pydantic-models

star 4

"'Pydantic frozen data models for trading: enums, annotated constraints" field/model validators, and computed properties'

paulpas By paulpas schedule Updated 6/4/2026
play_arrow Run Skill in Manus View GitHub

name: pydantic-models compatibility: opencode completeness: 95 content-types:

  • code
  • guidance
  • do-dont
  • examples description: '"''Pydantic frozen data models for trading: enums, annotated constraints" field/model validators, and computed properties''' license: MIT maturity: stable metadata: domain: coding output-format: code related-skills: null role: implementation scope: implementation triggers: enums, frozen, pydantic models, pydantic-models, trading archetypes:
    • tactical
    • generation anti_triggers:
    • brainstorming
    • vague ideation
    • code golf
    • over-engineering response_profile: verbosity: low directive_strength: high abstraction_level: operational version: "1.0.0"

Skill: coding-pydantic-models

Pydantic frozen data models for trading: enums, annotated constraints, field/model validators, and computed properties

Role / Purpose

This skill covers how to define immutable, self-validating data models for a trading system using Pydantic v2. Every model enforces its own invariants at construction time. Once created, a model object is trusted — no downstream code needs to re-validate its fields.

Key Patterns

1. (str, Enum) — JSON-Serializable Enums

All domain enums inherit from both str and Enum. This means enum members serialize directly to their string values in JSON, dicts, and Pydantic payloads — no extra serialization step needed.

from enum import Enum

class SignalType(str, Enum):
    """Signal type - parse at boundary, trust internally."""
    LONG = "long"
    SHORT = "short"
    EXIT_LONG = "exit_long"
    EXIT_SHORT = "exit_short"

class OrderType(str, Enum):
    MARKET = "market"
    LIMIT = "limit"
    STOP = "stop"
    STOP_LIMIT = "stop_limit"
    TRAILING_STOP = "trailing_stop"

class OrderSide(str, Enum):
    BUY = "buy"
    SELL = "sell"

class PositionSide(str, Enum):
    LONG = "long"
    SHORT = "short"
    FLAT = "flat"

2. ConfigDict(frozen=True) — Immutable After Creation

All trading models use frozen=True. Immutability prevents accidental mutation, makes objects safe to cache, and enables use as dict keys or set members.

from pydantic import BaseModel, ConfigDict

class SignalEvent(BaseModel):
    model_config = ConfigDict(frozen=True)
    # fields...

3. Annotated[float, Field(gt=0.0)] — Constraint Patterns

Constraints are encoded at the type level. Pydantic enforces them automatically — no manual guards needed for valid objects.

from typing import Annotated
from pydantic import Field

# Price must be strictly positive
price: Annotated[float, Field(gt=0.0)]

# Confidence must be in [0, 1]
confidence: Annotated[float, Field(ge=0.0, le=1.0)]

# Leverage between 1x and 100x
leverage: Annotated[int, Field(ge=1, le=100)] = Field(default=1)

# PnL percentage bounded to prevent data corruption
pnl_pct: Annotated[float, Field(ge=-1.0, le=10.0)] = Field(default=0.0)

4. @field_validator — Edge Value Detection and Format Validation

Field validators catch boundary values that pass range checks but still indicate data problems.

from pydantic import field_validator

class SignalEvent(BaseModel):
    model_config = ConfigDict(frozen=True)

    timestamp: datetime = Field(default_factory=datetime.utcnow)
    symbol: str = Field(..., pattern=r"^[A-Z/]+$")
    signal_type: SignalType
    confidence: Annotated[float, Field(ge=0.0, le=1.0)]
    price: Annotated[float, Field(gt=0.0)]
    timeframe: str = Field(default="1h")
    metadata: dict[str, Any] = Field(default_factory=dict)
    source: str = Field(default="technical")

    @field_validator("confidence")
    @classmethod
    def confidence_not_edge(cls, v: float) -> float:
        """Fail fast - edge values (0 or 1) indicate parsing issues."""
        if v == 0 or v == 1:
            raise ValueError("Confidence should not be edge value (0 or 1)")
        return v

related-skills: null

5. @model_validator(mode="after") — Cross-Field Validation

Model validators run after all fields are set and validated. Use them for invariants that span multiple fields (OHLC integrity, PnL consistency, component averages).

from pydantic import model_validator

class Candle(BaseModel):
    model_config = ConfigDict(frozen=True)

    timestamp: datetime
    symbol: str = Field(..., pattern=r"^[A-Z/]+$")
    open: Annotated[float, Field(gt=0.0)]
    high: Annotated[float, Field(gt=0.0)]
    low: Annotated[float, Field(gt=0.0)]
    close: Annotated[float, Field(gt=0.0)]
    volume: Annotated[float, Field(ge=0.0)]

