Benchmark: kultainen esimerkki + zensical-dokumentointiohjeet

- golden-examples/todo/: 6/6 PASS referenssitoteutus
  - SQLAlchemy 2.0 (DeclarativeBase, Mapped, mapped_column)
  - Pydantic v2 (ConfigDict)
  - PEP 621 pyproject.toml, Python >=3.14
  - Uniikki testidata per testi
- CODE_SYSTEM päivitetty: few-shot kultaisesta esimerkistä
- DOCUMENTATION.md: zensical-dokumentointiohjeet

network-poc/tests/golden-examples/DOCUMENTATION.md

@@ -0,0 +1,84 @@
+# Dokumentointiohjeet — Zensical
+
+Hyvä dokumentointi kertoo **mitä asia ON**, ei mitä se tekee. Se on kuin zen-koan: lyhyt, tarkka, riittävä.
+
+## Periaatteet
+
+1. **Yksi rivi riittää.** Jos tarvitset kappaleen, koodi on liian monimutkainen.
+2. **Kerro mitä, älä miten.** `"""Tietokantamallit — SQLAlchemy 2.0, SQLite."""` ei `"""This module creates database models using SQLAlchemy..."""`
+3. **Älä toista koodia.** Jos funktio on `create_todo`, docstring ei ole "Creates a todo".
+4. **Suomi tai englanti, ei molempia.** Valitse yksi kieli per projekti.
+5. **Ei täytesanoja.** "This module provides functionality for" → poista.
+
+## Mitä dokumentoidaan
+
+| Kohde | Dokumentointi | Esimerkki |
+|-------|--------------|-----------|
+| **Moduuli** (.py) | Aina. Yksi rivi: mitä tiedosto sisältää. | `"""Pydantic v2 -skeemat — Create ja Response."""` |
+| **Luokka** | Aina. Mitä entiteetti edustaa. | `"""Tehtävä — otsikko, deadline, prioriteetti."""` |
+| **Funktio** | Vain jos nimi ei kerro kaikkea. | `get_db` → `"""Tietokantasessio per pyyntö."""` |
+| **CRUD-endpoint** | Ei. Nimi + HTTP-metodi riittää. | `create_todo`, `list_todos` — itsedokumentoivia |
+| **Testi** | Ei. Testin nimi on dokumentaatio. | `test_get_todo_not_found` — selvä |
+| **Konfiguraatio** | Kommentti vain jos arvo yllättää. | `check_same_thread: False  # SQLite + FastAPI` |
+
+## Mitä EI dokumentoida
+
+- Importteja
+- Ilmeisiä parametreja (`item_id: int`)
+- Tyyppivihjeitä jotka kertovat saman asian
+- Geneerisiä "boilerplate"-docstringejä
+
+## Esimerkkejä
+
+### Hyvä (zensical)
+
+```python
+"""Tietokantamallit — SQLAlchemy 2.0, Mapped-tyypitys, SQLite."""
+
+class Todo(Base):
+    """Tehtävä — otsikko, kuvaus, deadline, prioriteetti ja status."""
+    ...
+
+def get_db():
+    """Tietokantasessio per pyyntö."""
+    ...
+```
+
+### Huono (verbose)
+
+```python
+"""
+This module defines the database models for the Todo application.
+It uses SQLAlchemy ORM to create the database tables and provides
+the session factory for database connections.
+"""
+
+class Todo(Base):
+    """
+    Represents a todo item in the database.
+
+    Attributes:
+        id: The unique identifier for the todo item.
+        title: The title of the todo item.
+        ...
+    """
+    ...
+```
+
+### Huono (tyhjä)
+
+```python
+# Ei docstringejä ollenkaan — lukija ei tiedä mikä tiedoston rooli on
+class Todo(Base):
+    __tablename__ = "todos"
+    ...
+```
+
+## Tarkistuslista
+
+Generoitu koodi on hyvin dokumentoitu kun:
+- [ ] Jokainen .py-tiedosto alkaa yksirivisellä docstringillä
+- [ ] Jokainen luokka kertoo mitä entiteetti edustaa
+- [ ] Docstringit ovat saman kielen kuin muu koodi
+- [ ] CRUD-endpointeilla ei ole turhia docstringejä
+- [ ] Kommentteja on vain siellä missä koodi yllättää

