structured RFC decision notes in wiki/research/rfc-gap-*.md.
• Mutates trm-research-gaps.md by linking in-progress gaps to their drafted RFCs.

---

4. Publish and Sync Knowledge Base & Wiki

Publish newly synthesized RFCs, concept nodes, and research topics to the GitHub Wiki:

# Synchronize and push all wiki docs to GitHub Wiki
npm run wiki:publish

---

Verifying & Testing

# Run TRM query expander unit tests (25 tests)
node --test kb-sync/tests/query-expander.test.mjs

# Run local SQLite cache tests
npm run test:cache

# Run TRM pipeline sandbox validation tests
npm run test:trm