# Research Methodology

## Dispatch Architecture

### Team Composition
Each batch is split across 3 parallel research teams, each covering 3-8 sites:
- **Team Alpha:** Marketing/landing optimization + scraping/monitoring tools
- **Team Bravo:** Micro-SaaS platforms + social/content tools  
- **Team Charlie:** Career/local discovery + document tools (typically highest priority due to HotNow competitive overlap)

### Model Selection
- Default: deepseek-v4-pro (cost-effective, strong at structured analysis)
- Premium: Claude Sonnet 5 for mission-critical competitive analysis (e.g., Locale-NYC vs HotNow)
- Fallback: deepseek-v4-flash for high-volume scraping passes

### Dispatch Pattern
```
delegate_task(tasks=[
  {goal: "...", context: "...", role: "orchestrator"},
  {goal: "...", context: "...", role: "orchestrator"},  
  {goal: "...", context: "...", role: "orchestrator"}
])
```

Each team subagent is an orchestrator — they can spawn their own leaf subagents for individual site scraping. This keeps context windows manageable and enables parallel site analysis.

## Site Analysis Protocol

### Per-Site Mandatory Extraction
1. **Core product/service** — What do they actually sell? (not what their tagline says)
2. **Pricing model** — Free tier? Monthly? Per-seat? Enterprise?
3. **Key features** — The 3-5 things that define the product
4. **UI/UX patterns** — Layout decisions, card designs, search patterns, filter approaches
5. **USP (Unique Selling Proposition)** — What's their "one thing"?
6. **Target audience** — Who is this for? (evidence from copy, not assumption)
7. **Tech stack indicators** — Framework hints, JS libraries, hosting clues
8. **Polish rating (1-5)** — Landing page quality, copy clarity, conversion strategy

### Competitive Comparison Dimensions
When comparing a target site to an ITPP product, evaluate across:
- Feature completeness
- UX/UI polish
- Monetization strategy
- Data pipeline (how do they get their content?)
- Mobile readiness
- Conversion funnel
- Brand voice and positioning

## Cross-Reference Protocol

### Feature Mapping
Every extracted feature is mapped to at least one ITPP product with:
- **What:** The specific feature (e.g., "their tiered pricing table with hover comparison")
- **From:** Source site
- **To:** Target ITPP product
- **Why:** Reason this feature would improve the product
- **Effort:** Low / Medium / High implementation estimate
- **Impact:** 1-5 rating of expected user impact

### Priority Tiers
- **P0 (Ship This Week):** 1-2 day features that close obvious gaps
- **P1 (Next Sprint):** 3-7 day features that add significant value
- **P2 (Roadmap):** Features requiring refactoring or new infrastructure
- **P3 (Nice to Have):** "Eventually" features with marginal impact

## New Product Identification

### Signal Detection
A new product opportunity exists when:
- 3+ sites in a batch cluster around the same problem
- A site solves a problem in one vertical that doesn't exist in another
- A "pick-and-shovel" opportunity emerges (tools for the toolmakers)
- A local version of a city-specific app is viable (e.g., Locale-NYC → HotNow Savannah)

### Evaluation Criteria
- Market size estimate (rough)
- Competition density
- Our unique advantage
- MVP scope and timeline
- AI cost for ongoing operation
- Domain availability

## Cost Tracking

### Per-Batch Cost Attribution
```
Research Batch 001:
  Team Alpha:  $X.XX (deepseek-v4-pro, N tool calls)
  Team Bravo:  $X.XX (deepseek-v4-pro, N tool calls)
  Team Charlie: $X.XX (deepseek-v4-pro, N tool calls)
  Integration: $X.XX (deepseek-v4-pro, N tool calls)
  Total:       $X.XX
```

Tracked via dedicated LiteLLM virtual key `competitive-research-prod` on admin-ai.

### Historical Cost Baseline
| Batch | Date | Total Cost | Sites | Cost/Site | Notes |
|---|---|---|---|---|---|
| 001 | 2026-08-10 | TBD | 20 | TBD | Initial batch |

## Report Template

Each batch produces a master report at `research/NNN-slug.md`:
1. Executive Summary (1 paragraph)
2. Site-by-Site Breakdown (summary table)
3. Feature Extraction Matrix (sortable table)
4. Product Enhancement Recommendations (grouped by product, prioritized)
5. New Product Opportunities (concept cards)
6. Risk Flags & Warnings
7. Cost Summary
8. Next Batch Queue (sites flagged for future analysis)