Weathering the Storm: How Trader-7 Survived 8 Regime Flips in 72 Hours
Weathering the Storm: How Trader-7 Survived 8 Regime Flips in 72 Hours
March 13, 2026 | Sprint 106+ Performance Review
Sometimes the best thing a trading system can do is nothing at all.
Over the past three days, Trader-7 faced its toughest test since going live: Bitcoin oscillated relentlessly around its 50-day moving average, triggering 8 regime flips in 48 hours. Every trend-following system's nightmare. The question wasn't whether we'd make money — it was whether we'd survive with our capital intact.
We did. Here's what happened, what we learned, and what's next.
The Market: A Masterclass in Chop
Between March 10-13, BTC played tug-of-war with its SMA50, swinging between $69,500 and $72,400. The price crossed the moving average eight times in two days, never straying more than 1% in either direction before reversing.
For context, our system uses BTC's position relative to SMA50 as a primary regime indicator: above = WEAK_BULL, below = WEAK_BEAR. Each flip triggers a 3-cycle cooldown where position limits drop to 1. When you're flipping every 2-5 hours, the math is punishing — the system spent roughly 50% of that 48-hour window in cooldown mode.
This is exactly the kind of market that liquidates aggressive trend followers.
Three Phases, One Story
Phase 1: Active Trading (first 8 hours) BTC started 3.5% above SMA50 — comfortable territory. The system opened 5 trades in 7 hours, all LONGs aligned with WEAK_BULL regime. Two winners (+$35.58 BTC trailing stop, +$12.55 BTC TP1), three losers (-$30.66 SOL stop, -$30.80 ETH stop, -$14.14 BTC reevaluator close). Net: -$27.47.
Not spectacular, but the losses were controlled. Every stop triggered at predetermined levels. The reevaluator saved us roughly $16 on trade 191 by closing at the regime flip instead of waiting for a full stop loss hit.
Phase 2: The Paralysis Zone (48 hours) This is where it got interesting. As BTC drifted toward SMA50, our boundary penalty system escalated from -0.2 points at 2% distance to -14.9 at 0.01%. Even a 95% confidence signal gets killed by a -14.9 penalty. The system generated proposals every cycle — it was thinking, analysing, reasoning — but the defensive gates blocked every single one.
Zero trades opened across 48 cycles. Eight regime flips came and went. The boundary penalty, regime cooldown, and direction-aware confidence thresholds worked in concert to keep us flat.
Phase 3: Cautious Recovery (final 12 hours) BTC finally pulled away from SMA50, reaching 2.75% distance. The system cautiously reopened: an ETH LONG at 0.72% distance (arguably early), a BTC LONG at 1.73%, and an XRP LONG at 2.64%. Three positions open at session end, all regime-aligned.
Sprint Validation: What's Proven
Since Sprint 106 (our performance baseline reset on March 7), we've deployed and validated six sprints. Here's the scorecard:
Sprint 115 (Regime Cooldown Fix) — VERIFIED The fix that moved cooldown decrement to unconditional execution worked flawlessly through all 8 regime flips. Previously, the counter got stuck at "2 cycles remaining" for 30+ hours. Now it decrements reliably: 3 → 2 → 1 → 0, reset at each new flip.
Sprint 107 (ETH/SOL Correlation Cap) — VERIFIED Over 20 blocks across the session. The system consistently prevented opening both ETH and SOL simultaneously, reducing correlated exposure. Simple rule, measurable impact.
Sprint 106 (Direction-Aware Thresholds) — VERIFIED In WEAK_BEAR phases, the confidence threshold correctly tightened to 75-82% (defensive stance). In WEAK_BULL, it relaxed to 72% (moderate). The system adapted at every single regime flip — no lag, no errors.
Sprint 111 (Confidence Variance Fix) — VERIFIED Signal confidence ranged 60-85% across the session. Before this sprint, signals clustered artificially at 82-85%, giving the validator nothing meaningful to work with. Now there's genuine variance that reflects actual conviction.
Sprint 110 (Strategist Context Expansion) — VERIFIED The strategist's 7-14 day rolling window produced excellent reasoning: correctly shifting from bullish to neutral to defensive as conditions deteriorated, citing specific patterns like "no meaningful trend" and "directionless market."
Boundary Penalty System — MVP OF THE SESSION Escalated precisely as designed. Blocked 48 consecutive cycles of entries during the chop zone. This single mechanism is arguably the most important capital preservation feature in the system.
What's Still to Be Proven
Sample size. We're at 8 closed trades since Sprint 106. Our monitoring target is 20-30 trades for statistical significance. With a 37.5% win rate on 8 trades, we can't draw meaningful conclusions about the direction-aware thresholds' impact on profitability.
Win rate in clear conditions. The first 8 hours showed a 40% win rate when BTC was well above SMA50 — promising, but still a tiny sample. We need to see whether the Sprint 106+ stack improves outcomes when the system is actually trading.
Boundary penalty calibration. The big question from this session: should the penalty curve be softer in the 1-2% range? During Phase 2, the system was completely paralysed for 48 hours. That was the right call in hindsight — all 8 regime flips proved the market was genuinely choppy. But there will be times when BTC approaches SMA50 and then trends through it. The current system would miss that move entirely.
Regime flip frequency as a signal. Eight flips in 48 hours is noise, not regime change. The system correctly handled each flip individually (cooldown, threshold adjustment), but it doesn't yet recognise that flip frequency itself is an indicator. A market that flips 8 times in 2 days isn't trending — it's ranging. That meta-signal could inform a different trading approach.
SHORT performance since Sprint 106. All 8 closed trades since baseline were LONGs. We haven't yet tested the defensive-stance SHORT signals in a sustained WEAK_BEAR regime. The validator correctly blocked marginal SHORT proposals during brief bear phases, but we need a real bear period to validate SHORT execution quality.
By the Numbers
| Metric | All-Time (97 trades) | Since Sprint 106 (8 trades) |
|---|---|---|
| Win Rate | 37.1% | 37.5% |
| PnL | -$72.12 (-2.4%) | -$69.16 (-2.3%) |
| Sharpe | -0.42 | -11.14 |
| Open Positions | 3 | — |
| Capital | $2,928 of $3,000 | — |
The all-time numbers tell a story of incremental improvement — we're down 2.4% after 97 trades across months of development and testing. The since-baseline Sharpe of -11.14 is ugly but statistically meaningless on 8 trades.
The Outlook
Cautiously optimistic.
The defensive systems work. That's no longer a hypothesis — it's proven across 72 hours of the worst possible market conditions. The system preserved capital when it should have, and it's starting to trade again now that conditions have improved.
The question is whether the system can be profitable in the windows where it IS trading. The first 8 hours showed promise: two clean winners from trailing stops and TP1 partials, losses contained to predetermined stops. If the Sprint 106+ stack can lift win rate from the historical 37% to even 42-45% during clear conditions, the math works.
Three positions are open as I write this. BTC is 2.64% above SMA50 — back in comfortable territory. The next 20 trades will tell us whether the recent sprint stack has genuinely improved signal quality, or whether we're still fighting the same win-rate battle with better defences.
Either way, the system just proved it can weather a storm. That matters more than any single trade.
Trader-7 is an LLM-powered automated paper trading system for crypto perpetual futures. This is a build-in-public project documenting the journey from concept to consistent profitability.
Previous updates: jamiewatters.work