Skip to content

Repository files navigation

prism-di-public-evaluation

Public claim-bearing artifacts for:

Tax-Aware Direct Indexing at Book Scale: A Batched Evaluation of Speed and After-Tax Value (Ghosh, 2026).

Asymmetry PRISM is evaluated only through externally observed input-output behavior: public problem data, returned weights, status codes, timings, quality diagnostics, after-tax wealth outcomes, failure modes, and audit records. Implementation details remain outside the public artifact.

Contents

  • results/ — sanitized claim-bearing result tables (CSV), one file per evidence artifact.
  • tables/ — published paper tables in machine-readable form.
  • figures/ — figure files as published (vector PDF).
  • validation/ — scripts that recompute external quality-gate residuals from public inputs and returned weights.
  • evidence_ledger/ — claim-to-evidence mapping and artifact policy.
  • environment/ — hardware and software disclosure for the recorded package.
  • data_sanitized/ — price-history manifest and universe list; no client data, no engine state.
  • paper/ — arXiv version and source.

Evidence artifacts

Artifact Use
e1_wholebook_timing_grid whole-book cold and re-solve timing, 7 regimes x 4 book sizes x 5 configurations
e2_singleaccount_probe single-account timing and quality, n=665, 4 configurations, 20-solve median
e3_quality_gate_diagnostics weight-RMSE and tracking-error-tail residuals against the independent reference, all cells
e4_additional_incumbent_configs tighter Clarabel (tol 1e-9) and OSQP cold and seeded re-solve, gap-closure measurements
e5_heldout_aftertax_wealth after-tax terminal wealth, 7 funding years x 2 client profiles x 4 strategies
e6_solver_configuration_ensemble seven-configuration path-noise ensemble on after-tax wealth

Claim policy (summary)

  • Timing and quality-gap claims are claim-bearing only where a reference solver completed on the same public objective, in a declared lane.
  • The single-account case is reported as a loss and is never folded into the whole-book speedup.
  • After-tax wealth results are a held-out backtest (workflow sanity check), not a claim about live trading performance or future returns.
  • Hyperparameters governing the tax-aware objective are frozen on 2016-2017 data, disjoint from every evaluation window reported.

See evidence_ledger/artifact_policy.md for the full policy.

About

Public evaluation artifacts for Asymmetry PRISM on tax-aware direct indexing at book scale: batched account solves, speed and after-tax value, quality diagnostics, failure modes, and audit records.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages