✨ Too Much Information

It started with why is my garage so hot? and did not stay there. One house, one year, 8 sources, 14 sections, and a garage that is still hot.

Every number below is recomputed from the raw exports on each build; none of it is typed in by hand. The code that does that is at github.com/banasiak/tmi. How it got this far is a separate, much longer, story in the Cesspool of Knowledge.

Overview

Every data source this house produces, joined on local calendar date and priced with tariffs extracted from the utility bills themselves. Everything is joined over 7 Aug 2025 – 5 Aug 2026, the span the weather station covers; the bills, the weather proxy and the hourly water meter all reach further back, and four late arrivals exist to check what the first five had already produced.

$2,921

to run this house for a year

Electricity, gas, water, wastewater and refuse, priced with the tariffs transcribed from your bills — every rate below reproduces its source bill to within two cents. See Data & Provenance.

Electricity

$1,905

18,594 kWh over 359 metered days · 52 kWh/day · 65% of the total

Natural gas

$258

18,216 cf · $14.50/mo of that is the fixed access fee

Water

$240

52,444 gal · first 3,000 gal each month are free

Wastewater + refuse

$518

effectively fixed — neither responds to what you do

What this is built from

Granularity matters more than volume here: the same house looks like different things at 5 minutes, a day and a billing period, and several of these exist only to check another one.

Show all 8 sources, their granularity and what each settles

Source	Granularity	Span	What it settles
AmbientWeather WS-2000	5 minutes	364 days	44 channels — five sensed zones with temperature, humidity and dew point, plus sun, wind, rain, lightning and the pool probe
El Paso Electric, interval	15 minutes	364 days	Separates what never switches off from what runs to a timer and what answers the weather
El Paso Electric, billing	per bill	35 periods, Jul 2023–Jul 2026	The only source carrying cost, and the only one reaching further back than the year of station data that can be exported
City of Las Cruces UtilityHawk	daily, and hourly	366 days daily; 400 days hourly, continuous	Water and gas. The hourly series is what finds a leak — it puts a clock on irrigation, measures the pool refill directly, shows the float valve replacing evaporation overnight, and caught a split line the daily series structurally could not. The same meter at two granularities, so the two never corroborate each other — where they disagree, the finer one is simply closer
Utility bills, PDF	per bill	28	Every rate on this page, written into the tariff engine as a literal value rather than parsed at run time, and re-checked against its source bill on each build — 39 checks, all reproducing their source to within two cents
Appliance nameplates	per machine	7	Turns assumptions into readings: furnace efficiency, pool heater rating, filter flow ceiling, and what the air conditioner is allowed to draw
NOAA GHCN-Daily	daily	2 stations	Stands in for the weather the station's export cannot reach — it has been recording far longer than the rolling year that can be downloaded — calibrated against it where the two overlap
NREL PVWatts the only modeled input here, and the only one not measured at this house	hourly, typical year	4 roof planes	Every plane-of-array figure in the solar section. It is here because the station's own pyranometer sits below the roof ridge and loses its eastern sky — which a due-south plane would average out but an east-west roof cannot

Two of these arrived to check figures already derived — the hourly water and the nameplates — and both overturned what they were meant to confirm, which is why the numbers here are the corrected ones. One source is absent by choice: a charge log from the motorcycle's OpenEVSE, a rolling eight-week window that no amount of care turns into a year.

What things cost

Every figure here is recomputed from the transcribed tariffs on each build, and priced at the rate that actually applied on the day.

Where the money goes

Every line recomputed from the transcribed tariffs, priced at the rate that applied on the day. $2,921 in total.

ElectricityNatural gasWater, wastewater and refuse

Color is the meter, not the category. The top of this chart is two electric bars, and the third is the one stream that answers to nothing you do. The three Other bars are each meter's own remainder — what it measured that the rest of this page could not assign to a named load. They are split by meter rather than pooled, because a single bar spanning three of them would be the one thing this chart's colors cannot say.

