I spent weeks rebuilding the Scout.
The original Scout was a Hash Rate Momentum bot. Over the full 6-year BTC history it returned +575% while HODL did +2,147%. It lost to just-holding by 1,572 percentage points. That's why it's on the post-mortems page as a BACKTEST-only warning example.
My assumption: the strategy concept wasn't broken, the execution was. So I built Scout v2 — a cleaner implementation around a different core signal. Donchian breakout on BTC daily, filtered by a 50-day SMA regime filter and a positive-funding requirement. TP at 15%, SL at 5%, 10-day time stop. Classic trend-following, well-documented edge-family, professional plumbing.
I ran the backtest. 2020-2026, 2,296 days, 71 trades. Bot return: +156%. HODL: +932%.
The bot is -776 percentage points behind HODL over the full period.
"Maybe it's a drawdown story?" I thought. The max drawdown WAS lower — -33.2% vs HODL's ~-77%. That could justify the return gap for a risk-averse investor.
Then I ran the walk-forward check. And Scout v2 died.
Here's the anatomy of another strategy that didn't make it, and what it taught me about the difference between "interesting research" and "live bot."
The Walk-Forward Kill
Three non-overlapping windows over 6 years:
| Window | Period | HODL | Scout v2 | Delta | Beats HODL? |
|---|---|---|---|---|---|
| W1 | 2020-01 → 2022-02 | +419% | +73% | -346 pp | ❌ No |
| W2 | 2022-02 → 2024-03 | +65% | +33% | -32 pp | ❌ No |
| W3 | 2024-03 → 2026-04 | +7% | +18% | +10 pp | ✅ Yes |
Scout v2 beats HODL in exactly one of three windows. The one where HODL was weakest (W3: +7% over 2 years — a quiet consolidation regime).
In W1 (the biggest BTC bull run in history), Scout v2 returned 73% while HODL delivered 419%. Scout v2 gave up 346 percentage points in exchange for a smaller drawdown. That's not risk-adjusted outperformance — that's dramatic underperformance.
In W2 (mid-cycle recovery), Scout v2 returned 33% while HODL did 65%. Bot trailed by 32 points.
One good window. Two bad windows. Net -776pp vs HODL.
This is the exact failure mode the 10 Biases series warns against. A strategy that looks okay on one regime and terrible in others, averaged into a "positive backtest" number that hides the truth. The walk-forward decomposition exposes the truth.
Why Donchian Breakout Fails on BTC Daily
The Donchian breakout strategy — buy when price closes above the N-day high, sell on time stop or TP/SL — has a well-documented historical edge. Richard Dennis's Turtles in the 1980s. Decades of academic literature on trend-following. It works on many asset classes across many decades.
It does not, as it turns out, work on BTC daily in the 2020-2026 period. Three reasons:
1. BTC's volatility is too high for fixed TP/SL
A 5% stop loss on BTC is routine noise. A 15% take profit is a single-day move in 2021. The fixed TP/SL structure lets the bot "work" on paper but in practice most trades hit the stop before the breakout has a chance to develop — or get taken out by routine intraday volatility.
Trend-following assets need stops scaled to realized volatility, not fixed percentages. Scout v2 doesn't have that.
2. BTC's trends are asymmetric
BTC goes up 30-80% per year on average for long stretches. Then it drops 70-80% in 6-12 months. This asymmetric return distribution means a breakout bot misses the biggest upward moves (entering late on confirmed breakouts, exiting early on TP) while still participating in the drawdowns.
Over W1 (2020-01 → 2022-02), BTC printed one of the biggest percentage moves in its history. HODL captured every dollar of it. Scout v2 captured about 17% of it. That 346pp gap is literally the cost of not participating fully in the biggest up-phase the asset has ever seen.
3. The positive-funding filter doesn't help when funding is noisy
The "require positive funding" gate was meant to filter out bear regimes. In practice, funding rates flip polarity constantly in BTC markets — especially during consolidation — so the filter adds whipsaw rather than regime clarity.
