“Full Flow” Inspiration Notes

Digest of apps/stps/deploy_n_v2/v2/claude/full flow.md (long-form chat). Use this when you want the qualitative guidance without rereading the entire thread.

1. Single App vs Multiple Apps

• Recommendation: keep one modular platform rather than separate agency vs learning apps.
• Reason: everything (glossary, PRD, tasks, calendar, pattern extraction) lives on the same graph; separating would break relationships and search.
• Extract into multiple apps only when user bases/compliance needs diverge significantly.

2. “Everything is a Node” Architecture

• Strong fit: shared graph allows nodes (Signal, Task, PRD, Glossary Term) to interlink.
• Implementation guardrails:
  • Keep common indexed fields (title, status, timestamps) at the top level for querying.
  • Store type-specific data in props but validate per node type (e.g., Zod schemas).
  • Accept JSON flexibility but plan migrations/versioning.

3. Glossary / Concept Model

CONCEPT (language-independent)
  └─TERM_OF
TERM (language-specific)
  └─IN_GLOSSARY
GLOSSARY (curated set)
  └─IN_SESSION
SESSION (context where glossary is used)

• Enables scenarios: simultaneous interpretation (combine glossaries per session), vocabulary building, persistent knowledge base with RELATES_TO edges.

4. Pattern Extraction Framework

• Pipeline idea: Input → ExtractionRun → Output, with phases (framework detection, principle extraction, pattern clustering, strategy split).
• Build incrementally:
  1. Basic text → pattern extraction.
  2. Add multi-phase detection.
  3. Introduce strategy/implementation branching.
• Store outputs as structured nodes (Pattern/PRD/Manual).

5. Document Types & Relationships

• Treat PRD, Manual, Tutorial, How-to, Research, Idea, Task as Document nodes with docType.
• Use edges to show evolution (Idea EVOLVES_TO Task, How-to TEMPLATE_FOR Task).
• Add statuses that acknowledge “overwhelming task” states (e.g., someday).

6. Notes App Integration

• Preferred approach: bring the note system into apps/stps as route groups so linking/backlinking works with the same node/edge model and UI kit.
• If notes already have block editors, reuse the existing BlockEditor and linking logic.

7. Suggested Phasing

Phase 0 (cleanup): UI kit consolidation, path aliases.
Phase 1 (data foundation): node/edge tables, core node types (Document, Task, Event, Concept, Term), repo layer.
Phase 2 (kill localStorage): migrate Signals/Sessions/Calendar to DB nodes.
Phase 3 (glossary + sessions): CRUD for Concept/Term, session composition, basic realtime retrieval.
Phase 4 (pattern extraction MVP): text→pattern pipeline, manual→PRD conversion.

Use these insights when refining the graph model, designing session/glossary workflows, or planning post-MVP phases.