network-poc/tests/golden-examples/todo/main.py

@@ -0,0 +1,61 @@
+"""FastAPI CRUD — yksi endpoint-setti per entiteetti."""
+
+from fastapi import FastAPI, Depends, HTTPException
+from sqlalchemy.orm import Session
+
+from models import SessionLocal, Todo
+from schemas import TodoCreate, TodoResponse
+
+app = FastAPI()
+
+
+def get_db():
+    """Tietokantasessio per pyyntö."""
+    db = SessionLocal()
+    try:
+        yield db
+    finally:
+        db.close()
+
+
+@app.post("/todos/", response_model=TodoResponse, status_code=201)
+def create_todo(item: TodoCreate, db: Session = Depends(get_db)):
+    db_item = Todo(**item.model_dump())
+    db.add(db_item)
+    db.commit()
+    db.refresh(db_item)
+    return db_item
+
+
+@app.get("/todos/", response_model=list[TodoResponse])
+def list_todos(db: Session = Depends(get_db)):
+    return db.query(Todo).all()
+
+
+@app.get("/todos/{item_id}", response_model=TodoResponse)
+def get_todo(item_id: int, db: Session = Depends(get_db)):
+    item = db.query(Todo).filter(Todo.id == item_id).first()
+    if not item:
+        raise HTTPException(status_code=404, detail="Todo not found")
+    return item
+
+
+@app.put("/todos/{item_id}", response_model=TodoResponse)
+def update_todo(item_id: int, item: TodoCreate, db: Session = Depends(get_db)):
+    db_item = db.query(Todo).filter(Todo.id == item_id).first()
+    if not db_item:
+        raise HTTPException(status_code=404, detail="Todo not found")
+    for key, value in item.model_dump().items():
+        setattr(db_item, key, value)
+    db.commit()
+    db.refresh(db_item)
+    return db_item
+
+
+@app.delete("/todos/{item_id}", status_code=204)
+def delete_todo(item_id: int, db: Session = Depends(get_db)):
+    db_item = db.query(Todo).filter(Todo.id == item_id).first()
+    if not db_item:
+        raise HTTPException(status_code=404, detail="Todo not found")
+    db.delete(db_item)
+    db.commit()

network-poc/tests/golden-examples/todo/models.py

@@ -0,0 +1,30 @@
+"""Tietokantamallit — SQLAlchemy 2.0, Mapped-tyypitys, SQLite."""
+
+from datetime import date
+
+from sqlalchemy import String, Text, Date, create_engine
+from sqlalchemy.orm import DeclarativeBase, Mapped, mapped_column, sessionmaker
+
+DATABASE_URL = "sqlite:///./app.db"
+engine = create_engine(DATABASE_URL, connect_args={"check_same_thread": False})
+SessionLocal = sessionmaker(autocommit=False, autoflush=False, bind=engine)
+
+
+class Base(DeclarativeBase):
+    pass
+
+
+class Todo(Base):
+    """Tehtävä — otsikko, kuvaus, deadline, prioriteetti ja status."""
+
+    __tablename__ = "todos"
+
+    id: Mapped[int] = mapped_column(primary_key=True, index=True)
+    title: Mapped[str] = mapped_column(String(255))
+    description: Mapped[str | None] = mapped_column(Text, default=None)
+    due_date: Mapped[date | None] = mapped_column(Date, default=None)
+    priority: Mapped[int] = mapped_column(default=1)
+    status: Mapped[str] = mapped_column(String(20), default="pending")
+
+
+Base.metadata.create_all(bind=engine)

network-poc/tests/golden-examples/todo/pyproject.toml

@@ -0,0 +1,11 @@
+[project]
+name = "todo-app"
+version = "0.1.0"
+requires-python = ">=3.14"
+dependencies = [
+    "fastapi",
+    "uvicorn[standard]",
+    "sqlalchemy",
+    "pytest",
+    "httpx",
+]