    @model_validator(mode="after")
    def candle_data_valid(self) -> "Candle":
        """Fail fast - verify OHLC data integrity."""
        if self.high < self.low:
            raise ValueError("High price cannot be lower than low price")
        if self.high < max(self.open, self.close):
            raise ValueError("High price must be >= max(open, close)")
        if self.low > min(self.open, self.close):
            raise ValueError("Low price must be <= min(open, close)")
        return self

class Trade(BaseModel):
    model_config = ConfigDict(frozen=True)

    id: str = Field(..., pattern=r"^[A-Z0-9_-]+$")
    position_id: str = Field(..., pattern=r"^[A-Z0-9_-]+$")
    symbol: str = Field(..., pattern=r"^[A-Z/]+$")
    side: OrderSide
    entry_price: Annotated[float, Field(gt=0.0)]
    exit_price: Annotated[float, Field(gt=0.0)]
    size: Annotated[float, Field(gt=0.0)]
    entry_time: datetime
    exit_time: datetime
    fees: float = Field(default=0.0)
    pnl: float = Field(default=0.0)
    pnl_pct: Annotated[float, Field(ge=-1.0, le=10.0)] = Field(default=0.0)
    conviction_at_entry: Annotated[float, Field(ge=0.0, le=1.0)]
    conviction_at_exit: Annotated[float, Field(ge=0.0, le=1.0)]
    exchange: str = Field(..., pattern=r"^[a-zA-Z0-9_-]+$")

    @property
    def gross_pnl(self) -> float:
        """Gross PnL before fees - pure function, no mutations."""
        if self.side == OrderSide.BUY:
            return (self.exit_price - self.entry_price) * self.size
        return (self.entry_price - self.exit_price) * self.size

    @model_validator(mode="after")
    def pnl_calculated_correctly(self) -> "Trade":
        """Fail fast - verify PnL matches calculation."""
        expected = self.gross_pnl - self.fees
        if abs(expected - self.pnl) > 0.01:
            raise ValueError(f"PnL mismatch: reported {self.pnl}, expected {expected}")
        return self

6. @property as Pure Functions

Properties on frozen models are pure: they compute derived values without mutations. They can safely be called multiple times with the same result.

class Position(BaseModel):
    model_config = ConfigDict(frozen=True)

    id: str = Field(..., pattern=r"^[A-Z0-9_-]+$")
    symbol: str = Field(..., pattern=r"^[A-Z/]+$")
    side: PositionSide
    entry_price: Annotated[float, Field(gt=0.0)]
    size: Annotated[float, Field(gt=0.0)]
    leverage: Annotated[int, Field(ge=1, le=100)] = Field(default=1)
    entry_time: datetime = Field(default_factory=datetime.utcnow)
    pnl: float = Field(default=0.0)
    fees_paid: float = Field(default=0.0)

    @property
    def value(self) -> float:
        """Position value in quote currency - pure function, no mutations."""
        return self.entry_price * self.size * self.leverage

    @property
    def margin_used(self) -> float:
        """Margin used for position - pure function, no mutations."""
        return self.value / self.leverage

class OrderBookSnapshot(BaseModel):
    model_config = ConfigDict(frozen=True)

    timestamp: datetime = Field(default_factory=datetime.utcnow)
    symbol: str = Field(..., pattern=r"^[A-Z/]+$")
    exchange: str = Field(..., pattern=r"^[a-zA-Z0-9_-]+$")
    bids: list[tuple[float, float]] = Field(default_factory=list)
    asks: list[tuple[float, float]] = Field(default_factory=list)
    mid_price: float = Field(default=0.0)

    @property
    def spread(self) -> float:
        """Order book spread - pure function, no mutations."""
        if not self.asks or not self.bids:
            return 0.0
        return self.asks[0][0] - self.bids[0][0]

7. ConvictionScore — Component Average Validation

The overall technical score must equal the weighted average of its components, enforced by a model validator. Floating-point tolerance of 0.01 accommodates rounding.

class ConvictionScore(BaseModel):
    model_config = ConfigDict(frozen=True)

    overall: Annotated[float, Field(ge=0.0, le=1.0)]
    technical: Annotated[float, Field(ge=0.0, le=1.0)]
    momentum: Annotated[float, Field(ge=0.0, le=1.0)]
    trend: Annotated[float, Field(ge=0.0, le=1.0)]
    volatility: Annotated[float, Field(ge=0.0, le=1.0)]
    volume: Annotated[float, Field(ge=0.0, le=1.0)]

    @model_validator(mode="after")
    def weights_sum_to_one(self) -> "ConvictionScore":
        """Fail fast - internal weights must be consistent."""
        total = (self.momentum + self.trend + self.volatility + self.volume) / 4
        if abs(total - self.technical) > 0.01:
            raise ValueError(
                f"Technical score ({self.technical}) should be average of components"
            )
        return self

8. RiskMetrics — Performance Statistics Model

class RiskMetrics(BaseModel):
    model_config = ConfigDict(frozen=True)

    max_drawdown: Annotated[float, Field(ge=0.0, le=1.0)]
    daily_var_95: float = Field(default=0.0)
    daily_var_99: float = Field(default=0.0)
    sharpe_ratio: float = Field(default=0.0)
    sortino_ratio: float = Field(default=0.0)
    win_rate: Annotated[float, Field(ge=0.0, le=1.0)] = Field(default=0.0)
    profit_factor: float = Field(default=1.0)
    expected_value: float = Field(default=0.0)
    max_consecutive_losses: int = Field(default=0)
    max_consecutive_wins: int = Field(default=0)

    @field_validator("profit_factor")
    @classmethod
    def profit_factor_valid(cls, v: float) -> float:
        """Fail fast - invalid profit factor."""
        if v < 0:
            raise ValueError("Profit factor cannot be negative")
        return v

9. AccountBalance — Balance Snapshot

class AccountBalance(BaseModel):
    model_config = ConfigDict(frozen=True)

    timestamp: datetime = Field(default_factory=datetime.utcnow)
    exchange: str = Field(..., pattern=r"^[a-zA-Z0-9_-]+$")
    total_balance: float = Field(default=0.0)
    free_balance: float = Field(default=0.0)
    used_balance: float = Field(default=0.0)
    equity: float = Field(default=0.0)
    margin_used: float = Field(default=0.0)
    margin_available: float = Field(default=0.0)
    positions_mkt_value: float = Field(default=0.0)
    unrealized_pnl: float = Field(default=0.0)
    realized_pnl: float = Field(default=0.0)

Code Examples

Creating a Signal and Publishing It

from apex.core.models import SignalEvent, SignalType

# Parsed at the boundary — raises immediately if invalid
signal = SignalEvent(
    symbol="BTC/USDT",
    signal_type=SignalType.LONG,
    confidence=0.82,     # Must not be 0 or 1
    price=65_000.0,
    timeframe="1h",
    source="ma_crossover",
    metadata={"momentum_score": 0.75},
)

# Internally: trusted, typed, immutable
print(signal.signal_type)   # SignalType.LONG
print(signal.confidence)    # 0.82

Building a Candle With OHLC Integrity

from apex.core.models import Candle
from datetime import datetime

# Pydantic enforces OHLC invariants at construction
candle = Candle(
    timestamp=datetime.utcnow(),
    symbol="ETH/USDT",
    open=3500.0,
    high=3600.0,  # must be >= max(open, close)
    low=3450.0,   # must be <= min(open, close)
    close=3580.0,
    volume=12345.6,
)

Philosophy Checklist

  • Early Exit: @field_validator and @model_validator halt object construction on invalid data
  • Parse Don't Validate: All validation happens at model instantiation; code receiving a model instance trusts it
  • Atomic Predictability: Frozen models + pure @property methods; same inputs always give same outputs
  • Fail Fast: Edge values, cross-field inconsistencies, and format violations all raise ValueError immediately
  • Intentional Naming: confidence_not_edge, candle_data_valid, pnl_calculated_correctly — validators read like English assertions

Constraints

MUST DO

  • Include at least one BAD/GOOD code example pair
  • Reference a relevant standard (OWASP, SOLID, DRY, KISS, etc.)
  • Use type hints on all function signatures

MUST NOT DO

  • Use magic numbers or hardcoded configuration values
  • Bypass error handling for assumed-valid inputs
  • Write functions longer than 50 lines without decomposition

Live References

Authoritative documentation links for this skill's domain. The model follows markdown links at load time to resolve external references and inline content.

Install via CLI
npx skills add https://github.com/paulpas/agent-skill-router --skill pydantic-models
Repository Details
star Stars 4
call_split Forks 0
navigation Branch main
article Path SKILL.md
More from Creator
paulpas
paulpas Explore all skills →