We Ran a CoinRoc Out-of-Sample Forward Simulation Through a Bear Market — Here’s What Happened to Drawdown

The research sitting at yodacom.com/research has a limitation every honest researcher has to own: it is historical. Backtest a strategy across two and a half years of data and the natural question comes back: "But does it actually work when you run it forward?"

That question is specifically about the RXI regime detection layer — the fuzzy inference engine that CoinRoc uses to decide which assets get a live grid, which assets get suspended when market conditions deteriorate, and when to resume. The prior study tested a fixed 2.5-year window. What we had not done was run that engine forward in real time, responding to new data as it arrived, with no knowledge of what came next.

So we built that test. The result is worth reporting — with the caveats that make it honest.

The Methodology

This is a simulation, not a live trading account. No real money moved. But the mechanics were designed to replicate live operation as faithfully as possible, specifically on the dimension that matters most for methodology credibility: look-ahead.

The simulation used only data available at each point in time; no future data informed any rule execution or regime decision.

In most backtests, the engine has silent access to future data — it “knows” how prices move before it places the trade. We went to unusual lengths to prevent that here.

At each 4-hour bar in the simulation, the RXI signal was computed using only the price data available up to and including that bar. Python array slicing enforced this at the code level, not by convention or assumption. The initial grid bounds were set from the 60 days of data before the out-of-sample period began. No future bar entered any calculation.

The cost model used OKX US Institutional VIP 2 fees: 0.06% maker, 0.08% taker, with a slippage floor of max(0.02%, 0.15 × ATR/price) per fill side. Grid resets — the friction cost when price exits the grid range and the grid must be rebuilt — were charged at 2× round-trip fee. Twelve resets occurred across ten symbols over the period. Suspended periods credited zero return; idle capital earned nothing. This is the conservative modeling choice.

The fee tier conditionality is also material: these results apply at OKX institutional fees only. At OKX consumer rates (0.08/0.10) results would be worse. At Binance.US retail (0.40/0.60), extrapolating from the fee differential, the sleeve’s net margin would be significantly reduced at Binance.US retail rates and may have produced a negative net result — though this scenario was not separately modeled. That context matters and is not buried in footnotes.

A 74-Day Bear Window

This is the most favorable regime window the research has examined to date; the broader 2.5-year study covers a more representative range of conditions, including windows where the sleeve did not produce favorable results relative to the benchmark.

This window was not chosen for favorable results. It is the period immediately following the prior study’s end date. It is simply what happened next.

Ten symbols ran in the grid sleeve, each independently rated by the RXI engine: ZEC, DASH, IOTA, XLM, SKY, TRAC, DOT, AAVE, ETH, BCH. The sleeve constituted 10% of the simulated portfolio (P3). The remaining 90% was allocated to a 60/40 SPY/AGG baseline (P2), the same structure tested in the prior study. This portfolio structure is a modeled construct for research comparison purposes and does not represent a recommended allocation.

The RXI Engine in a Bear Window

The visual above tells the mechanism story more clearly than numbers can. In the stress window beginning around May 22, 2026, BTC entered a sustained drawdown — eventually falling approximately 27% from its peak across the out-of-sample period. During this same window, the RXI-rated grid sleeve held roughly flat during this bear-market regime window.

That is the mechanism illustrated, not a return claim. The grid sleeve did not fall with BTC because it was not holding BTC directionally. It was running a rules-based oscillation strategy that the RXI engine actively managed — suspending exposure when conditions turned trending and hostile to grid mechanics, then resuming when conditions improved.

Over the 74-day period, the RXI engine fired 92 suspend events and 83 resume events across 10 symbols. One symbol, DASH, ended the period in a suspended state. The rules were firing continuously.

The full-period visual above shows all four tracked positions from April 14 through June 27. The separation between the grid sleeve and BTC buy-and-hold is the clearest illustration of what regime-aware grid mechanics look like in a bear window: the sleeve tracks a different mechanism entirely.

Three Things the Prior Study Could Not Isolate

Against the prior study (RES-REPOSITION-PORTFOLIO-01, October 2023 through April 2026), this 74-day window demonstrates three things that the prior study’s longer window could not isolate:

These are directionally meaningful. Our prior reconciliation analysis predicted exactly this pattern: a ranging or high-volatility-oscillation environment would narrow or close the performance gap between the rated grid sleeve and the buy-and-hold benchmarks. This window confirms that prediction.

The correct interpretation: the RXI regime detection performed as designed in the environment it was designed for.

The prior study’s 2.5-year window remains the more representative finding for long-run context. A 74-day bear window cannot override 2.5 years of evidence. This adds evidence; it does not rewrite the story.

What You Should Know Before Drawing Conclusions

We are stating these plainly because the honesty is the point.

The Mechanism in Plain Language

The RXI filter does something simple that is hard to do emotionally: it reduces exposure when conditions deteriorate, and restores it when conditions improve. Over this 74-day window, it held the portfolio roughly flat while BTC lost more than a quarter of its value from peak to trough — a result specific to this bear-market regime window, not a standing property of the mechanism in all conditions.

The mechanism is not prediction. The grid does not need to know where BTC goes. It needs the right conditions — oscillation within a range, not a trending directional move. The suspend and resume thresholds are fixed rules applied to incoming indicator values. No learning, no model updating, no prediction. When those conditions break, the RXI suspend logic steps back. When they return, it re-engages.

In a bear window like this one, drawdown protection is where the mechanism earns its keep. Whether that benefit persists across regimes is what the 2.5-year prior study answers — and the answer there is more nuanced.

The two and a half years of prior research context is still the more representative historical reference for understanding how the mechanism has behaved across a broader range of conditions. This 74-day result is evidence that the regime detection layer does what it says it does when the regime it is designed to protect against actually arrives.