Verify the record.
An outcome-resolved, forward-accrued track record of the agent's strategies — not a backtest fit to the past, but a forward equity curve scored on days that happened after each strategy went live. The portfolio's hit rate, average R, profit factor and expectancy, a per-strategy leaderboard — each figure independently verifiable. The forward-accrued headline publishes only once outcome-resolved returns land; until then the sealed forward test below is the live, accruing receipt.
The strategy track record publishes only outcome-resolved, forward-accrued returns, and none have accrued yet. The sealed forward test below is the live receipt. Nothing is simulated in its place.
raw JSON →Real backtest — 8-year history, out-of-sample vs buy-and-hold
realThe answer to "does the strategy win?" — every strategy run through a no-look-ahead walk-forward backtest on real multi-year history across 10 liquid names. 15 of 30 strategy×stock tests beat buy-and-hold out-of-sample (50%). Buy-and-hold itself returned a median +70.30% (Sharpe 0.43) — a high bar we don't hide.
| Strategy | Beat B&H (OOS) | Median return | Median Sharpe | Win rate | Trades |
|---|---|---|---|---|---|
| Trend (50/200 SMA cross) | 4/10 | +99.90% | 0.57 | +50% | 5 |
| Mean reversion (RSI) | 7/10 | +79.30% | 0.44 | +75% | 22 |
| 12-month momentum | 4/10 | +50.50% | 0.37 | +39% | 29.5 |
Real walk-forward backtest (no look-ahead; round-trip costs) on real market daily history. Win-rate is shown but is NOT the headline — beating buy-and-hold out-of-sample is. Your decisions, your risk.
Live forward test — a sealed paper basket, marked to real closes
realThe forward proof — an equal-weight 9-name basket sealed at REAL entry closes on 2026-07-19 (data-hash 5096b28fb0a44dcf), then marked to the live close on every load. Nothing is re-fit; the realized return vs SPY grows over real days.
| Name | Entry | Now | Return |
|---|---|---|---|
| AAPL | 333.74 | 333.74 | +0.00% |
| MSFT | 393.82 | 393.82 | +0.00% |
| NVDA | 202.81 | 202.81 | +0.00% |
| AMZN | 247.23 | 247.23 | +0.00% |
| GOOGL | 346.77 | 346.77 | +0.00% |
| META | 646.01 | 646.01 | +0.00% |
| JPM | 341.10 | 341.10 | +0.00% |
| XOM | 147.36 | 147.36 | +0.00% |
| UNH | 426.09 | 426.09 | +0.00% |
Equal-weight, equal-conviction. Entry prices are sealed in the DB (re-derivable from the data-hash); current prices are real closes fetched fresh on every load. No realized number is ever fabricated — day-0 is exactly 0% and the book diverges from SPY only as real prices move. Your decisions, your risk.
The 100-stock backtest
running…Running a mean-reversion strategy across 100 large-caps on 12 years of real history — thousands of round-trip trades. Computing…
Drawdown control — a risk-managed portfolio vs buy-and-hold
real · backtestDrawdown is a risk problem, not a signal problem. A trend signal sized inverse-to-volatility, vol-targeted, with a drawdown circuit-breaker that de-risks as equity nears the cap — backtested on 8-year real history (no look-ahead, round-trip costs). The result: max drawdown held to 4.8% (target ≤ 8% ✓) — vs SPY's 25.4% — at a comparable Sharpe. Monthly win-rate is 54.7%; an 80% bar is unrealistic for any equity strategy, and we won't fake it.
Drag either lever to explore the real trade-off — a tighter cap holds drawdown lower but gives up return; a higher vol target takes more risk for more return until the drawdown brake bites. Every number is a real re-run of the backtest for that (cap, vol-target), not a fabrication.
| Metric | Risk-managed | SPY buy & hold |
|---|---|---|
| Max drawdown | 4.8% | 25.4% |
| Sharpe (risk-adjusted) | 0.31 | 0.66 |
| CAGR | 1.0% | 10.5% |
| Annualized volatility | 3.4% | 17.6% |
| Monthly win-rate | 54.7% | — |
Risk-managed = vol-targeted (10.0%) inverse-vol portfolio with a 8.0% drawdown brake, rebalanced every 5d, 3bps round-trip, over 1076 real trading days. The point isn't to out-return SPY — it's SPY-class risk-adjusted return at a fraction of the drawdown. Your decisions, your risk.