Initial kitchen scaffold — Phase 1 kitchen port (build-verified 2026-07-11)

FastAPI backend (Python 3.11, MSSQL ODBC for SambaPOS, Azure DI OCR),
kitchen_db on central PG. React/TS/Vite frontend with navy sidebar layout.

Backend: auth.py (APP_SLUG=kitchen, SimpleNamespace — archive routes use
.kitchen_id/.is_admin without modification), main.py (51 migrations, scheduler,
internal router for KDS bookings feed), api/internal.py, full archive API
(31 routers: invoices, recipes, menus, sambapos, resos, newbook, disputes,
purchase_orders, etc.), models, migrations, OCR pipeline.
kitchen_id pinned to 1 (B1 — single hotel).

Frontend: AuthGate (app=kitchen, token shim for archive compat — B5b pending),
Layout (navy sidebar, 6 sections, Lucide icons, teal --app-primary),
App.tsx (Outlet pattern, UploadApp outside Layout), index.css (full :root block).
strict: false — archive components have type issues; build clean.

Note: 45 archive components call fetch('/api/...') without /kitchen/ prefix
(B5b). Runtime 404s; deferred until after initial testing.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

backend/models/llm.py

@@ -0,0 +1,43 @@
+"""
+LLM integration models — usage tracking and response caching.
+LLM FEATURE — see LLM-MANIFEST.md for removal instructions
+"""
+from datetime import datetime
+from typing import Optional
+from sqlalchemy import String, DateTime, Integer, Text, Boolean, UniqueConstraint
+from sqlalchemy.dialects.postgresql import JSONB
+from sqlalchemy.orm import Mapped, mapped_column
+from database import Base
+
+
+class LlmUsageLog(Base):
+    """Log every LLM API call for cost tracking and debugging"""
+    __tablename__ = "llm_usage_log"
+
+    id: Mapped[int] = mapped_column(primary_key=True)
+    kitchen_id: Mapped[int] = mapped_column(Integer, nullable=False, index=True)
+    feature: Mapped[str] = mapped_column(String(50), nullable=False, index=True)  # label_analysis, invoice_assist, etc.
+    model: Mapped[str] = mapped_column(String(100), nullable=False)
+    input_tokens: Mapped[int] = mapped_column(Integer, default=0)
+    output_tokens: Mapped[int] = mapped_column(Integer, default=0)
+    latency_ms: Mapped[int] = mapped_column(Integer, default=0)
+    success: Mapped[bool] = mapped_column(Boolean, default=True)
+    error_message: Mapped[Optional[str]] = mapped_column(Text, nullable=True)
+    created_at: Mapped[datetime] = mapped_column(DateTime, default=datetime.utcnow, index=True)
+
+
+class LlmAnalysisCache(Base):
+    """Cache LLM responses to avoid redundant calls for identical inputs"""
+    __tablename__ = "llm_analysis_cache"
+    __table_args__ = (
+        UniqueConstraint("feature", "input_hash", "prompt_version", name="uq_llm_cache_feature_hash_version"),
+    )
+
+    id: Mapped[int] = mapped_column(primary_key=True)
+    kitchen_id: Mapped[int] = mapped_column(Integer, nullable=False, index=True)
+    feature: Mapped[str] = mapped_column(String(50), nullable=False)  # label_analysis, ingredient_match, etc.
+    input_hash: Mapped[str] = mapped_column(String(64), nullable=False, index=True)  # SHA-256 of input
+    result_json: Mapped[dict] = mapped_column(JSONB, nullable=False)
+    model_used: Mapped[str] = mapped_column(String(100), nullable=False)
+    prompt_version: Mapped[str] = mapped_column(String(10), nullable=False)
+    created_at: Mapped[datetime] = mapped_column(DateTime, default=datetime.utcnow)