Open research instrumentEducational, not evidence

Find an edge that is not there.

The series below is a driftless random walk. Its true Sharpe ratio is 0, by construction, and no setting of any control can change that. Search it anyway. The lab counts every parameter set you try and deflates your best result against your own search, so you can watch a good-looking number turn into what it actually is.

What this is. A teaching instrument. It runs on a synthetic series with no edge, spends no research hypothesis, writes to no ledger, and produces no ALPHAC performance claim. Nothing a visitor generates here is evidence about any strategy.

Series seed 750 bars, driftless geometric random walk Fast window bars in the fast average Slow window bars in the slow average Cost basis points per unit of turnover

Annualised Sharpe: 0
Max drawdown: 0
Total return: 0
Trades: 0

Your trial ledger

Distinct settings evaluated: 0
Best annualised Sharpe found: 0
Sharpe luck alone would find: n/a
Deflated Sharpe of your best: n/a

Best found at nothing yet

Move a control to record your first trial.

How it works, and why it is synthetic

The execution rule is the engine's

decide at the close of bar t, hold from the open of bar t+1 to the open of bar t+2. the same no-lookahead contract the ALPHAC backtester enforces. The property is re-checked on every run and reported under the chart, rather than asserted once in prose.

Why not real prices

config/data_source_rights_policy.json records every market-data source this project uses as prohibited or withheld for raw redistribution, with a project default of do not bundle raw third-party rows. Shipping real bars into a browser is not available.

On real data a visitor can always argue the edge they found was genuine. On a driftless random walk the true Sharpe is zero by construction, so every good result is provably luck. The demonstration is stronger for it, not weaker.

What the deflation uses

Your trial count is the number of distinct settings you have evaluated on this series. The dispersion is measured across your own results, not assumed. The arithmetic is the same calculator published elsewhere on this site, js/dsr-tool.js, golden-vector tested against the Python source, on the non_excess kurtosis convention, annualised at 252 periods.

The gate

ALPHAC admits a full-union book at a deflated ratio of 0.95. That is the bar your scenario is read against here. Reaching it on a series with no edge is a false positive, and the instrument is built to let you produce one and then see it named.

Sandbox contract. every input is stored in the URL, so a scenario is shareable and reproducible. no write path to any ledger, artifact, broker or published record exists. results are labelled USER_GENERATED_SCENARIO_NOT_RESEARCH_EVIDENCE.

Rendered from selection_risk_lab_contract.json, whose content hash the build reproduces before this page is written. Further reading: the arithmetic of not fooling yourself.