The thin orange line is the backyard weather station’s air temperature, for comparison. Click any legend entry (or tab to it and press Enter) to hide or show that line; hiding the water line takes its shaded daily band with it. The °F scale refits to what is left showing — the air swings tens of degrees in a day while the pool moves two or three — but never to a frame narrower than eight degrees.

Two more of the station’s series start switched off: its feels-like temperature, which sits exactly on the air line except when heat index or wind chill applies, and the outdoor humidity, drawn dashed against its own 0–100% axis on the right. That percent scale is pinned, so a humid week and a dry week look different, and humidity never counts toward the °F scale.

Inches per calendar month, July through June, one bar colour per season so the same month can be compared across seasons.

Each bar is one calendar month across every year on record: the bottom of the bar is the typical daily low, the top is the typical daily high. Tick “compare years” to lay each individual year over the top, so a warm summer or a cold spring stands out from the usual.

How much warmer the pool is than the outdoor air, month by month, one line per year — the same colours the “compare years” overlay above uses, so a year is the same colour on both charts. Positive means the water is warmer than the air, which it nearly always is, so the line rising means the gap widened. Click a year in the legend (or tab to it and press Enter) to hide it; the axis holds still either way, so the lines that remain never move.

Each point is built from the hours where BOTH sensors reported, differenced hour by hour — never one series’ average minus the other’s. Those differences are then grouped by hour of the day, averaged within each of the 24 hours, and the month is the average of those 24 figures. The gap has a daily rhythm of its own — widest before dawn, narrowest mid-afternoon — so this is what stops a month that lost more nights than afternoons from reading low: every hour of the day counts once, whether thirty days supplied it or one.

The tree overhanging the shallow end came down in mid-April 2026 (marked on the 2026 line). April itself is transitional — half of it shaded, half not — so it should not be read as either state. The clean comparison for the step is May and June 2026 against May and June of the years before it, and, if the step is real, those months again in 2027.

What this can and cannot settle: because a hot week lifts the air and the water together, weather largely cancels out of the difference, so a gap that has grown points at the pool or at the sensor rather than at the season — which is how a genuinely warm year is told apart from a thermometer drifting upward. It cannot say which of the two on its own, and it is a short record: five years, in which shading, cover use and rainfall all changed as well.

The pump sits at a different pressure at each speed it runs at, so this finds those levels in the data and tracks each one separately — the pale line is the day’s settled pressure at that speed, the bold one is the two-week trend. The levels are worked out afresh for each stretch of months the schedule stayed the same, and the bottom one is never assumed to be the pump off — a low speed can read the same as a stopped pump, and for years this one did. The filter verdict is read from the highest speed, where clogging shows most clearly: a slow climb means the filter is clogging up, a lasting drop means it was cleaned, and the pump merely slowing or stopping is neither.