Use Python Excellence Prover with your AI.

Validate python excellence

PILLAR I. PYTHONIC: type hints on EVERY function (params + return + generics). Pydantic BaseModel for external data (API, forms, files). dataclasses for internal DTOs. Protocol for structural subtyping. f-strings, pathlib, context managers, match/case, EAFP over LBYL, comprehensions over loops where readable, enumerate() over range(len()). PILLAR II. PERFORMANCE: async/await for ALL I/O (httpx, asyncpg, aiofiles). Generators/itertools for large data (never materialize 1M rows into a list). __slots__ on data-heavy classes. functools.lru_cache for pure functions. Bulk DB operations (executemany, COPY). N+1 prevention (select_related, joinedload). PILLAR III. ZERO TOLERANCE: no bare except (catches SystemExit). No mutable defaults (def f(x=[])). No print() (use structlog/loguru). No os.path (use pathlib). No magic values (use Enum/StrEnum/Settings). No global mutable state. No # type: ignore without specific error code. If rejected, fix the gap before shipping. Structured reflection capability for Python code excellence. forces type safety with Pydantic/mypy, Pythonic idiom enforcement, structured error handling, async-first I/O patterns, and zero-tolerance for anti-patterns before any Python code ships. Grounded in PEP 8, PEP 484 (type hints), PEP 557 (dataclasses), PEP 3156 (asyncio), and Python 3.12+ best practices. Catches Type Erosion (untyped functions that silently pass wrong data. def process_order(data, user, amount): # What type is data? dict? OrderModel? str? # What type is user? User object? user_id string? email? # What type is amount? float? Decimal? int (cents)? str ("$49.99")? result = calculate(amount 1.08) # Tax calculation on... what? If amount is passed as "$49.99" (string from form submission): amount 1.08 raises TypeError at runtime. in production, on a Saturday. Fix: def process_order(data: OrderRequest, user: User, amount: Decimal) -> OrderResult: mypy catches the string×float multiplication at commit time, not production. External data (API, forms, files) MUST pass through Pydantic validation: class OrderRequest(BaseModel): amount: Decimal = Field(ge=0, description="Order total in USD") Internal data uses @dataclass for zero-overhead DTOs. Protocol for structural subtyping. Never Any. Never untyped), Anti-Pattern Blindness (using Python like Java instead of writing Pythonic code. Java-in-Python: for i in range(len(items)): item = items[i]; process(item) Pythonic: for item in items: process(item) Java-in-Python: result = ""; for s in parts: result = result + s Pythonic: result = "".join(parts) Java-in-Python: if x != None: ... Pythonic: if x is not None: ... Java-in-Python: try: d[key] except KeyError: d[key] = default Pythonic: d.setdefault(key, default) or defaultdict os.path.join("data", "file.csv") → pathlib.Path("data") / "file.csv" "Hello, " + name + "!" → f"Hello, {name}!" open(f); try: ... finally: f.close() → with open(f) as handle: ... Every non-Pythonic pattern is a readability tax on every future reader), Error Swallowing (catching exceptions without handling them. try: process_payment(order) except Exception: pass This catches EVERYTHING: ValueError (bad data), ConnectionError (Stripe is down), KeyboardInterrupt (operator pressing Ctrl+C), SystemExit (server shutting down). The payment silently fails. The order shows "completed." The customer is charged $0. Nobody knows until the finance team runs reconciliation 3 days later. Fix: specific exception classes, structured logging, re-raise or return typed error. try: result = stripe.Charge.create(amount=order.total_cents) except stripe.CardError as e: logger.warning("card_declined", order_id=order.id, reason=str(e)); raise PaymentDeclined(order.id, str(e)) except stripe.APIConnectionError: logger.error("stripe_unreachable", order_id=order.id); raise ServiceUnavailable("Payment processor") Custom exception hierarchy: AppError → PaymentError → PaymentDeclined | PaymentTimeout), Sync I/O in Async Context (blocking the event loop with synchronous operations. async def fetch_user_data(user_id: str): response = requests.get(f"https://api.example.com/users/{user_id}") # BLOCKS event loop data = json.loads(open("config.json").read()) # BLOCKS event loop db_result = cursor.execute("SELECT FROM users WHERE id = %s", (user_id,)) # BLOCKS event loop Every synchronous I/O call in an async function blocks ALL concurrent coroutines. 100 concurrent requests become sequential. throughput drops from 100 req/s to 3 req/s. Fix: httpx.AsyncClient for HTTP. aiofiles for file I/O. asyncpg/databases for DB. async with httpx.AsyncClient() as client: response = await client.get(url) async with aiofiles.open("config.json") as f: data = json.loads(await f.read()) result = await conn.fetch("SELECT FROM users WHERE id = $1", user_id)), and Mutable Default Trap (Python's most infamous gotcha. shared mutable defaults. def add_item(item: str, items: list = []): items.append(item); return items add_item("a") → ["a"]. add_item("b") → ["a", "b"]. The default list is created ONCE at function definition, then SHARED across ALL calls. This is not a bug. it is Python's object model. But it is a trap. Fix: def add_item(item: str, items: list | None = None) -> list: if items is None: items = []; items.append(item); return items Same applies to dict defaults, set defaults, and any mutable object. Rule: default arguments must be immutable (None, str, int, float, tuple, frozenset)). Call once per Python module, function, or feature implementation