Trend Traders
Three live paper traders running the strategies that survived verification in LIVE_CANDIDATES.md -- re-measured from the archive rather than quoted from an earlier report. All three are long/flat only; no shorts (the short leg was negative in 6 of 6 up years across three variants, twice, in two separate projects). Ada and Halden decide once a day at the 00:00 UTC rollover, exactly the bar they were backtested on. Wren decides on 15-minute bars.
Read this before reading the equity numbers
Expect single digits, not the headline CAGR. These strategies returned
50%+ a year historically, but +4.6% in 2025 and +2.9% in 2026. That is
consistent with a decay measured directly elsewhere in this project: a
mean-reversion effect worth +16.00 bps per trade in 2017-2022 was worth
+2.41 bps in 2023-2026, an 85% fall. They still beat buy-and-hold in both
recent years -- but buy-and-hold was negative in both, and a benchmark
cratering is not the same thing as a strategy earning.
Expect long flat stretches. At 0.57x the EMA-50 spent 95% of its
history in drawdown, with a worst run of 784 days underwater.
Removing its best 20 days drops the CAGR from 30.5% to 11.5%. A month of
nothing happening is the normal case, not a fault.
Anything much better than the backtest is a bug, not a win. The live
code was checked against the measured backtest bar by bar -- the signal is
bit-identical across all 3,303 archive bars, and the equity curves diverge
0.16% to 0.21% over 1,500 days from one deliberate accounting difference
(live holds fixed units between rebalances; the backtest silently re-sized
to target every day for free).
What each one was measured at
| Trader | Strategy | Lev | Fee | CAGR | Max DD | Calmar | Sharpe | Win yrs | Trades/yr |
|---|---|---|---|---|---|---|---|---|---|
| Ada Fairweather | EMA ensemble 20/30/50/75 | 1.00x | 8 bps | +54.2% | -59.7% | 0.91 | 1.18 | 70% | 66 |
| as deployed | 0.57x | 8 bps | +31.3% | -38.6% | 0.81 | 1.18 | 80% | 66 | |
| Halden Rooke | Single EMA-50 | 1.00x | 8 bps | +52.4% | -61.6% | 0.85 | 1.14 | 80% | 27 |
| as deployed | 0.57x | 8 bps | +30.5% | -41.1% | 0.74 | 1.14 | 80% | 27 | |
| Wren Castellan | EMA-50 base + 8h momentum overlay | 1.00x | 8 bps | +66.8% | -66.9% | 1.00 | 1.16 | -- | 52 |
| matched-leverage control | 1.10x | 8 bps | +52.1% | -69.2% | 0.75 | 1.10 | -- | 0 | |
| -- | buy and hold BTC | 1.00x | 0 | +37.7% | -83.2% | 0.45 | 0.82 | -- | 0 |
Wren is the one that is not de-levered, and that is deliberate rather than an oversight. The 0.57x figure comes from de-levering a pure long/flat exposure series, which scales cleanly. Wren sizes its overlay off a risk-per-unit rule against equity and a shared 2x cap, so scaling its base does not scale the strategy -- it produces a different strategy nobody measured. It therefore runs at the leverage it was verified at, carrying the -66.9% drawdown that came with it. Wren is also the only one with a real fee ceiling: it breaks even against its own control somewhere between 12 and 15 bps per side, so it is a spot-exchange strategy and a CFD would kill it. Ada and Halden still clear a Calmar of 0.67-0.70 at 30 bps.