Post-Crisis Response
Turning a -5.65σ Event Into Stronger Protection
In February 2026, Bitcoin suffered its worst drawdown since the FTX bankruptcy — a -5.65σ extreme event that has only occurred 4 times since 2010. Ultron drew down ~25%, with our team’s own capital bearing the same losses alongside subscribers. We spent extensive time studying past crash patterns and indicator desensitization, and developed new long-side protection layers. Backtests show these would have nearly halved the maximum drawdown.
--
ROI Improvement
--
MDD Improvement
Improvement from adding Long Filter (vs Short Filter only)
We studied past major BTC crashes in depth and discovered that before each event, the Binance BTCUSDT perpetual contract’s Long-Short Ratio (LSUR) and Open Interest consistently showed detectable anomalies. We had already developed the Short Filter in 2025 Q4. After this event, we further developed the Long Filter, completing the risk protection framework. Below is an introduction to the Filter mechanisms.
Short Filter
Monitors Cumberland OTC transfer activity on-chain. When large transfers exceed sell-side liquidity thresholds, short entries are blocked to avoid false signals during strong uptrends.
CoinKarma Indicator Review & Guide
Long Filter: Multi-Layer Protection
Layer 1 — Moving Average Filter
Gradually reduces long exposure as price falls below key moving averages, preventing catastrophic losses while retaining partial upside. With 3× leverage, a 50% coefficient caps exposure at 1.5×, and 40% caps it at 1.2×—effectively spot-level exposure—dramatically reducing liquidation risk.
Normal: 100% exposure
Below MA120: 50%
Below MA180 + Declining: 40%
Layer 2 — Crash Detection & Cooldown
Detects crash conditions when multiple signals fire simultaneously: 24h price drop ≥ 5%, combined with abnormally elevated LSUR z-score and Open Interest z-score. When triggered, all positions are immediately closed.
Normal Trading
Crash Detected
Mandatory Cooldown
72 hours
Expires
Monitoring Phase
≤ 120 hours
Conditions Met
Resume Trading
After the 72-hour mandatory cooldown, the system enters a monitoring phase. Trading resumes when any one of the following conditions is met:
- Volatility Normalization — Current 24h realized volatility drops below 70% of the volatility at the time of the crash trigger.
- Price Reversal Signal — Previous day's intraday range exceeds 10%, indicating a potential reversal.
- Time Expiry — 120-hour monitoring window expires without any condition being met, and the system resumes cautiously.
Results Comparison
BTC backtest from August 2022 to present, starting with $100,000 USD. 3x leverage, MLD (Multi-Layer Dynamic) rolling thresholds.
| Metric |
No Filters |
Short Filter Only |
Full Protection |
| Loading... |
Event Deep-Dive
How each configuration handled major market crashes. The table below compares the most impacted position during each event.
Always Evolving
Ultron is not a static strategy. Each market event teaches us something new, and we continuously refine our risk management framework.
Feb 2025
Market Pulse Index Integration
Migrated from Overall LIQ batch entry to Market Pulse Index (MPI) dynamic position sizing, enabling more precise market timing.
Sep 2025
Short Filter
Developed Cumberland OTC monitoring to avoid premature short entries during sustained uptrends.
Oct 2025
Dynamic Thresholds (MLD)
Introduced Multi-Layer Dynamic thresholds that adapt entry and exit levels based on current market conditions.
Feb 2026
Long Filter
Market stress analysis revealed additional protection opportunities. Developed the MA Filter and Crash Detection system with mandatory cooldown periods.
Q2 2026
Next: Macro Regime Detection
Researching cross-asset macro correlations and additional on-chain anomaly patterns to further expand the protection framework.
Frequently Asked Questions
Why does the backtest differ from live performance?
Ultron has been live-trading since May 2024, from day one of CoinKarma’s launch. The strategy has gone through multiple iterations since then. This page shows backtests using the latest MLD + Filter version, which incorporates all historical improvements. Since the strategy is continuously evolving, earlier live results naturally differ from the current backtest.
Why wasn’t the Long Filter in place before?
The Short Filter was already deployed and running in 2025 Q4. Developing the Long Filter required sufficient crash samples for modeling. The -5.65σ event in January 2026 provided the critical data patterns that enabled us to build and validate this protection layer. Additionally, after November 2025, to avoid the impact of fat-tail events in a bear market, the team proactively halved long exposure, which already reduced the original ~50% drawdown to ~25%. During the same period, BTC spot retraced over 50% from its highs, while Ultron with 3x leverage only drew down ~25% — already significantly outperforming spot holders — yet we continue to strengthen our protection mechanisms.
Is there a risk of overfitting in the newly added Long Filter?
The Long Filter has two layers. Layer 1 (MA Filter) is a trend-based position management mechanism: when price falls below key moving averages, the system automatically reduces long exposure rather than trying to predict specific crash events. This logic uses few parameters and doesn’t depend on specific patterns, making overfitting risk very low. In two of the five major backtest events, the Layer 2 crash stop-loss wasn’t even triggered — the MA Filter’s exposure reduction alone was sufficient to control losses. Layer 2’s crash detection does rely on a smaller sample set, but its role is supplementary insurance, not the primary line of defense.
Does the Filter reduce returns in bull markets?
Quite the opposite. Backtests show that adding Filters significantly increased ROI. The reason: avoiding deep drawdowns means the portfolio doesn’t have to climb out of a deep hole, and compounding effects accelerate equity growth.
Is the backtest period cherry-picked?
No. The backtest covers August 2022 to present, spanning a full bull-bear cycle and 5 major crisis events: FTX bankruptcy, BOJ rate hike, tariff escalation, November 2025 sell-off, and the January 2026 crash.
Does the team trade with their own capital?
Yes. Since CoinKarma launched in May 2024, the team’s own capital has been trading the exact same strategy alongside subscribers, sharing the same gains and losses.
Why does the backtest start from August 2022? Can’t it go further back?
The earliest available data from our data sources begins on August 1, 2022. This period already covers multiple distinct market phases — including bear markets, recoveries, bull runs, and several major crash events — providing a comprehensive validation of the strategy’s performance.