Defined-risk options structures make honest sizing possible because the worst case is a known number at entry. A sizing discipline caps that worst case per position (a few percent of capital), caps the sum of worst cases across open positions, and normalizes contract counts by each stock's actual volatility so risk per position is comparable.
Premium measures reward; margin measures the broker's collateral requirement; neither measures what a position can lose. A credit spread collecting $1.00 on a $5.00 width can lose $4.00, and sizing on the $1.00 makes a losing streak four times larger than intended. Worst-case sizing (width minus credit for spreads, debit paid for debit structures, wider wing minus total credit for condors) is the only base that keeps the arithmetic honest: a per-trade cap of 5% of allocated capital means five simultaneous full losses cost 25%, painful and survivable, by construction rather than by luck.
Because losses cluster. Short-premium positions share a common factor, volatility, and the regimes that break one strike tend to test several. A per-trade cap alone permits unlimited stacking of individually reasonable positions into one large volatility bet; a total-open-risk cap (for example 20% of capital across all worst cases) bounds what a single regime shift can take. The same logic extends to per-symbol limits, one position per underlying, so a single company's news cannot hit twice.
Equal worst-case dollars on a sleepy consumer stock and a volatile semiconductor name are not equal risks per day: the volatile name reaches its worst case more often and faster. Normalizing size by realized volatility, targeting a fixed number of dollars of daily movement per position (computed from something like a 20-day EWMA of returns), evens the pace of risk across the book. The practical effect is smaller contract counts on fast movers, which is exactly the adjustment intuition wants and rarely applies consistently.
On top, as multipliers, not replacements. Volatility regime bands cut size in panic conditions; term-structure inversion cuts it further; portfolio-level caps on aggregate greeks (net delta, vega) refuse positions that concentrate the book even when each passes its own sizing. Layered this way, any single failure of judgment is bounded by the layer above it.