arch-mvp-roadmap

# MVP & Roadmap Planning Patterns for defining what to build first and in what order. ## Core Principle > **"Build the smallest thing that proves value."** An MVP is not a half-built product—it's a complete vertical slice that validates assumptions. ## MoSCoW Prioritization | Priority | Definition | Criteria | |----------|------------|----------| | **Must Have** | System doesn't function without | Core journey incomplete, no workarounds | | **Should Have** | Important but not blocking | Workarounds exist, high value | | **Could Have** | Nice to have | Enhances experience, low priority | | **Won't Have** | Explicitly out of scope | Prevents scope creep, document for later | ## MVP Scoping Checklist - [ ] Single complete user journey end-to-end - [ ] Validates core assumption/hypothesis - [ ] Deployable and demonstrable - [ ] Measurable success criteria defined - [ ] No features without corresponding tests ## Phased Delivery Pattern Each phase should: 1. Build on previous phase (not parallel development) 2. Be independently deployable 3. Have clear success criteria 4. Include tests for new functionality ## Dependency Ordering Order features by: | Factor | Question | |--------|----------| | **Technical** | What must exist first? | | **Value** | What provides most value soonest? | | **Risk** | What validates riskiest assumptions? | | **Learning** | What teaches us most about the domain? | ## Roadmap Template | Phase | Features | Success Criteria | Dependencies | |-------|----------|------------------|--------------| | MVP | [Must-haves] | [Measurable outcomes] | None | | Phase 2 | [Should-haves] | [Measurable outcomes] | MVP complete | | Phase 3 | [Could-haves] | [Measurable outcomes] | Phase 2 complete | ## Extensible Algorithm Design Design algorithms for evolution using stable interfaces. MVP delivers value immediately while building ground truth for future ML phases. **Evolution path:** 1. **MVP:** Deterministic algorithm (keyword matching, rule-based) 2. **Phase 2:** Statistical approach (TF-IDF, collaborative filtering) 3. **Phase 3:** ML/embeddings (neural networks, transformers) 4. **Phase 4:** LLM-powered (if needed) **Interface pattern:** ```python class Categorizer(Protocol): def categorize(self, text: str) -> tuple[str, float]: """Returns (category, confidence_score)""" ... ``` **Key principles:** - MVP uses zero ML infrastructure (no model serving, no embeddings DB) - Interface stays stable across phases (same input/output signature) - Each phase is independently measurable (track accuracy improvement) - Switch implementations via dependency injection, not rewrite See `reference.md` for detailed patterns and `examples.md` for sample roadmaps. --- > Converted and distributed by [TomeVault](https://tomevault.io/claim/ai-enhanced-engineer) — claim your Tome and manage your conversions. <!-- tomevault:4.0:skill_md:2026-04-14 -->