Table view

What	Per year	Share	Detail
Always-on floor	$676	23.1%	0.76 kW that never switches off
Cooling	$532	18.2%	4,411 kWh — main HVAC plus about 15% from the patio mini-split
Wastewater + refuse	$518	17.7%	wastewater is tiered on an allowance re-set yearly; refuse is flat
Pool pump	$400	13.7%	3,998 kWh on a 6.75 h timer
Other electricity	$249	8.5%	every electric load too small or too irregular to separate from the meter — vehicle charging among them, plus the customer charge no load can shed
Gas access fee	$174	6.0%	$14.50/month before a single cubic foot
Other water	$169	5.8%	indoor use — showers, laundry, dishes, the tap. Metered, never separately, so it is a remainder rather than a measurement
Space heating, electric	$48	1.6%	581 kWh — 97% of it the mini-split, only 19 kWh the furnace blower
Irrigation	$28	0.9%	12,600 gal the controller meant to deliver
Space heating, gas	$27	0.9%	6,244 cf — the furnace alone
Water heating + cooking	$22	0.8%	6,845 cf standing baseline
Other gas	$20	0.7%	what the heating signature could not assign to space heating, the water heater or the pool
Pool evaporation	$19	0.7%	8,053 gal the float valve replaced, unasked
Pool refill (one-off)	$17	0.6%	5,228 gal over two days in March — not an annual cost
Pool + spa heating	$14	0.5%	4,704 cf on days with no heating demand
Zone 1 leak	$6	0.2%	1,938 gal since 2 June and still rising

The patio is a third conditioned zone, and it pays its own way

The house is not one thermal zone. The enclosed patio has its own mini-split — a 12,000 BTU/h Premium Levella heat pump, 1.73 kW at full draw — and neither of the two lines above is wholly the central system. Its share is recoverable without a submeter: the furnace's own gas consumption fixes how many hours it ran, which fixes its blower energy at about 19 kWh, so the remaining 562 kWh of the cold-weather line has to be the mini-split — and the heat-loss coefficient that implies prices its cooling season at 649 kWh, 15% of the cooling line.

Total: 1,211 kWh, roughly $132 a year. The heating line is almost entirely this unit; the cooling line is mostly the main system. How that zone actually behaves — its glazing, what the low-E coating costs and saves, and why a 12,000 BTU/h unit is five times the size it needs to be — is measured alongside the other three buildings in Four boxes in the same weather.

Where the electricity goes

Interval data separates three things a monthly bill blends together: what never turns off, what runs to a clock, and what responds to the weather.

Always-on floor

0.76kW

the quietest sustained draw on a median day — 6,548 kWh/yr, 35% of total

Floor, cost

$676/yr

at marginal rates — what the house costs before anyone does anything

Seasonal drift in floor

0.24kW

between the lowest and highest month — largely weather-independent

Weather-independent load

38kWh/day

regression intercept — 73% of consumption survives with zero degree-days

Every interval of the year

Each column is one day; each row a 15-minute slot from midnight to midnight. Power in kW.

The horizontal band across the afternoon and evening is the timer load — it holds the same clock position through every season, which is what distinguishes it from weather-driven use. The summer bulge above and below it is air conditioning. The pale strip across the small hours is the always-on floor.

Table view

Date	kWh	Peak kW	Floor kW
2025-08-07	79.8	7.28	0.92
2025-08-21	63.5	7.08	0.68
2025-09-04	54.9	9.4	0.68
2025-09-18	30.9	4.2	0.6
2025-10-02	50.4	6.6	0.6
2025-10-16	50.3	4.76	0.88
2025-10-30	41.3	4.52	0.72
2025-11-13	38.9	8.72	0.6
2025-11-27	35.0	3.24	0.76
2025-12-11	41.6	3.72	0.68
2025-12-25	43.0	5.08	0.64
2026-01-08	37.9	3.56	0.84
2026-01-22	36.7	3.68	0.88
2026-02-05	40.9	3.32	0.84
2026-02-19	38.8	3.64	0.64
2026-03-05	40.4	3.56	0.76
2026-03-19	41.8	5.68	0.72
2026-04-02	48.9	4.72	0.68
2026-04-17	41.6	6.24	0.68
2026-05-01	40.7	5.24	0.72
2026-05-15	63.8	7.04	0.64
2026-05-29	43.9	6.12	0.84
2026-06-12	59.3	7	0.72
2026-06-26	67.8	7.48	0.72
2026-07-10	83.1	9.8	0.8
2026-07-24	87.6	7.88	0.84

Something runs on a timer from 15:15 to 22:00

A block of 1.65 kW switches on at 15:15 and off at 22:00 — 6.75 hours, every day. That is 11.1 kWh/day, or 3,998 kWh across the 359 metered days — 22% of everything you use, about $400 a year.

It is equipment, not habit: mild-day consumption is 43.8 kWh on weekdays and 44.1 kWh at weekends — a difference of 0.3 kWh. Nothing people do is that consistent. It is the pool pump — confirmed below by a second sensor that has nothing to do with the electricity meter.

The motor is an A.O. Smith SQ1102 — 1 HP nameplate, service factor 1.65, 3450 RPM, capacitor start. That rating predicts 1,641 W at full load; the meter measured 1,650 W. Three independent confirmations now agree — the electrical step, the water temperature jumping at the same quarter-hour, and the nameplate.

