A Transformer that reads the regime
Tensor’s custom architecture conditions forecasts on whether the market is trending up, trending down or ranging. The same price pattern can mean different things in different market states.
AlphaNet Strategy Series / 2026
Archimedes.
A new standard in
deep-learning directional trading.
The market is non-stationary and chaotic. Adapt every decision based on a dynamic system.
A premium strategy series developed by Tensor, with regime-aware intelligence from entry to exit.
01 / The philosophy
Give me a
place to stand.
Markets are non-stationary and chaotic. Archimedes is designed to adapt its decisions as the relationships between price, liquidity and volatility change.
Archimedes pairs a proprietary, regime-aware time-series Transformer with an adaptive, unsupervised exit engine. It reads trend, range and volatility before deciding how to enter, size, hold and exit a position.
Each preview strategy is a portfolio of complementary long and short sub-strategies, spanning trend, mean-reversion and volatility regimes, with upgraded shorting capability relative to earlier engines such as Hackworth.
4Preview markets
2,000+Engineered features
2,000,000+Trained model variations
02 / The engine
Intelligence at
every decision.
From reading the market to managing the last fill, Archimedes brings prediction, portfolio construction and execution into one adaptive system.
An adaptive exit engine
A proprietary, lightweight unsupervised model continuously re-evaluates open positions. Across the preview strategies, 72–83% of exits were model decisions rather than stop-loss triggers.
2,000+ engineered features
Curated inputs span price action, market microstructure, volatility and cross-asset context, giving the model a broad view of the state it is trading.
Complementary long and short books
Each strategy combines directional, mean-reversion and volatility sleeves, with balanced long/short exposure and improved shorting capability relative to earlier engines such as Hackworth.
Deep-RL execution on Hyperliquid
A deep reinforcement learning execution layer uses dynamic TWAP to slice orders, accelerating when liquidity is deep and backing off when it is thin.
Signal-level benchmarking
Every sub-strategy is evaluated independently through standalone Sharpe, parameter perturbations, trade bootstraps, matched-control exit benchmarks and year-by-year stability checks.
03 / The preview strategies
Four markets.
One standard.
Four frozen portfolios, evaluated under the same protocol. Explore the historical results, then open each strategy for its equity curve and benchmark findings.
Research and live testing period Jan 2024 - Sep 2026INITIAL EQUITY INDEXED TO 100%300% TARGET GROSS EXPOSURE
| Strategy | Cumulative return | Sharpe ratio | Max drawdown | Win rate | Profit factor | Trades |
|---|---|---|---|---|---|---|
| BNB | +1,509% | 2.89 | −21.8% | 59.5% | 1.50 | 2,184 |
| ETH | +1,536% | 3.54 | −16.3% | 49.6% | 1.80 | 1,287 |
| SOL | +1,525% | 3.31 | −15.4% | 55.9% | 1.43 | 2,297 |
| XMR | +5,513% | 3.87 | −21.8% | 65.4% | 1.66 | 3,692 |
Frozen sub-strategy portfolios; no re-optimization during the test window. Baseline conditions, with Hyperliquid execution using deep-RL dynamic TWAP. These are backtest results, not live returns. Historical performance does not guarantee future results.
Equity indexed to 100%
FROZEN PORTFOLIOS / LOG SCALE
Reading the numbers
Sharpe ratioReturn relative to risk. The preview series spans 2.89–3.87.
Maximum drawdownThe largest peak-to-trough decline in the account’s value during the backtest.
Win rateThe share of trades that closed profitably.
Profit factorGross profits divided by gross losses. A factor of 1.50 means gross profits were 150% of gross losses.
BNBThe blue-chip workhorse+1,509%RESEARCH PERIOD RETURN
2.89Sharpe ratio
−21.8%Max drawdown
59.5%Win rate
1.50Profit factor
2,184Trades
BENCHMARK HIGHLIGHTS
- Equity indexed from 100% to 1,609% over the research period, across rising, falling and flat markets.
- At a one-hour horizon, model exits avoided further losses in 56% of 1,743 matched cases. The brochure reports a confidence interval entirely below zero for the subsequent returns of closed positions.
- All signals exceeded standalone Sharpe 1.0 (range 1.12–2.15). Resampling checks supported consistency across historical trade samples.
