Risk Controls Belong Before Order Construction

Most trading systems implement risk as a check: size the position, then verify that the resulting risk is acceptable, then either proceed or complain. The position already exists by the time the check runs, which means the check has to be obeyed rather than enforced.

Moving the same controls earlier changes their nature. If position size is derived from the stop distance and the risk cap, then the cap is not a limit applied to a size — it is the definition of the size.

automated_trading_os enforces a 1% per-trade risk cap, a minimum 1:4 R:R floor, fee and slippage modelling, and a cost-budget filter — all before order construction. A setup that fails the R:R floor or the cost budget does not get reduced; it never becomes an order.

This matters because it changes the admitted-trade population, not just the P&L of the surviving trades. The distribution you end up analysing is a different distribution from the one a post-hoc filter would produce.

The same structural argument applies to timeframes. Higher timeframe context establishes direction, mid timeframe maps structure, lower timeframe times the entry, and no lower layer may override a higher constraint. Contradictory signals are eliminated by construction rather than resolved by a scoring function.

A scoring function that weighs conflicting timeframes is a discretionary decision wearing a numeric costume. Enforcing one-way flow removes the decision entirely.

The pipeline exists; the attribution study does not. Which of these controls contributes most — the risk cap, the R:R floor, or the cost budget — is an open question, and the ablation work to answer it is registered rather than finished.

Until that study reports, the honest claim is about structure: every admitted trade satisfies every constraint by construction. It is not a claim about profitability.