It is single-speed, so runtime is the only variable without new hardware. And there is a great deal of runtime to give back. At around 60 GPM this pump turns the 5,000 gallon system over once every 1.4 hours, so 6.75 hours a day is about 4.9 turnovers. The usual guidance is one, two if you want margin.

Daily runtime	Turnovers	kWh/yr	Cost/yr	Saved
6.75 h (today)	4.9	3,998	$400	—
5.00 h	3.6	2,962	$297	$104
4.00 h	2.9	2,369	$237	$163
3.00 h	2.2	1,777	$178	$222

Cutting to three hours still gives 2.2 turnovers and saves $222 a year, at no cost. Before acting on that, check what else is tied to pump runtime: a salt chlorine generator only makes chlorine while the pump runs, the heater needs flow to fire, and surface skimming needs some hours. Any of those can set a floor that filtration alone does not.

Replacing the motor with a variable-speed unit changes the arithmetic entirely, because power follows the cube of speed while flow follows the first power — half speed moves half the water for an eighth of the power:

Speed	Draw	Hours for 2 turnovers	kWh/yr	Cost/yr	Saved
75% speed	0.70 kW	3.7 h	926	$93	$308
50% speed	0.21 kW	5.6 h	411	$41	$359
40% speed	0.11 kW	6.9 h	263	$26	$374

At half speed the same filtration costs $41 instead of $400 — a saving of $359 a year. Worth noting that US efficiency rules have barred new single-speed pumps above 0.711 total HP since July 2021, so a replacement would be variable-speed regardless, and utility rebates for the swap are common.

Timing still matters as exposure rather than cost. The block runs straight through the late-afternoon system peak, and four months of it fall in the summer season where the marginal rate is $0.1304/kWh against $0.0841 in winter — 1.55× more. If EPE introduces time-of-use pricing, this schedule is your largest exposure to it.

The shape of a day, by how hot it was

Median power at each 15-minute slot, grouped by the day's mean outdoor temperature.

Cold day (57d)Mild day (98d)Hot day (42d)

The timer block sits at the same clock position in all three curves, cold days included — it does not care about the weather. Differencing the hot and mild shapes isolates the cooling load without a submeter: about 35 kWh/day, peaking at 3.16 kW around 14:00.

Table view

Time	Cold day (57d)	Mild day (98d)	Hot day (42d)
00:00	1.3	0.9	1.32
01:00	1.32	0.82	1
02:00	1.62	1.07	0.99
03:00	1.63	0.77	0.95
04:00	1.66	0.78	0.98
05:00	1.78	0.79	0.96
06:00	2.09	0.84	0.98
07:00	2.12	0.99	3.62
08:00	1.94	1.08	2.42
09:00	1.51	1.15	2.38
10:00	1.22	1.35	3
11:00	1.3	1.54	3.15
12:00	1.31	1.87	3.93
13:00	1.23	1.99	4.31
14:00	1.29	2.07	4.92
15:00	1.28	2.21	4.81
16:00	3.08	2.69	4.37
17:00	3.24	3.74	5.91
18:00	3.25	3.87	6.09
19:00	3.1	3.56	5.97
20:00	3.21	3.55	5.91
21:00	2.49	3.04	5.06
22:00	1.31	1.88	2.9
23:00	1.4	1.15	2.16

Every interval of the year, sorted by size

All 34,460 fifteen-minute readings ranked from largest to smallest. The x-axis is rank, not time.

The same three loads the curves above separate, seen here as one shape. The cliff on the left is air conditioning — 2,201 hours a year above 3 kW, 25% of the time. The long flat tail on the right is the floor, and it never reaches zero: the quietest quarter-hour of the entire year still drew 0.48 kW.

Table view

Share of the year	Days	At or above (kW)
Peak	—	13.8
0.1%	0.4	10.8
0.25%	0.9	9.56
0.5%	1.8	8.32
1%	3.6	7.36
5%	18	5.52
10%	36	4.4
25%	90	3.04
50%	179	1.36
75%	269	0.96
90%	323	0.76
99%	355	0.64
100% (floor)	359	0.48

Half the year is spent under 1.36 kW, and the peak is 10× that

Ranking the readings instead of plotting them against the clock answers a different question: not when the house draws power, but how much of the year it spends at each level. The median quarter-hour is 1.36 kW — barely more than the floor — while the peak reached 13.8 kW.

That gap is why the annual bill is so much flatter than the peaks suggest. The top 1% of intervals run above 7.36 kW but there are only 86 hours of them; the bottom 10% sit below 0.76 kW for 862. The area beneath the 0.76 kW line — that draw sustained through every hour of the year, not just the hours that sit on it — is 35% of the year's energy: the house's own idling, paid for in full whatever the weather does.

