System Architecture

┌─────────────────┐
│  DREAMCATCHER   │  ← Founder's raw idea (e.g., "I want X from CES")
│  (Idea Intake)  │
└────────┬────────┘
         │ Structured idea brief
         ▼
┌─────────────────┐
│  POC DREAM      │  ← Grooms What (requirements) + How (specification)
│  FACTORY        │  → Generates N candidate approaches
└────────┬────────┘
         │ N candidates with specs
         ▼
┌─────────────────┐
│  THE FORGE      │  ← Daytona: Build each candidate in sandbox
│  (Build & Test) │  ← Sentry: Monitor errors/performance
└────────┬────────┘  → Score each with weighted objectives
         │ Top 3 candidates
         ▼
┌─────────────────┐
│  POLISHING      │  ← CodeRabbit: AI code review
│  (Code Quality) │  → Final tests, select winner
└────────┬────────┘
         │ Winner + polished code
         ▼
┌─────────────────┐
│  THE SHOWCASE   │  ← Generate docs: Eng, Marketing, Exec
│  (Documentation)│  ← ElevenLabs: Voice for podcasts/videos
└────────┬────────┘
         │ Full package
         ▼
┌─────────────────┐
│  CHRYSALIS      │  ← Company dogfooding period
│  (Validation)   │  → ROI analysis, market evaluation
└─────────────────┘  → Launch decision

@dataclass
class IdeaBrief:
    id: str
    title: str
    raw_description: str  # Founder's original brainfart
    context: str          # Where it came from (CES, customer request, etc.)
    aspirations: list[str]  # What success looks like
    constraints: list[str]  # Budget, timeline, tech limitations
    created_at: datetime

@dataclass
class Candidate:
    id: str
    idea_brief_id: str
    approach_name: str
    approach_description: str
    tech_stack: list[str]
    estimated_complexity: Literal["low", "medium", "high"]
    code_scaffold: str  # Generated starter code
    objectives: list[Objective]

@dataclass
class Objective:
    name: str
    description: str
    measurement: str  # How to measure (e.g., "response_time < 200ms")
    bias: Literal["H", "M", "L"]  # Weight multiplier

    @property
    def multiplier(self) -> int:
        return {"H": 3, "M": 2, "L": 1}[self.bias]

@dataclass
class ForgeResult:
    candidate_id: str
    daytona_workspace_id: str
    build_success: bool
    test_results: dict
    sentry_issues: list[dict]
    performance_metrics: dict
    objective_scores: dict[str, float]  # objective_name → score (0-100)
    total_score: float  # Weighted sum

class Message:
    from_agent: str
    to_agent: str
    payload: Any
    timestamp: datetime

forge:
  max_candidates: 5
  build_timeout_seconds: 300
  test_timeout_seconds: 120

scoring:
  bias_multipliers:
    H: 3
    M: 2
    L: 1
  top_n_for_polish: 3

showcase:
  voice_model: "eleven_multilingual_v2"
  output_formats: ["mp3"]