Alexandre Catarino
35b32b401e
API Tests / build (push) Has been cancelled
Benchmarks / build (push) Has been cancelled
Build & Test Lean / build (push) Has been cancelled
Regression Tests / build (push) Has been cancelled
Report Generator Tests / build (push) Has been cancelled
Research Regression Tests / build (push) Has been cancelled
Syntax Tests / build (push) Has been cancelled
Python Virtual Environments / build (push) Has been cancelled
* Add margin-aware option strategy match selection OptionStrategyMatcher.MatchOnce greedily matched definitions in descending leg-count order, never consulting the objective function hook. Books of overlapping debit spreads were carved into ladders whose uncovered short leg is charged naked option margin, producing phantom margin deltas, inconsistent accept/reject decisions and TotalMarginUsed churn on fully covered, defined-risk books. MatchOnce now evaluates a second candidate solution that deprioritizes definitions leaving a short leg uncovered, and selects the best solution via the objective function. The new default objective function minimizes the quantity of uncovered short contracts, a deterministic proxy for the margin required to hold the positions. Ties preserve the previous grouping, so behavior only changes where the greedy carve left a short uncovered that another grouping of the same positions covers. Fixes #9638 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Cache strategy definition ordering and skip redundant match pass Materialize the definition enumerations once per matcher options instead of re-sorting them on every MatchOnce call, and only evaluate the second candidate solution when some short contract can actually be covered by a long of the same right or by the underlying lots held. A book of naked shorts, by far the most common one reaching that point, now runs a single matching pass. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * Bound credit-side short coverage and add overlapping spreads regression The uncovered short proxy treated any same-right long as covering a short leg. A long on the credit side (higher strike for calls, lower for puts) caps the risk at the strike width, which for a distant long can exceed the naked short margin, so preferring it could raise the margin required instead of lowering it. Coverage from the debit side stays free, while credit-side coverage only counts within 10% of the short strike, the price-free stand-in for the naked short margin floor of the option margin model. Beyond that width the short counts as uncovered, the candidate solutions tie and the previous grouping is preserved, so the selection can only ever lower the margin required to hold the positions. Also adds a regression algorithm for the reported defect: two overlapping bull call debit spreads with interleaved strikes resolve into two margin free spreads instead of a bull call ladder charging naked call margin plus an unmatched long. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Skip provably useless match passes and drop scoring allocations Matching again cannot help once the first solution already leaves no more shorts uncovered than the positions can possibly cover, since a long contract covers at most its own quantity of shorts of the same right, and so does an underlying lot. Checking that bound generalizes the naked shorts precondition it replaces and removes the second pass from books holding fewer longs than shorts, such as a plain ladder, which measured 2.2x slower than a single pass before and is now level with it. The credit side width test also subsumes the debit side one, whose width is never positive, so coverage collapses into a single predicate and one pass over the legs. Strategies with a single short leg, which is every spread, butterfly, condor, backspread and covered call, now take a fast path that needs neither ordering nor allocation, and the remaining ladders and short butterflies sort a small array in place instead of allocating lists, objects and sort closures per score. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Require a covering long to outlive the short it covers The coverage proxy compared strikes only, so a short calendar spread, long the near expiration and short the far one at the same strike, read as fully covered on a zero strike width. The margin models disagree: once the long expires the short is naked for the rest of its life, and short calendar spreads are charged the stand-alone naked short margin while ordinary calendar spreads, whose long outlives the short, require none. Requiring the covering long to expire no earlier than the short makes the proxy mirror that distinction exactly, and leaves same expiry books untouched. The skip added for provably useless second passes reads the score as a quantity of uncovered contracts, which only the default objective function guarantees, so a custom one now always gets both candidates. Also documents that the definition ordering is cached, freezing the first output of a user supplied enumerator, and drops the stale claim that nothing in the options type is consulted by the matcher. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Apply the uncovered short bound to the default objective function only A function deriving from the default one is free to score by different rules, so taking its score for a quantity of uncovered contracts could skip a second candidate it would have preferred. Match the type exactly instead, which leaves derived functions always evaluating both. Also documents that the legacy objective function scores are not bounded above by zero, so configuring it ends candidate evaluation and preserves the single matching pass, and describes the regression algorithm strikes by their order in the chain rather than as the highest ones, which only held for a chain of exactly four strikes. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com> Co-authored-by: Martin Molinero <martin.molinero1@gmail.com>