It also sizes any battery conversation before it starts. Carrying the house through the top 1% of intervals is a demand problem; carrying it through the flat tail is an energy problem, and they need very different hardware.

What solar would do here

A grid-tied, net-metered array split across the east and west slopes, priced by rebuilding the actual bill month by month rather than multiplying production by an average rate.

First: the station's solar sensor reads 26% low, and only 20% of that is the sensor

This section does not use the weather station for how much sun falls on the roof. It is worth explaining why, because the reason is more interesting than the correction.

The sensor reports 1,557 kWh/m² a year, peaking at 995 W/m². NSRDB puts Las Cruces between 2,050 and 2,150 — a shortfall of about 26%. It can be caught out without leaving the dataset: the clearness index should approach 0.78 on a clear desert day, and restricted to 4,124 moments with the sun above 50°, where geometry cannot be blamed, the median is 0.63 with a 99th percentile of 0.72.

But that shortfall is two faults, not one. An instrument that under-reads and an instrument with something in front of it are indistinguishable in an annual total, and separate by sun angle: a hardware fault is there at every altitude, an obstruction runs out once the sun climbs above it. Above 50° elevation, where nothing blocks either instrument, the station reads a steady 80% of PVWatts. That 20% is the silicon photodiode a consumer station carries in place of a thermopile: it answers to a narrow slice of the spectrum where a thermopile absorbs the whole of it. That the gap holds flat as the sun climbs is what identifies it — a spectral shortfall is the same at every angle, where a cosine error would widen as the sun fell. The remaining 6% does exactly that, and it is the subject of the next panel.

Where the station is still used — the timing chart below, and every non-solar figure on this page — it is scaled by ×1.35, applied to the horizontal reading before any tilt geometry: the beam-and-diffuse split depends on clearness, so correcting afterwards would keep an overstated diffuse fraction. That scale is now confirmed from outside. A flat PVWatts plane is just global horizontal, and it returns 2,145 kWh/m² against the corrected sensor's 2,097 — agreement to about 2%, from a source that shares none of its hardware.

The other 6%: something stands in the sensor's eastern sky

The pyranometer sits on an old satellite mount 31 inches above the eave, on the west slope below the ridge — which makes the ridge the obvious suspect, and the arithmetic rules it out. Cutting the sky at 21° from a 31-inch pole needs 85 ft of roof between sensor and ridgeline; this house is about half that across, and the conclusion holds even if that estimate is three times off. The roof accounts for 13–18° of the 21°, depending which ridgeline the sightline clears, and cannot be the whole of it.

An aerial photograph of the property supplies the likelier candidate: mature trees stand due east of the sensor at the same latitude, tall enough to throw shadows several times the length of the shrubs beside them. A canopy is also the better fit for the shape measured — an obstruction that ramps rather than cuts. Their height has not been measured, so how the remaining degrees divide between roof and trees is open.

What is in the way is unsettled; that something is in the way is not. The evidence is in the sky, not the specification. Binning every clear-day reading by where the sun actually was, the eastern sky is blocked to roughly 21° of elevation while the western sky is clear to the horizon. At matched altitude and therefore matched airmass, morning clearness runs 0.11 against an afternoon 0.34 — a three-fold gap that no atmosphere produces. It is not the instrument either: above 65° the two halves of the day agree to 0.003.

Three explanations were tested and failed. Dew predicts the deficit growing on humid nights; it is largest on the driest. A nocturnal inversion predicts a winter maximum; the deficit is twice as large in summer. The Organ Mountains to the east subtend only a few degrees from this side of the valley and would cut a sharp edge — the eastern sky instead fades in gradually over an 18°-wide span of elevation, and the block gets worse toward the north, where the range is further away and lower.

The same test convicts the roof. A ridgeline a dozen feet away is the sharpest edge in this problem and should cut rather than fade — which is the second reason to look past it.

Why it matters here and nowhere else. On the due-south roof the error would cancel exactly — a south plane weighs morning and afternoon symmetrically about noon. On an east-west roof it does not cancel; it is the entire measurement. The station puts the west slope 7.4% ahead of the east. PVWatts, from a clear horizon, puts them within 1.7%. That is why production below comes from a model rather than from the backyard.

Worth pricing before anyone climbs up. Moving the pole to the nearest ridgeline gains about three feet and buys perhaps three degrees — 21° down to 18. The main ridge would gain nine and help more. Neither clears a tree, which is why the canopy wants measuring before the obvious fix is worth doing.

The test that catches the sensor, using nothing outside the data

k_t=GHIG_sc · E₀ · sin α

k_tclearness index — the share of the extraterrestrial beam that reaches the groundGHIwhat the station reports, W/m²G_scsolar constant, 1,367 W/m²E₀eccentricity correction for the day of yearαsolar altitude — restricted here to above 50°, so a low sun cannot be blamed for a low reading

A clear desert moment should reach 0.78. The median is 0.63, which is why everything below is scaled by ×1.35.

