BrightHedge AI
Systematic trading on our own capital. Fully automated. Track record in progress.
We build daily machine-learning models, prove them on data they have never seen, and only then trade them with our own money. Code places every order.
Status as of 27 September 2026 · Live figures re-derived from broker records (Alpaca, OKX) on that date
What we do
- Own capital only. No client money, no fund.
- One decision per strategy per day. Tabular ML models score daily market features. Code turns each score into an order, a stop and an exit. There are no discretionary trades.
- Trading with capital today: Hash, a Bitcoin perpetual future on OKX at 3× leverage that can hold for several days. Steadfast V2 and Maverick V2 are single US stocks on Alpaca, split 50/50, bought at the open and flat by the close.
- In testing or paused: five more strategies across gold, FX, copper and US equities. See the pipeline section below.
Live results
Real fills, real money. One trade = one position from entry to exit.
| Strategy | Live since | Trades | Win rate | Avg win | Avg loss | Profit factor | Sum of trade returns | Worst trade |
|---|---|---|---|---|---|---|---|---|
| Hash | 25 Jun 2026 | 35 | 62.9% | +7.4% | −3.4% | 4.31 | +119.8% | −6.4% |
| Steadfast V2 | 27 Jul 2026 | 27 | 55.6% | +3.7% | −2.5% | 1.56 | +26.0% | −3.6% |
| Maverick V2 | 18 Sep 2026 | 0 | No trades yet. Its market filters kept it in cash on all trading days since launch. | |||||
- How returns are measured. Each trade's return is its profit on the capital committed to it, as traded (Hash at 3×). The total is the plain sum of per-trade returns. It is not compounded and ignores position size, so it is not an account return. Profit factor is gross dollar profit ÷ gross dollar loss.
- Costs. Hash is net of exchange fees and funding. The stock broker charges no commission; small regulatory fees on sales are not deducted.
- Hash history. Pilot-size positions (about $70) ran until 12 Jul 2026, and full size after that. Per-trade percentages don't depend on size. Nine manual connectivity-test fills (21–24 Jun) are excluded.
- Late executions. Three Hash entries were placed 6–9 hours late by re-running that day's model signal. On 12 Jul and 19 Aug the scheduled order was rejected for insufficient margin; the 19 Aug position became the largest winner (+42.1%). On 28 Aug a network outage blocked the scheduled run.
- Concentration. Without the +42.1% trade, Hash's sum of trade returns is +77.7%.
- Sample size. About three months of live history for Hash and two for Steadfast V2. That is too short to draw firm conclusions.
External check: FTMO Challenge passed
On 21 September 2026 our automated strategies passed the FTMO Challenge. This was a €80,000 evaluation account traded from 24 August to 21 September 2026 by Hash (Bitcoin) and our FX strategy, Nexus. The Challenge requires hitting a profit target while staying inside daily and overall loss limits.
The FTMO Challenge runs on a simulated account. It is an independent test of risk control, not live-capital performance.
Pipeline
Strategies not in the live table. Each one keeps generating and recording its daily signals so it can be judged on new data.
Midas
Gold perpetual future on OKX at 3×, trading small size since July 2026. A new short-side filter went live on 23 Sep 2026.
Nexus
FX: ranks 10 currency pairs and takes at most one trade a day. Traded on FTMO during the Challenge. Execution paused 23 Sep 2026; signals still recorded.
Horizon
S&P 500 ETF (SPY). Execution paused 20 Sep 2026 to fund the Maverick V2 test; signals still recorded.
Steadfast V1
The first single-stock model, now superseded by V2. Still trades a single share when balance allows, so we can track how its edge decays.
Copper
Copper CFD. Early live testing; one trade so far.
Research results (backtests, not live)
Only figures we could re-run from saved code and data on 27 September 2026. Each is labelled by how clean its test window is:
- Blind Neither the model nor any rule threshold saw this data.
- Selection The configuration was chosen after seeing results on this window, so the figure is optimistic.
- In-sample Rule thresholds were fit on this same window. It shows how the rule behaves; it is not evidence of an edge.
| Strategy | Window | Label | Sharpe | Return | Max drawdown |
|---|---|---|---|---|---|
| Hash | 23 Dec 2025 – 20 Jun 2026 | Blind | 1.32 | +24.1% | −16.9% |
| Hash + Robustness Rule | 1 Jan 2025 – 15 Sep 2026 | In-sample | 2.75 | +358% | −23.2% |
| Nexus | 180 days to 16 Jul 2026 | Blind | 2.64 | +7.7% | −1.8% |
| Steadfast V2 | 24 Mar – 8 Dec 2025 | Selection | 3.58 | +34.3% | −11.6% |
| Maverick V2 | 1 Jan – Sep 2026 | Selection | 2.32 | +51.7% | −13.2% |
Sharpe ratios are annualised as in each original study and are not directly comparable across rows.
How we work
- Test before money. We split each model's history into training, tuning and untouched windows, and a strategy goes live only after it holds up on the untouched one.
- Replace, don't defend. When a live strategy decays, we build its successor with the same process. Steadfast V2 replaced V1 in July 2026, and V1 still runs at one share so the decay stays measurable.
- ML and AI first, small by design. Models, not people, do the work that scales: research, signal generation, execution and monitoring are automated end to end, and AI tools extend what each person can build and oversee. We intend to grow capital and strategy count while keeping the team at three people or fewer.
Risks
- Short live history. Months, not years. Early results can reverse.
- Leverage. Hash trades at 3×. Single trades have ranged from −6.4% to +42.1% on committed capital.
- Small capital. Live positions are small. Execution at larger size is untested.
- Key person. One builder. The system is documented, automated, and keeps all state on disk and at the brokers.
- Backtests are not forecasts. Even blind windows are a few months long.