Research Loop

How to Execute

Run the research loop phases in order:

# Discover new sources
wikirecall knowledge search "<topic>"

# Ingest discovered sources
wikirecall knowledge create --title "<entity>" --type concept --tags "<tags>"

# Query for insights
wikirecall knowledge search "<query>"

The Loop: Discover → Ingest → Query → Compound

Phase 1 — Discover

Identify new sources of knowledge relevant to your active scenarios.

1. Sweep existing entities — look at related fields and [[wikilinks]] for references you haven't ingested yet
2. Search by tags — use wikirecall knowledge search to find clusters with few entities (potential gaps)
3. External sources — check arXiv, GitHub trending, blog aggregators, and conference proceedings for topics matching your tags
4. Stakeholder signals — review recent PRs, issues, and discussions for emerging concepts not yet captured

Output: A ranked list of candidate sources, each with a one-line description and relevance score (1–5).

Phase 2 — Ingest

Transform raw sources into structured knowledge entities.

1. Create entities for each high-relevance source:

   wikirecall knowledge create --title "Concept Name" --type concept --tags "tag1,tag2"
   

2. Write content using Karpathy-style mental models:
   • What it is (1–2 sentences)
   • Why it matters (context in your domain)
   • How it connects (link to existing entities via related and [[wikilinks]])
   • Open questions (what you don't know yet)
3. Cross-link aggressively — every new entity should reference 2–3 existing entities
4. Tag consistently — use existing tags before creating new ones

Phase 3 — Query

Use search and visualization to surface patterns.

1. Full-text search — wikirecall knowledge search "query" to find entities by content
2. Visualize the graph — wikirecall visualize to see the network topology
   • Isolated nodes = poorly connected knowledge (needs more links)
   • Dense clusters = well-understood areas
   • Bridge nodes = key concepts connecting different domains
3. Topic clusters — wikirecall visualize --type topic-clusters to find tag imbalances
4. Timeline view — wikirecall visualize --type timeline to track knowledge growth rate

Phase 4 — Compound

Let accumulated knowledge generate new insights.

1. Identify bridge opportunities — entities in different clusters that should be connected
2. Write synthesis entities — create "concept" entities that connect multiple domains:
   • "Retry Patterns + Circuit Breakers" → new entity: "Resilience Architecture"
   • "OAuth2 + RBAC" → new entity: "Authorization Patterns"
3. Update existing entities with new context from recent additions
4. Generate the research landscape — wikirecall visualize --type research-landscape for a full dashboard
5. Archive stale entities — mark entities with status: needs_update if they haven't been revised in 90+ days

Setting Up Automated Curation

Daily sweep (5 minutes)

# Check for entities updated today
wikirecall knowledge list | grep "$(date +%Y-%m-%d)"

# Generate fresh graph
wikirecall visualize --output ./daily-graph.html

Weekly review (30 minutes)

1. Run the full research landscape: wikirecall visualize --type research-landscape --output ./weekly-review.html
2. Open the dashboard and identify:
   • New clusters forming
   • Isolated nodes needing connections
   • Stale clusters with no recent updates
3. Create 3–5 new entities from the discovery phase
4. Update 2–3 existing entities with fresh connections

Monthly audit (1 hour)

1. Export the full entity list and review coverage
2. Compare entity count growth month-over-month
3. Identify the top 5 most-connected entities (your knowledge hubs)
4. Write one synthesis entity that connects your most active clusters
5. Archive any entities that are no longer relevant

Expected Outputs

After running a research loop cycle, you should have:

• New entities: 3–10 new knowledge entities per cycle
• Updated links: 5–15 new cross-references between existing entities
• Visualization: An updated knowledge graph showing the current state
• Gap analysis: A list of topics that need deeper coverage
• Synthesis notes: At least one new entity connecting previously separate clusters

Tips

• Start small — begin with 10 entities and grow organically
• Prefer depth over breadth — 5 well-connected entities beat 20 isolated ones
• Use wikilinks liberally — [[entity-name]] in content creates implicit graph edges
• Review the graph weekly — visual patterns reveal gaps your linear reading misses
• Tag taxonomy matters — keep tags to 10–15 core categories, merge similar ones