From plane-of-array insolation to kilowatt-hours

E=H_poa · P_dc · PR

Eannual generation, kWh. The tables below land about 1.6% under what this line gives, because net metering settles per billing month: they price the 11 complete months and scale to twelve, which leaves out a high-sun AugustH_poainsolation on the roof planes, 2,065 kWh/m² — the mean of 2,082 east and 2,047 west, across the 364 days coveredP_dcarray size, kW — rated at 1,000 W/m², which is why the areas cancel and no panel efficiency appearsPRperformance ratio, 0.78 — inverter, wiring, soiling, and the temperature derate that matters at 110°F

Your pitch costs you, and there is nothing worth doing about it

On a south roof, any common pitch is near enough the right one — the whole 14° to 32° span sits within a couple of percent of optimum. That is not true here. On an east-west roof, flat is optimal and every degree of pitch is a loss.

A horizontal plane here collects 2,145 kWh/m². The same panels laid on your two slopes collect 2,072 — 3.4% less. A south roof at the same pitch would have collected 2,436, an 18% premium that this roof simply does not have available to it.

The reason the conclusion survives is that the alternative is worse. Tilt-up racking on an east or west slope has to fight the roof rather than follow it: to reach a south-facing tilt it needs ballast, wind loading and row spacing that a flush mount never pays for, on a roof whose planes are only 3.4% off horizontal to begin with. Flush-mount both slopes and take the 3.4%.

The pitch is measured, not assumed — 22.5° off the roof by phone inclinometer, an instrument good to about ±1°.

The figures here come from PVWatts runs at 22.6°, which sits inside that band, and re-running them at the reading itself would be theatre. Plane-of-array moves 2.4 kWh/m² per degree on the east slope and 4.0 on the west, so the full width of the measurement's uncertainty is worth about 0.2% — well under the uncertainty in the performance ratio these numbers are multiplied by. The measurement's job was never to refine the model; it was to confirm the roof is not the 25–30° the sensor's blocked horizon once suggested.

An 8 kW array against the months it has to cover

Generation split evenly across the east and west slopes, against metered consumption. Complete months only.

ConsumptionGeneration

Consumption is the context and generation the proposal, so only one of them is drawn as a series. Note how much flatter generation is than consumption: the array varies about 2.4× across the year against the house's 2.1×, because summer heat derates panels and a tilted plane loses to a high sun. The binding month is spring, not summer — that is where generation first catches consumption and surplus starts.

Table view

Month	Consumption	Generation
09/25	1,608	1,079
10/25	1,422	960
11/25	1,179	719
12/25	1,351	626
01/26	1,490	684
02/26	1,142	807
03/26	1,344	1,133
04/26	1,324	1,325
05/26	1,537	1,503
06/26	2,132	1,476
07/26	2,404	1,317

What each array size returns

Priced by rebuilding both bills from the real tariff, month by month.

Annual bill saving

Payback improves with size — 13.9 years at 4 kW against 11.0 at 10 kW — because installed cost per watt falls while every kWh keeps the same value. Cost, panel count, roof area and spill are in the table below.

Table view

Array size	Annual bill saving
4 kW	623
6 kW	934
7 kW	1,090
8 kW	1,243
10 kW	1,495

Every size, in full

Net cost is after the 30% federal credit and New Mexico's 10%. Spill is production that overran the month it was made in.

Array	Panels	Roof ft²	kWh/yr	% of use	Gross	Net	Saved/yr	Spill	Payback	25-yr net
4 kW	10	215	6,342	34%	$14,400	$8,640	$623	0	13.9 yr	$6,034
6 kW	15	322	9,513	52%	$18,900	$11,340	$934	0	12.1 yr	$10,671
7 kW	17	366	11,099	60%	$21,000	$12,600	$1,090	0	11.6 yr	$13,080
8 kW	20	430	12,685	69%	$23,200	$13,920	$1,243	1	11.2 yr	$15,351
10 kW	24	516	15,856	86%	$27,500	$16,500	$1,495	813	11.0 yr	$18,719

8 kW — 20 panels, $1,243 a year, 11.2 year payback

At $2.90/W installed that is $23,200 gross, or $13,920 after the 30% federal credit and New Mexico's 10%. It produces 12,685 kWh — 69% of the 18,472 kWh this house uses — and needs about 430 ft² of roof, which an even split puts at 215 ft² on each slope. That consumption figure is the complete billing months scaled to twelve, so it sits a little under the 18,594 kWh totalled at the top of this page from every metered day.