network-poc/tests/golden-examples/todo/schemas.py

@@ -0,0 +1,22 @@
+"""Pydantic v2 -skeemat — Create sisääntulolle, Response vastaukselle."""
+
+from datetime import date
+
+from pydantic import BaseModel, ConfigDict
+
+
+class TodoCreate(BaseModel):
+    """Uuden tehtävän luonti. Pakolliset: title."""
+
+    title: str
+    description: str | None = None
+    due_date: date | None = None
+    priority: int = 1
+    status: str = "pending"
+
+
+class TodoResponse(TodoCreate):
+    """Palautettava tehtävä — sisältää id:n."""
+
+    id: int
+    model_config = ConfigDict(from_attributes=True)

network-poc/tests/golden-examples/todo/test_main.py

@@ -0,0 +1,69 @@
+"""Pytest — TestClient, erillinen test.db, uniikki data per testi."""
+
+from fastapi.testclient import TestClient
+from sqlalchemy import create_engine
+from sqlalchemy.orm import sessionmaker
+
+from main import app, get_db
+from models import Base
+
+test_engine = create_engine(
+    "sqlite:///./test.db", connect_args={"check_same_thread": False}
+)
+TestSession = sessionmaker(autocommit=False, autoflush=False, bind=test_engine)
+Base.metadata.create_all(bind=test_engine)
+
+
+def override_get_db():
+    db = TestSession()
+    try:
+        yield db
+    finally:
+        db.close()
+
+
+app.dependency_overrides[get_db] = override_get_db
+client = TestClient(app)
+
+
+def test_create_todo():
+    response = client.post("/todos/", json={"title": "Osta maitoa", "priority": 2})
+    assert response.status_code == 201
+    assert response.json()["title"] == "Osta maitoa"
+    assert "id" in response.json()
+
+
+def test_list_todos():
+    client.post("/todos/", json={"title": "Listattava tehtävä"})
+    response = client.get("/todos/")
+    assert response.status_code == 200
+    assert len(response.json()) >= 1
+
+
+def test_get_todo_by_id():
+    created = client.post("/todos/", json={"title": "Haettava tehtävä"}).json()
+    response = client.get(f"/todos/{created['id']}")
+    assert response.status_code == 200
+    assert response.json()["id"] == created["id"]
+
+
+def test_get_todo_not_found():
+    response = client.get("/todos/99999")
+    assert response.status_code == 404
+
+
+def test_update_todo():
+    created = client.post("/todos/", json={"title": "Vanha otsikko"}).json()
+    response = client.put(
+        f"/todos/{created['id']}", json={"title": "Uusi otsikko"}
+    )
+    assert response.status_code == 200
+    assert response.json()["title"] == "Uusi otsikko"
+
+
+def test_delete_todo():
+    created = client.post("/todos/", json={"title": "Poistettava"}).json()
+    response = client.delete(f"/todos/{created['id']}")
+    assert response.status_code == 204
+    response = client.get(f"/todos/{created['id']}")
+    assert response.status_code == 404

network-poc/tests/model-benchmark.mjs

@@ -11,7 +11,11 @@
  */
 
 import { execSync } from 'child_process';
-import { writeFileSync, mkdirSync, rmSync, existsSync } from 'fs';
+import { writeFileSync, readFileSync, mkdirSync, rmSync, existsSync } from 'fs';
+import { dirname, join } from 'path';
+import { fileURLToPath } from 'url';
+
+const __dirname = dirname(fileURLToPath(import.meta.url));
 
 // === CLI-argumentit ===
 const args = process.argv.slice(2);
@@ -141,15 +145,29 @@ Blog → Author: name,email,bio(Text|None) / Post: title, content(Text), author_
 
 const FIX_SYSTEM = 'You are a Python code fixer. Return ONLY the corrected Python file. No markdown fences, no explanations — just valid Python code.';
 
+// === Kultainen esimerkki ===
+const GOLDEN_DIR = join(__dirname, 'golden-examples', 'todo');
+const GOLDEN_FILES = ['models.py', 'schemas.py', 'main.py', 'test_main.py', 'pyproject.toml'];
+function loadGoldenExample() {
+    if (!existsSync(GOLDEN_DIR)) return '';
+    let example = '\nREFERENCE IMPLEMENTATION (todo project — follow this exact structure, style, and conventions):\n\n';
+    for (const f of GOLDEN_FILES) {
+        const path = join(GOLDEN_DIR, f);
+        if (existsSync(path)) example += `=== ${f} ===\n${readFileSync(path, 'utf-8').trim()}\n\n`;
+    }
+    return example;
+}
+const GOLDEN_EXAMPLE = loadGoldenExample();
+
 const CODE_SYSTEM = `You are a Python backend developer. Generate a complete FastAPI project with SQLAlchemy and SQLite.
 
