Option-proxy design: BSM, delta proxy, and VIX-scaled IV versus real SPXW quotes

The research ladder prices option exposure three increasingly honest ways — delta proxy, BSM with a VIX-scaled IV, BSM with real/fitted IV — and reserves the tradability verdict for real quote-chain replay. Each proxy has a defined role and a defined ceiling.

Why proxies exist at all

Real OPRA tick data costs money and time; a parameter sweep over thousands of configurations cannot price every leg against real quotes. The Donchian research engine therefore runs a ladder: Layer 1 establishes whether the signal has an edge in SPX points; Phase 6 maps surviving candidates onto approximate option PnL; Phase 7 replays the short list against real SPXW bid/ask. The non-negotiable interpretation rule: a Layer 1 pass is a ticket to Phase 6, never evidence of 0DTE profitability.

The three proxies and their ceilings

Delta proxy. gross_pnl = |entry_delta| × underlying_pnl_points. Linear, no gamma or theta, nearly free to compute. Good for ranking configurations and sensitivity sweeps; meaningless as absolute PnL — by construction it cannot see convexity, which is the point of being long 0DTE.
BSM with VIX-scaled IV. iv = VIX/100 × tod_adj × regime_adj, with a time-of-day table (1.5 at the open, 1.2 midday, 1.8 in the last half hour) and a VIX-regime factor (1.1 low / 0.9 high). Explicitly classified diagnostic-only in the spec — it can justify buying real data, never a “strong pass”.
BSM with real-chain or fitted intraday IV. The only IV sources allowed to support a strong pass. Even then, entries and exits get spread haircuts by moneyness (±3% ATM with a $0.05 floor, ±8% at 5-point OTM, ±20% at 15-point OTM), and a T_min sensitivity grid (5/15/30/60 minutes) must not flip the sign — if it does, the verdict is inconclusive_requires_real_chain.

What real quotes add that proxies cannot

Phase 7 replays trades against the actual SPXW chain with quote-quality filters (locked/crossed removed, maximum quote age 0.25–5 seconds, default 1.0) and pessimistic execution (buy at ask, sell at bid). Two things consistently surprise relative to proxies: fill feasibility — at a 0.25s freshness threshold some trades simply have no quote to fill against — and premium level. Measured against real quotes, BSM-with-VIX1D entry premium was off by a median factor of 1.85 (see the under-pricing note); no time-of-day multiplier table would have predicted that cleanly.

Practice

Use the delta proxy to rank, VIX-scaled BSM to diagnose, real/fitted-IV BSM to shortlist, and real quotes to decide.
Make the proxy’s ceiling explicit in the output schema — the engine literally tags results diagnostic_only: true so a proxy number cannot be quoted as a tradability claim later.
Budget for the real-data step from the start. In the negative-result study the full-population download cost ~$280 — cheap relative to the false conclusion it prevented.