Welcome, S7.
S7 began as a falsification project, not a product. The question was simple: does the market edge that free data and AI are supposed to unlock actually exist? It runs on real capital, under one rule: a strategy stays alive only on out-of-sample evidence.
Five families of strategy were tested against the S&P 500 from 2007 to 2026, survivorship-free and pre-registered: which stocks to own, how much to hold of each, when to rotate between stocks, bonds, and cash, whether momentum adds an edge, and whether spreading across countries helps. Each was scored under a deflated Sharpe ratio, walk-forward validation, and a single one-shot holdout. None cleared the bar on both return and risk. No free-data edge in picking or timing survived honest testing.
What survived was risk control. A regime signal, built from price trend and credit spreads, moves the book to cash when the market breaks and sits tight through ordinary dips. That single rule is what separates the two lines in the chart. A related momentum rule, applied to eight foreign markets without any refitting, cut drawdown in every one it had never seen.
| Metric | S&P 500 | S7 |
|---|---|---|
| Return / yr | 10.8% | 9.8% |
| Worst drawdown | −55.2% | −28.4% |
| Sharpe | 0.55 | 0.68 |
| Time defensive | 0% | 21% |
The claim is narrow by design. S7 does not beat the market on raw return; stepping to cash costs upside while the market climbs. What it buys is loss control: a drawdown roughly half as deep, and the discipline to not sell at the bottom. S7 runs live on real capital, and every change ships with its evidence.