The first panel on the Distributions tab is a rolling Sharpe, and on the capture I am working from it is an ugly picture. Twenty-eight plotted points running from 2026-04-11 to 2026-07-11, a vertical axis spanning -5.5782 to 0.6361, a line that spends nearly its whole length below zero, one spike up through zero in early May, and then a drop of more than three points followed by a slow grind back down toward the bottom of the axis.
Anyone looking at that chart concludes the same thing in about a second: the strategy fell apart. That may well be true. But you cannot read it off this chart, because a rolling Sharpe has two moving parts and the chart only shows you their quotient. Getting into the habit of asking which part moved is the single most useful thing you can do with this panel, and it takes about two minutes.
The two ways a Sharpe line moves
A Sharpe is roughly the average return over a window, minus a risk-free rate, divided by the standard deviation of those returns, then scaled up to an annual figure. Two inputs, and the second one lives in the denominator.
Take a book averaging -0.30% a day with a daily standard deviation of 2.0%. In periodic terms that ratio is -0.15, and annualising a daily crypto series by multiplying by the square root of 365, which is about 19.1, gives roughly -2.9.
Now change nothing about the returns and let the daily standard deviation fall to 1.2%. The mean is still -0.30% a day. The ratio is now -0.25, and annualised that is about -4.8.
The book performed identically in both cases. The Sharpe fell by nearly two points. That is the denominator effect, and for a losing book it runs in the direction nobody expects: getting calmer makes your Sharpe worse, and getting wilder makes it better. A negative numerator divided by a smaller number is a larger negative.
This is why the page stacks the two panels the way it does. The rolling volatility chart sits directly under the rolling Sharpe chart with its own 30D, 90D and 1Y toggle, and if you line the two windows up you can answer the question directly.
Reading the pair together
Four combinations, and each one implies a different next step.
Sharpe fell, volatility fell. The denominator did some or all of the work. Your returns may not have changed at all. The question that matters now is why volatility dropped, and the usual answers are that you cut position size, you stopped trading, or the book collapsed into a single quiet position. All three are worth knowing and none of them is "the strategy stopped working".
Sharpe fell, volatility rose. The numerator did the damage, and it did enough to overcome a denominator that was working in your favour. This is a genuine deterioration in returns and it deserves the reaction you were about to have.
Sharpe rose, volatility rose. Be careful here. A negative Sharpe improving while volatility expands can be pure arithmetic, with your losses unchanged and the divisor larger. Check the actual return over the window before you take any comfort from a line that went up.
Sharpe rose, volatility fell. The only one of the four that is unambiguously good news, and correspondingly the rarest.
Why the spike and the cliff may both be noise
There is a second thing to know about this chart before you act on it, and it is uncomfortable.
The sampling error of a Sharpe ratio estimated from n periods is roughly the square root of one plus half the squared Sharpe, all over n. On a 30-day window with a periodic ratio near zero that is about the square root of one over thirty, which is 0.18 in daily terms. Annualise it the same way you annualised the ratio itself, by multiplying by 19.1, and the standard error of the annualised 30-day Sharpe comes out around 3.5.
The entire visible range of this chart, from 0.6361 at the top to -5.5782 at the bottom, is a span of 6.2. That is under two standard errors of the estimator. The spike in early May and the drop that follows it are, on this arithmetic, well inside what an unchanged process produces by chance.
There is a related detail on the same chart. The description of the series says 28 data points across a window running from 11 April to 11 July, which is 91 days. That means the plotted points sit roughly three days apart rather than one, so what looks like a cliff between two adjacent dots is several days of moves compressed into one line segment. The chart is not lying to you, but it has less resolution than the eye assumes.
What to check instead this week
Three things, in order, and none of them takes long.
Start with the monthly returns heatmap at the bottom of this same tab. On the capture I am working from it reads April at -0.4%, May at -20.5%, June at -8.1% and July at -16.8%. Those are plain returns with no denominator in them at all, which makes them immune to the entire problem described above. If the months are red, the returns are genuinely bad, and no amount of care about the Sharpe changes that. If the months are flat while the Sharpe chart looks catastrophic, you have your answer and it was the denominator.
Second, go to the trade distribution panel on the same tab. It reads n equals 87 with a mean of -0.53 and a median of 0.00 on this capture. Eighty-seven observations is a far better sample than four monthly cells or twenty-eight rolling points, and it tells you about the shape of individual outcomes rather than about a ratio.
Third, if you are going to act, act on position sizing rather than on the strategy. A rolling Sharpe is too noisy at 30 days to justify switching a method off. It is not too noisy to justify halving size while you gather more data, and halving size is a reversible decision that costs you very little if the signal turns out to be noise.
Pin the source before you quote the number
One habit worth building whenever you take a figure off this page. The same statistic can appear on more than one Blockcircle surface carrying different values, because a metric computed over trade records and a metric computed over an equity curve answer different questions, and because panels can differ in their annualisation factor, their risk-free assumption and the window they cover. None of that means anything is broken. It means the name of a statistic is not its definition.
If you want to know which convention produced a given tile, there is a direct test. Use the export control in the page header with the account scope, the source filter and the date range all set explicitly, then rebuild the number yourself in a spreadsheet. Compute it once from the daily equity series and once from the individual trade returns, try both annualisation factors and both risk-free assumptions, and see which of the four candidates reproduces what is on screen. Whichever one matches tells you the convention that panel uses. Write it down next to the number, and from then on you are comparing like with like instead of guessing.