quantdata.dysanos.grid.gendysanos.run_incremental_candidate_adversarial_replay#

Search frozen period portfolios over their conditional Gaussian support.

This diagnostic complements finite conditional Monte Carlo. It writes each five-day state continuation as an affine function of independent standard normal shocks and differentiates the complete frozen portfolio P&L through the DYSANOS decoder. Projected Adam searches nested Mahalanobis balls for a loss. A negative minimum is a counterexample to arbitrage; a non-negative minimum is only a numerical search result, not a global certificate.

quantdata.dysanos.grid.gendysanos.run_incremental_candidate_adversarial_replay.project_rows_to_l2_balls_(values, radii)[source]#

Project each row of values onto its corresponding Euclidean ball.

quantdata.dysanos.grid.gendysanos.run_incremental_candidate_adversarial_replay.first_loss_radius_on_ray(pnl_function, shocks, *, iterations=50)[source]#

Bisect the first sign change from zero toward a loss-producing shock.

quantdata.dysanos.grid.gendysanos.run_incremental_candidate_adversarial_replay.run(args)[source]#

Run the automatic-differentiation support search.