Every kWh is worth $0.098, and that figure holds flat across every size below 10 kW. That is the whole difference net metering makes: without it an array this size would be discarding a fifth to a third of its output, and a battery would be an expensive way to buy some of it back.

Held at today's rates and degrading the panels 0.5% a year, 25 years returns $15,351 net of the install. Rates rising is the upside case — roughly a year off the payback per 1%/yr of increase.

What no array can touch, and what could still go wrong

$137 a year survives a zero-kWh month. Customer charge, franchise fee and gross receipts tax are not per-kWh, so they are outside solar's reach entirely. Nobody's quote will mention it.

Above about 10 kW it turns over. At that size 813 kWh a year overruns the month it was generated in, and surplus is modeled at $0.035/kWh rather than retail. New Mexico utilities also generally cap system size near historical use, so this is a regulatory ceiling as well as an economic one.

The tariff is the real risk, not the hardware. These paybacks assume the rate schedule transcribed elsewhere on this page still applies in fifteen years. EPE has pursued separate rate classes for solar customers before; ask directly which tariff you would be moved onto, because that question is worth more than a percentage point of panel efficiency.

One thing that stops mattering: under net metering it no longer matters when the pool pump runs. An exported kWh and a self-consumed one are worth the same, so the case for re-timing it disappears. Cutting its runtime still saves what it always did, and now that saving is clean — it also shrinks the array you need.

The sprinkler saga

Water is the one stream with no weather signature worth fitting: it answers to a clock rather than a thermometer. The meter recovers that clock on its own — the schedule it keeps, the two seasonal settings it steps between, and the volume each cycle delivers.

The drip controller is legible from daily meter reads alone

Three weekdays sit far above the other four, every week, for a year: Tue/Thu/Sat. Daily resolution cannot see a valve open, but a controller running to a weekly schedule cannot hide from it either — the gap between the two groups is the water the controller delivered.

In the warmest months it puts out about 384 gallons a week; in the coldest, 168 — 44% of the summer rate, on the same three days. Across the year irrigation accounts for roughly 12,600 gallons, 24% of everything metered.

The watering days are not quite fixed, and the reason is not horticultural. The daily meter reports them changing one time across the year. It is right that they moved and wrong about why — read at the hour each cycle actually ran, the start time moved with them, which no reprogramming would do. That is a clock, not a gardener, and it has its own note in the hourly section.

The controller delivers 47 gallons a cycle in winter and 107 in summer — the runtime, and nothing else

Each cycle can be read straight off the meter, because the controller owns the 20:00 hour outright — 150 of them across 14 months. No baseline to subtract, no household noise to argue with.

The seasonal step is the runtime. The X-Core holds two settings rather than a seasonal curve — 14 minutes of valve-open time in winter against 30 in summer — and the water follows: a clean winter cycle is 44% of a summer one, against 47% of the runtime. The small remainder is a slightly slower winter flow.

Neither figure is one number all season. Winter cycles run 41–48 gallons and clean summer ones 92–112, drifting down through the autumn as the ground cools and the line with it. That drift is why the fault below takes its reference from the weeks immediately before the break rather than from a season-wide average.

Steady enough to catch a fault. Across the clean months a cycle holds its level to within a few gallons. That is what made a 92-gallon rise unmistakable, and why this is the only channel on the page that could have caught it.

The same events estimated from daily totals scatter about 47 gallons, so only a sustained step of roughly 95 gallons would show — obvious in summer, but in winter a fault would have to nearly double the cycle first. Subtracting a household baseline also charges anything landing on a watering night to the controller, so smooth seasonal drift in that estimate is the pool and the hose, never a controller that ramps.

Irrigation versus everything else

Complete months only, and the two refill days of 29–30 March held out. Each month's metered water, split into the controller's own events and everything left over.

IrrigationLeakEverything else

Meter readings, not a model: the three series sum to the 42,829 gallons actually billed across these 11 months.

The irrigation series steps and the gray one drifts, and those are different things. A cycle has two sizes and changes between them on the day somebody moves the dial — the winter block from November to February is unmistakable. The monthly totals are less tidy than the cycles because they also carry how many watering nights fell in the month, and because a summer cycle tapers through the autumn as the ground cools. The gray series has no settings at all: it runs 108 gal/day across the summer months against 66 in the depths of winter, and it slides between the two because that is what evaporation does. Most of that 41 gal/day gap is the pool topping itself up through the float valve — the pool section predicts it from the weather and gets the seasonal shape right to an R² of 0.86. The rest is hose work and washing vehicles.

The third series is the split line in Zone 1, which is what a controller that cannot ramp looks like when something downstream of it can. It appears in June, it grows, and it is the subject of its own section. Folded into the irrigation series — where a daily meter has no way to separate it — it reads as an irrigation program doing something no irrigation program does.