CONSISTENCY & ROBUSTNESS
- All sub-strategies were profitable in 2024 and 2025; most were profitable in 2026 to date.
- Parameter tests demonstrated robustness under the tested stress conditions.
- 80% of exits were model decisions rather than stop-loss triggers.
Frozen portfolio backtest, Jan 2024 – Sep 2026 · initial equity indexed to 100% · 300% target gross exposure · baseline conditions. Exit benchmark: 1,743 matched control events. Historical results do not guarantee future performance.
ETHThe precision engine+1,536%RESEARCH PERIOD RETURN
3.54Sharpe ratio
−16.3%Max drawdown
49.6%Win rate
1.80Profit factor
1,287Trades
BENCHMARK HIGHLIGHTS
- Equity indexed from 100% to 1,636%, with a Sharpe ratio of 3.54 and a maximum drawdown of 16.3%.
- A 49.6% win rate combined with a 1.80 profit factor: gross profits were 180% of gross losses.
- All signals exceeded standalone Sharpe 1.0 (range 1.18–2.54), with the highest median in the series. Resampling checks supported consistency across historical trade samples.
CONSISTENCY & ROBUSTNESS
- Most sub-strategies were profitable in each of 2024, 2025 and 2026 to date, including all in 2025.
- Parameter tests demonstrated robustness under the tested stress conditions.
- 83% of exits were model decisions, the highest share in the series.
Frozen portfolio backtest, Jan 2024 – Sep 2026 · initial equity indexed to 100% · 300% target gross exposure · baseline conditions. Exit benchmark: 1,065 matched control events. Historical results do not guarantee future performance.
SOLThe smooth operator+1,525%RESEARCH PERIOD RETURN
3.31Sharpe ratio
−15.4%Max drawdown
55.9%Win rate
1.43Profit factor
2,297Trades
BENCHMARK HIGHLIGHTS
- Equity indexed from 100% to 1,625%. Its 15.4% maximum drawdown was the smallest in the preview series.
- All sub-strategies were profitable in both 2025 and 2026 to date.
- All signals exceeded standalone Sharpe 1.0 (range 1.01–2.08). Resampling checks supported consistency across historical trade samples.
CONSISTENCY & ROBUSTNESS
- Parameter tests demonstrated robustness under the tested stress conditions.
- Long and short books span trend, mean-reversion and volatility sleeves.
- The backtest contained 2,297 trades, with profits distributed across many trading decisions.
Frozen portfolio backtest, Jan 2024 – Sep 2026 · initial equity indexed to 100% · 300% target gross exposure · baseline conditions. Exit benchmark: 1,645 matched control events. Historical results do not guarantee future performance.
XMRThe high-octane outlier+5,513%RESEARCH PERIOD RETURN
3.87Sharpe ratio
−21.8%Max drawdown
65.4%Win rate
1.66Profit factor
3,692Trades
BENCHMARK HIGHLIGHTS
- Equity indexed from 100% to 5,613%, the highest cumulative return in the four-strategy preview.
- The series’ highest Sharpe ratio (3.87) and win rate (65.4%), across 3,692 trades.
- The exit engine avoided further downside in over 55% of 2,870 matched cases. All signals exceeded standalone Sharpe 1.0 (1.02–2.07). Resampling checks supported consistency across historical trade samples.
CONSISTENCY & ROBUSTNESS
- Parameter tests demonstrated robustness under the tested stress conditions.
- All sub-strategies were profitable in 2024 and 2026 to date; most were profitable in 2025.
- Each sub-strategy signal was benchmarked independently.
Frozen portfolio backtest, Jan 2024 – Sep 2026 · initial equity indexed to 100% · 300% target gross exposure · baseline conditions. Exit benchmark: 2,870 matched control events. Historical results do not guarantee future performance.
04 / The proving ground
Edge, examined
signal by signal.
A strong portfolio result is only the beginning. Each selected sub-strategy signal was evaluated independently.
- Standalone edge
Each signal exceeded a standalone Sharpe ratio of 1.0.
- Matched-control exits
Model exits were compared with matched “what if we held” controls to measure the value of closing the position.
- Parameter perturbation
Parameter tests demonstrated robustness under the tested stress conditions.
- Trade bootstrap
Trade histories were resampled to assess whether results remained consistent across different historical samples.
- Time stability
Every sub-strategy was re-scored year by year, covering 2024, 2025 and 2026 to date.