What Penrose is

Penrose is independent peer review for quantitative trading claims. Give it a claim from a paper, a researcher, or a strategy generator and it attempts to disprove the claim before anyone treats the backtest as evidence.

The problem it solves

A strong backtest can emerge from a weak process. If a researcher tests hundreds of signals, changes parameters repeatedly, and reports only the winner, the reported Sharpe ratio no longer means what it appears to mean. Ordinary backtesting tools simulate the chosen strategy; they do not usually account for the entire search that selected it.

Penrose adds that missing inference layer. It records the search population, reconstructs the claim, checks that the implementation is faithful, and subjects the evidence to deflation, power, stability, cost, capacity, and confirmation controls.

What Penrose is not

Penrose is not:

• a broker or order-execution system;
• a portfolio optimizer;
• a source of trading signals;
• an automatic publication authority;
• proof that a surviving strategy will remain profitable.

Its strongest verdict, research-supported, means only that a claim survived the configured falsification process and a one-time confirmation check. A human still decides what happens next.

The compounding asset

Most claims fail. Penrose treats those failures as useful evidence rather than discarded work. A clean invalidation records what was tested, why it failed, its data and reconstruction provenance, and the search context that determined the statistical bar.

Across independent studies, these records can support advisory principles and connections. Those connections may inform researchers, but they never change another claim's verdict automatically.

The design promise

Penrose prefers an honest stop over a fabricated answer:

• missing data becomes needs_data;
• an unfaithful reconstruction becomes cannot_replicate;
• an unresolved implementation becomes pending_module;
• a broken execution becomes engine_error;
• ambiguous or implausible output becomes needs_review.

Only a successfully reconstructed and adequately evaluated claim receives a research verdict.

Next

Continue to the Quickstart, or learn the system model.