March is short two days here: 29–30 March put 5,229 gallons through the meter refilling the pool, which would have dwarfed every other bar on the chart.

Table view

Month	Irrigation	Leak	Everything else
09/25	1,034	0	2,769
10/25	1,118	0	2,458
11/25	414	0	2,465
12/25	612	0	2,029
01/26	705	0	2,083
02/26	573	0	1,862
03/26	902	0	2,268
04/26	1,505	0	3,427
05/26	1,224	0	3,746
06/26	1,476	684	3,380
07/26	1,362	1,066	3,667

Water, by the hour — and the one thing that is broken

The daily meter could say which days water moved. 400 days at hourly resolution say at what hour, how fast, and — on the night one cycle stopped matching every cycle before it — exactly when something broke. That fault closes the section: it is the only live problem on this page, and the only one the daily meter is structurally blind to.

The shape of a day, season by season

Mean gallons in each hour of the day, across every complete day of the year bar the two the pool refill occupies. Every panel on the same scale, each against the annual mean for reference.

SeasonWhole year

One spike, in every season, at 20:00 — the controller, which never moves. What changes is its height. Each line averages every day of the season including the four a week the controller sleeps, so the peak here is roughly three-sevenths of a cycle. The rest of the day is flat and low: this household's water is a schedule with a little noise on top, which is why a change in that one hour is visible at all.

Table view

	Whole year	Winter — 90 days	Spring — 89 days	Summer — 128 days	Autumn — 91 days
00:00	1.97	1.41	2.12	3.01	1.33
01:00	2.6	3.88	1.84	2.63	2.05
02:00	4.32	4.03	4.04	5.52	3.66
03:00	3.36	1.19	4.01	5.43	2.8
04:00	1.76	1.21	1.79	2.84	1.22
05:00	1.59	1.1	1.57	2.46	1.23
06:00	2.82	1.82	2.82	3.21	3.43
07:00	3.65	2.48	2.54	3.59	6
08:00	3.43	3.03	3.13	4.06	3.49
09:00	3.77	3.32	4.02	4.73	3.01
10:00	4.4	3.59	4.13	6.7	3.18
11:00	5.06	2.92	6.03	7.31	3.96
12:00	6.63	3.51	6.11	13.3	3.59
13:00	4.73	2.58	4.75	8.73	2.88
14:00	4.63	3.32	4.87	6.9	3.44
15:00	5.58	3.86	6.67	6.47	5.33
16:00	8.03	3.17	9.64	10.4	8.92
17:00	5.82	3.02	6.01	9.17	5.09
18:00	5.95	3.74	5.25	10.1	4.66
19:00	7.63	5.4	8.75	10.7	5.63
20:00	38.8	22.5	41.9	60.6	30.0
21:00	5.1	2.27	8.08	6.38	3.67
22:00	2.93	2	3.45	3.88	2.4
23:00	2.34	1.98	2.38	3.3	1.71

Hour by hour, the refill comes to 5,002 gallons — the daily meter said 5,000

Pool volume is the keystone here: the per-degree costs, the spa volume and every turnover count derive from it. It comes from the March drain and refill, by subtracting an estimated daily baseline from two days of billing — 5,000 gallons.

The hourly register measures it directly. The fill ran 28 hours from 14:00 on Sunday, holding a steady 274 gallons an hour (4.6 a minute) for fourteen hours straight — a hose, left on overnight. Net of baseline that is 5,002 gallons.

The two land 2 gallons apart on 5,000 — closer than either method deserves, and closer than the drain-and-refill assumption underneath both of them. The check is worth having because it could have gone the other way: the daily figure rests on an estimated baseline, and an hour-by-hour read owes it nothing.

The weather over the pool predicts the water going into it

The strongest check does not measure volume at all — it measures the surface, and it comes from instruments that have nothing to do with the water meter. The pool has a float valve, so evaporation is replaced automatically and shows up as flow in the small hours. Carrier's evaporation relation predicts that flow from three measured quantities: the pool probe's water temperature, the outdoor dew point, and the anemometer.

Evaporation from an open water surface

w=(95 + 0.425·V) · (P_w − P_a)Y

wwater lost, lb per hour per ft² of surfaceVair speed over the water, mph — measuredP_wsaturation vapour pressure at the water temperature, inHg — from the pool probeP_avapour pressure of the air, inHg — the saturation pressure at the outdoor dew pointYlatent heat of vaporisation, 1,050 BTU/lb

Run across the year at the measured 181 ft² of surface, and set beside what the meter actually passed in the small hours:

What the weather asks for, and what the meter delivered

Monthly means. The prediction uses no water data; the measurement uses no weather data.

