Why We Call Our Backtests “Illustrative,” Not “Validated”

The results, up front

Our Anchor Accumulate strategy buys more Bitcoin in weeks our risk model reads as low (accumulate) and less in weeks it reads as high, versus a plain weekly buy of the same base size. Across the four regime windows we test, it out-returns plain weekly DCA in every one of them— on a per-dollar (ROI) basis, including the sustained 2023–25 rally we originally expected to be its weak spot:

Figures as of July 2026, from the same live risk model shown on the Risk Model page. Anchor Accumulate out-returns plain weekly DCA in every window we test — that is not the reassurance it sounds like, and the next section is about why. We publish every window we tested on the performance page. Now the important part.

Why that's not proof

A backtest is a story told with the benefit of hindsight, and there are three specific reasons ours cannot be called validated:

1. The published weights were formalized after the fact. The strategy itself isn't new — its author has traded these same indicators live since 2021, normalizing them by hand on TradingView long before this site existed. That's real skin-in-the-game history, and it's why the model exists at all. But the codified version you see here, with its exact factor weights, was written down with full knowledge of how past cycles ended — and a backtest over those same cycles cannot distinguish genuine foresight from a well-fitted curve. Only forward results can, which is why we treat the backtest as illustration and let the frozen forward record carry the burden of proof.

2. Even four cycles is a small sample. Bitcoin has traded through roughly four boom-bust cycles since 2010 — enough to watch the same pattern repeat, and enough for the factors behind this model to prove useful more than once. But by the standards of statistical evidence, four is still a small number: over a handful of windows, a strategy can win through skill or through luck, and the data alone can't yet separate the two. We'd rather under-claim on a real edge than over-claim on a small sample.

3. Winning every window tested is a caution sign, not a victory lap. A strategy with a known weak spot is easier to trust than one that appears to have none. Risk-tiered buying is built to give something up in a market that never gives it fear to buy into — a smooth, uninterrupted rally with no meaningful pullbacks. In our current sample, it hasn't given anything up: Anchor Accumulate out-returns plain weekly DCA in every window we test, including 2023–25. We don't read that as proof the strategy has no weak spot — we read it as a sign we haven't yet seen the regime that would expose one. A rally with truly no dips to buy into has arguably never happened in Bitcoin's fifteen-year history, so the strategy's real adversarial case remains untested, not disproven.

What we can say concretely: even where it wins on return, it doesn't always win on risk. Across the full history and in the 2023–25 rally specifically, its own worst drawdown was deeper than plain weekly DCA's (−84.1% vs −79.8% since 2018; −47.6% vs −33.8% in 2023–25) — deploying more capital into dips means riding out sharper pullbacks along the way, not fewer. That's the real, present trade-off: more return, but not automatically less pain to get there.

What “illustrative” actually means for you

We use the word deliberately. Illustrative means the backtest shows you the shape of the idea — buy more when the market looks cheap on our factors, less when it looks stretched — and roughly how that behaved in the past. It is a way to understand the strategy, not a forecast of your returns.

Validated would mean the edge had been demonstrated out-of-sample, across independent data the model never saw, enough times to rule out luck and curve-fitting. We haven't earned that word, so we don't use it — and you should be wary of anyone in this space who does.

What would change our minds

There's exactly one thing that turns an illustrative backtest into evidence: a forward track record on data the model was fixed before seeing. So we've frozen the model's weights and started publishing the live signal, date-stamped, going forward. Over the next few cycles that record will either hold up out-of-sample or it won't, in public, either way.

That's the honest version of a track record: slow, unglamorous, and impossible to fake after the fact. Until it accumulates, the numbers on this page stay exactly what we called them — illustrative.