M2: strategy engine and optimizer (pages 5-6)

[SecondOrderEdge]

Summary

Implements M2 — Strategy engine and optimization on top of the merged M1 library.

• osl.strategy.legs — Leg/Strategy dataclasses (OptionType/Side enums) with stock support and signed P&L at the primary expiry (later-dated legs repriced via BSM for calendars/PMCC).
• osl.strategy.strategies — 22 builders via a ChainContext: long call/put, verticals, covered call, cash-secured put, collar, protective put, straddles/strangles, iron condor/fly, broken-wing butterfly, ratio spread, jade lizard, calendar, diagonal, PMCC.
• osl.strategy.metrics — breakevens, bounded/unbounded max P/L, POP under four labelled measures (delta proxy, RN closed-form, RN Monte Carlo, empirical bootstrap), EV (closed-form edge + RN MC with 95% CI + empirical), expected shortfall, return-on-risk, aggregated greeks, convexity. POP is never shown without an EV and tail-loss companion.
• osl.strategy.liquidity — score (spread/OI/volume/ATM) + slippage; zero-bid legs penalized.
• osl.strategy.enumerate — view-aware candidate enumeration; excludes zero-bid legs by default.
• osl.strategy.optimizer — nine ranking objectives; pop_capped never ranks unbounded risk first.
• osl.pricing.monte_carlo — antithetic GBM terminal sampler.
• osl.viz + pages 5–6 — scatter/radar builders; Strategy Generator (top-N per objective, radar) and Strategy Optimizer (sortable ranking, EV/POP-vs-risk frontiers).

Test plan

• ruff / black / strict mypy osl — clean
• pytest — 138 tests pass, 89% coverage, including on CI's managed standalone interpreter (3.11.13)
• Acceptance: iron-condor breakevens match the closed form; uncovered short strangle reports ∞ max loss (never a finite number); POP delta/RN/MC agree within 2pp on a long call; a deep-OTM short put spread shows POP > 90% with EV < 0; optimizer ranking is deterministic and objective-dependent; zero-bid contracts excluded from enumeration unless explicitly allowed
• Streamlit pages 5–6 compile-checked only (no browser in CI)

Honesty notes

• EV here is risk-neutral (edge vs fair value + RN MC); real-world/GARCH EV and the Probability Lab arrive in M4.
• POP labels make the measure explicit; the delta proxy is documented as overstating POP.

https://claude.ai/code/session_011QQZCSnQXWFafw3MEnMCHe

---

Generated by Claude Code