Question 1

What is PRISM by Asymmetry Computing?

Accepted Answer

PRISM is a tax-aware portfolio-optimization engine that prices an entire book of personalized accounts before the open — deterministically, auditably, with no per-seat solver license. Measured on real US-equity data (Mar–Apr 2026): ~10× fleet throughput (100,000 accounts in ~11.6 minutes), a 100,004-asset real backtest at ~1.8s per rebalance (Sharpe 0.737), and the full ~$238k tax budget captured on a $5M book.

Question 2

How does PRISM perform against published baselines?

Accepted Answer

An exact commercial solver times out near 5,000 names (~33s) and cannot price a 100,000-account book before the open. PRISM runs that workflow at ~144 accounts/second/core — about 10.2× a commercial CPU baseline — deterministically and with no license, and it parallelizes linearly. Every figure traces to a recorded run on real US-equity data, with losing cases kept in.

Question 3

How does PRISM compare to other solver options?

Accepted Answer

PRISM is a routed production engine rather than a single point result. It is competitive with an exact commercial baseline on after-tax outcome (matching it in 58–63% of regimes, winning outright in ~33%), beats open-source solver baselines ~88% and rules-based tax-loss harvesting ~95%, and completes the full 100,004-asset path where exact solvers time out well before that ceiling.

Question 4

What is the solution quality of PRISM?

Accepted Answer

PRISM matches the exact reference optimum to ~1e-7, and on after-tax outcome is competitive with an exact commercial baseline — matching it in 58–63% of regimes and winning outright in ~33%. Outputs are deterministic and content-hashed, so every result is reproducible and auditable. PRISM is marketed around auditable, production-ready outcomes rather than speed-only tradeoffs.

Question 5

What asset scale can PRISM handle?

Accepted Answer

PRISM runs a full 100,004-asset real backtest spanning 47+ exchanges at ~1.8 seconds per rebalance (walk-forward Sharpe 0.737, max drawdown −4.22%), and prices 100,000 personalized accounts in ~11.6 minutes on a single core versus ~118 minutes for a commercial CPU baseline — parallelizing linearly across cores.

Question 6

What is direct indexing and how does PRISM support it?

Accepted Answer

Direct indexing is the account-level construction of a personalized portfolio from individual securities rather than a pooled fund, so each account can be tax-managed and customized. PRISM is the optimization core for that workflow — tax-aware by design, with wash-sale handling and lot-level accounting — scaling from a single account up to a 100,000-account book priced before the open.

Question 7

How is PRISM deployed and integrated?

Accepted Answer

PRISM is delivered as a REST API with a Python SDK, deployable in dedicated cloud, your VPC, or on-prem / air-gapped — with no per-seat solver license. Every solve produces a deterministic, replayable audit log, and an exact-reference comparator ships so you can validate PRISM against ground truth on your own data.

Question 8

What is PRISM-Q?

Accepted Answer

PRISM-Q is Asymmetry Computing's research lane on its long-term R&D roadmap. It is an honest research track that adds zero tax alpha today — PRISM's production speed and tax results are classical and measured. We publish outcomes only; the methods are proprietary.

Question 9

What is a matched-workload pilot with Asymmetry Computing?

Accepted Answer

A matched-workload pilot is a 30-day, buyer-owned benchmark of PRISM against your current production stack on your own universe, constraints, costs, and tax rules, with a pass/fail metric agreed before kickoff. Deliverables include a full results pack (every account, losing cases shown), deterministic audit logs, an exact-reference comparator on your data, and an ROI computed with your real numbers.

PRISM — Tax-Aware Portfolio Optimization at Fleet Scale | Asymmetry Computing