xingwudao/open-xquant/build-rule

Rule Builder

You help the user add risk and exit logic without overstating current spec support.

Current Execution Model

Rules are Python objects passed to Engine.run(..., rules=[...]) or added by the spec compiler internally. The audited CLI spec path does not support a generic top-level rules: YAML section or oxq backtest run --rules, but it does support a narrow whitelist under portfolio.rules.

In current Engine.step():

• pre-trade rule effects consumed: RuleResult.weights, RuleResult.hold
• post-trade rule effects consumed: RuleResult.target_positions
• declared but not currently consumed by Engine: RuleResult.constraints

Do not tell the user that a constraints field will affect execution unless you have verified the engine path being used consumes it.

Built-in Rules

Inspect the live registry:

uv run python - <<'PY'
import oxq
print(sorted(oxq.list_rules()))
PY

Common built-ins:

• ExitRule
• StopLossRule
• TakeProfitRule
• TrailingStopRule
• MaxDrawdownRisk
• DailyLossLimitRisk
• MaxHoldingsRule
• RebalanceFrequencyRule
• CalendarRebalanceRule
• BlacklistRule

Safe CLI Path

For Crossover specs, compile_run() automatically adds ExitRule using the declared fast and slow columns. For every explicit portfolio.rules item, all required params must be present; do not rely on runtime constructor defaults.

For rebalance throttling by trading-session count, use:

execution:
  rebalance:
    frequency: daily
    interval_days: 5

or:

portfolio:
  rules:
    rebalance:
      type: RebalanceFrequencyRule
      params:
        interval_days: 5

For calendar rebalance, use the execution schedule. The compiler emits CalendarRebalanceRule internally:

execution:
  rebalance:
    frequency: monthly
    schedule: month_start

Supported audited portfolio.rules entries are:

portfolio:
  rules:
    stop_loss:
      type: StopLossRule
      params:
        threshold: 0.05
    take_profit:
      type: TakeProfitRule
      params:
        threshold: 0.20
    trailing_stop:
      type: TrailingStopRule
      params:
        trail_pct: 0.08
    max_drawdown:
      type: MaxDrawdownRisk
      params:
        max_drawdown: 0.15
    daily_loss:
      type: DailyLossLimitRisk
      params:
        max_daily_loss: 0.03
    max_holdings:
      type: MaxHoldingsRule
      params:
        max_holdings: 10

For portfolio target constraints, use:

portfolio:
  constraints:
    max_weight: 0.2
    min_weight: 0.01
    max_holdings: 10
    cash_reserve: 0.05

portfolio.constraints.min_position_value is parsed but blocked by validation because the audited runtime cannot execute it yet.

Use SDK execution for rules outside the whitelist until the SPEC validator, compiler, runtime, and artifact audit support them.

SDK Pattern

from oxq.core import Engine
from oxq.rules import StopLossRule, TakeProfitRule
engine = Engine()
result = engine.run(
    strategy=strategy,
    market=market,
    broker=broker,
    start="2020-01-01",
    end="2024-12-31",
    rules=[
        StopLossRule(threshold=0.05),
        TakeProfitRule(threshold=0.20),
    ],
)

After SDK runs, make sure the user still gets equivalent artifacts or a clear statement that the run is exploratory rather than the standard CLI artifact pipeline.

When A New Rule Is Needed

Load agent/skills/create-component/SKILL.md, then route to agent/skills/create-rule/SKILL.md.

Before creating a rule, confirm:

• pre-trade or post-trade category
• trigger condition
• exact RuleResult field to use
• whether internal state is required
• how it will be tested with hand-crafted portfolio and bar data

Red Lines

• Do not document unsupported YAML rules: as working CLI behavior.
• Do not claim constraints changes execution unless verified in code.
• Do not mutate Portfolio inside evaluate().
• Do not treat a stop-loss rule as a substitute for research audit.