The idea was defensible. The execution didn't survive the data.
What This Says About the Original Scout
The original Scout post-mortem already documented that Hash Rate Momentum lost to HODL over the full period. Scout v2 was the supposed fix. Instead it demonstrated the same pattern, with a different underlying signal.
Two versions of Scout. Both underperform HODL over the honest benchmark. Both have one regime where they shine and two where they lose.
The unifying lesson: Scout-the-concept — a mid-frequency trend-follower on BTC daily — does not beat HODL after costs in the 2020-2026 sample. Not because the specific signal was wrong. Because the broader category (discretionary trend-following on a single asset with fixed TP/SL) is at a structural disadvantage vs HODL in a strongly-trending asymmetric market.
This is worth knowing. It means the entire family of similar strategies — Bollinger breakouts, channel breakouts, MA-crossover bots, fixed-stop trend-followers — probably also fails this benchmark on BTC daily. Save yourself the time: don't build them.
What Lives, What Dies
Scout v2 does not graduate to the /bots page. The local directory /110-BotLab/bots/scout-v2/ stays for research reference. The Supabase lab_bots table does not currently show Scout v2 (it was never promoted), so no downgrade is required there.
The original Scout stays on /bots with its BACKTEST tier badge — as an educational artifact showing a bot that loses to HODL but gets honest billing.
Scout v2 joins the post-mortems ledger. Fourth confirmed failure alongside:
- Watchdog+Tactician combos (walk-forward fails on both B and C variants)
- Alpha Hunter Light (retired for brand clarity)
- RSI bot (documented failure, archived)
The Bigger Lesson: When Walk-Forward Says No
This is the third time in a week that walk-forward has rejected a strategy that looked fine on continuous backtest:
- Watchdog+Tactician Option C: continuous +218%, walk-forward 1/3 windows. Rejected.
- Watchdog+Tactician Option B: continuous +170%, walk-forward 1/3 windows on both return AND drawdown. Rejected.
- Scout v2: continuous +156%, walk-forward 1/3 windows. Rejected.
Each one had a continuous-period result that would have been presentable on a chart. Each one failed the regime-robustness check.
This is not bad luck. This is the base rate. Most strategies that have a positive continuous backtest fail walk-forward. The few that pass are the ones worth deploying.
BearBullRadar's 5 live bots are what's left after running this filter on 200+ strategy variants. One real capital (Watchdog), three paper (Tactician, Alpha Hunter, Genius), one backtest-only (Scout v1) as a transparent warning example. That's the filter output. The rejected variants fill the post-mortems page.
If you ever see a strategy claim without walk-forward validation, it has 3 in 4 chance of being Scout v2. A pretty continuous backtest, a fatal hole the continuous view hides.
Run walk-forward or don't deploy.
Raw Numbers for the Skeptic
Full backtest config:
Donchian period: 20 days
SMA regime filter: 50 days
Take profit: +15%
Stop loss: -5%
Time stop: 10 days
Entry filter: positive funding rate required
Initial capital: $10,000
Fee: 0.1% per trade (Bybit spot taker)
Data: BTC/USD daily, 2020-01-01 to 2026-04-14
Results:
- Total trades: 71
- Win rate: 45.1%
- Total return: +156.06%
- HODL same period: +931.68%
- Max drawdown: -33.2%
- HODL max drawdown: ≈-77%
Source code and walk-forward JSON: /110-BotLab/bots/scout-v2/ in the BotLab repository.
Related reading:
- The original Scout post-mortem — Hash Rate Momentum, same family, same verdict
- Beat HODL or Don't Bother — the benchmark every strategy must clear
- Overfitting bias — how parameter-fitting produces Scout-v2-style failures
- Regime Bias — testing on one regime and calling it generalizable
- Post-mortems ledger — the full list of retired strategies
Not financial advice. This article documents a research failure for educational purposes. Past performance does not guarantee future results.