-Given the project requirements and JSON specification, generate these 5 files:
+Given the project requirements, JSON specification, and a REFERENCE IMPLEMENTATION, generate these 5 files:
 
-1. models.py - SQLAlchemy models with database setup (create_engine, declarative_base, sessionmaker, Base.metadata.create_all)
-2. schemas.py - Pydantic schemas (Create + Response for each entity, use ConfigDict(from_attributes=True))
-3. main.py - FastAPI application with full CRUD endpoints for each entity
-4. test_main.py - Pytest tests using TestClient with separate test database and dependency override
-5. pyproject.toml - Project configuration with dependencies
+1. models.py — SQLAlchemy 2.0: DeclarativeBase, Mapped, mapped_column (NOT legacy declarative_base)
+2. schemas.py — Pydantic v2: ConfigDict(from_attributes=True) (NOT class Config)
+3. main.py — FastAPI CRUD endpoints for each entity
+4. test_main.py — Pytest with TestClient, separate test.db, unique test data per test
+5. pyproject.toml — PEP 621 [project] format (NOT [tool.poetry])
 
 OUTPUT FORMAT — use these exact markers to separate files:
 
@@ -168,18 +186,17 @@ OUTPUT FORMAT — use these exact markers to separate files:
 === pyproject.toml ===
 <toml content>
 
+DOCUMENTATION — every file must have a one-line module docstring. Classes get a one-line docstring. Keep it zensical: say what it IS, not what it does. No filler.
+
 RULES:
-- SQLite: create_engine("sqlite:///./app.db", connect_args={"check_same_thread": False})
-- Each model: auto-increment "id" Column(Integer, primary_key=True, index=True)
-- Schemas: BaseModel with ConfigDict(from_attributes=True) for Response variants
-- Endpoints per entity: POST (create, 201), GET (list), GET by id (404 if missing), PUT (update), DELETE (204)
-- Tests: separate test.db, override get_db dependency, use TestClient
-- pyproject.toml: fastapi, uvicorn[standard], sqlalchemy, pytest, httpx
-- Status fields: String(20) with default, NEVER Enum
+- Follow the REFERENCE IMPLEMENTATION patterns exactly
+- SQLAlchemy 2.0: DeclarativeBase + Mapped + mapped_column (not Column())
+- Python type unions: str | None (not Optional[str])
+- pyproject.toml: PEP 621 [project] format, requires-python = ">=3.14"
+- Tests: unique descriptive data per test, NOT generic "test_title" strings
 - Absolute imports only (from models import ..., from schemas import ...)
-- Python booleans: True/False/None (not true/false/null/none)
 - NO markdown fences inside file content — just raw code
-- Every _id foreign key field MUST have ForeignKey("table.id") constraint`;
+- Only test endpoints that exist in main.py — no extra tests`;
 
 // === Tiedostoparseri LLM-vastauksesta ===
 function parseGeneratedFiles(text) {
@@ -285,7 +302,7 @@ async function runPipeline(model, scenario) {
 
         // 3. LLM-koodigenerointi
         console.log(`    [3/5] Koodigenerointi (LLM)...`);
-        const codePrompt = `PROJECT REQUIREMENTS:\n${req.text}\n\nJSON SPECIFICATION:\n${JSON.stringify(spec, null, 2)}\n\nGenerate the complete project with all 5 files.`;
+        const codePrompt = `${GOLDEN_EXAMPLE}\n---\n\nPROJECT REQUIREMENTS:\n${req.text}\n\nJSON SPECIFICATION:\n${JSON.stringify(spec, null, 2)}\n\nGenerate the complete project with all 5 files. Follow the reference implementation patterns exactly.`;
         const codeResp = await ollamaChat(model, codePrompt, CODE_SYSTEM, 8192);
         timings.push(codeResp);
         writeFileSync(`${dir}/_code_raw.txt`, codeResp.text);