PredictedMetered

Two instrument chains with nothing in common — a pool thermometer, a dew point and an anemometer on one side, a water meter on the other — tracing the same four-fold seasonal swing. The gap between them is close to constant, which is the giveaway: it is the household's own overnight draw, 14 gal/day, sitting under the evaporation the whole year.

Table view

Period	Predicted	Metered
Aug	34.1	51.0
Sep	20.8	30.0
Oct	24.6	36.0
Nov	17.2	24.0
Dec	12.7	24.0
Jan	12.3	30.0
Feb	15.9	24.0
Mar	27.2	36.0
Apr	31.9	42.0
May	37.1	42.0
Jun	43.5	48.0
Jul	42.5	48.0

What that agreement is worth

The two track each other across a four-fold seasonal swing — slope 0.82, R² 0.86 over 12 months, with nothing fitted to make it happen: every term is a sensor reading and the only free choice was the surface area.

The slope is 0.82 rather than 1, and this fit cannot say why. The pool's surface area might be smaller than the 181 ft² measured; Carrier's relation is documented to run high on still water with nobody in it; or the anemometer on its mast reads a breeze the sheltered pool never feels. Only the first is about the pool.

Widened from the pool to all the water this house cannot account for, the same shape holds at a steeper slope — 1.65, R² 0.77. Which puts evaporation at about 61% of the seasonal swing rather than all of it; the rest is what Carrier's relation excludes — swimmers, hoses and longer showers, all leaning the same way the weather does. That fit's own intercept, 46 gal/day, is the indoor water underneath both.

None of this measures the pool. Inverting the fit implies 149 ft² against the 181 ft² measured, which would assume the model exactly right in order to prove the tape wrong. What it establishes is that two instrument chains sharing no hardware reproduce one seasonal shape to R² 0.86 — and that either was free to fail.

Two programs, one flow rate — and the pipe is the honest witness

Every cycle fires on the same schedule — 20:00, Tuesday, Thursday and Saturday — inside a single hour, in all 14 months of the record. Only the volume moves: 47 gallons a cycle on the winter program against 107 on the summer.

Dividing each by the runtime it was set to gives the one thing the controller does not control — how fast the pipe delivers.

Month	Program	Gallons a cycle	Implied flow
Jul 2025	30 min	112	3.73 GPM
Aug 2025	30 min	109	3.62 GPM
Sep 2025	30 min	94	3.14 GPM
Oct 2025	30 min	92	3.07 GPM
Nov 2025	14 min	42	2.97 GPM
Dec 2025	14 min	47	3.36 GPM
Jan 2026	14 min	50	3.60 GPM
Feb 2026	14 min	48	3.41 GPM
Apr 2026	30 min	109	3.64 GPM
May 2026	30 min	108	3.59 GPM

Both programs land between 2.97 and 3.73 GPM — the same plumbing, moving water at the same rate, whichever runtime it was given. That is what makes the runtimes credible without anything here being able to measure them: an hourly bucket records volume and never duration, so the two figures are the owner's account, and their only corroboration is that dividing by them produces one flow rate instead of two.

These months are the clean ones. From June the same arithmetic returns 6.87 GPM, which no pipe does on its own — that is the leak, and it has its own section above.

When the controller actually fired

One dot per cycle, placed at the hour the meter recorded it — against the 20:00 it was programmed to use.

The record holds no cycle at a wrong hour that is also a wrong volume, which is what separates a lost clock from a changed program.

Table view

From	To	Fires at	Cycles	What it was
2025-07-01	2025-08-09	20:00	17	as programmed
2025-08-13	2025-08-31	12:00	9	power cut
2025-09-04	2025-10-21	20:00	20	as programmed
2025-10-26	2025-10-31	07:00	3	power cut
2025-11-06	2026-03-07	20:00	52	as programmed
2026-03-10	2026-03-21	21:00	6	daylight saving
2026-03-24	2026-08-04	20:00	55	as programmed

Why a controller loses the time

The weekdays tell the two causes apart. A controller keeping its own schedule badly and a controller that has lost the schedule entirely both look like a moved start time, but only one of them also moves the days.

Two are power cuts. The controller has a backup battery and it had never been connected — the insulating tab was still in it, straight from the factory. An outage takes the clock with it and the controller comes back knowing neither the hour nor the day, which is why these fire on Wed/Fri/Sun and Tue/Fri/Sun instead of Tue/Thu/Sat. What did not change is the volume: the midday cycles delivered 112 gallons against 107 either side. The program survived; only the clock was lost, which is why the volume evidence elsewhere on this page is undisturbed by it.

One is daylight saving. The controller runs its own clock and does not observe it, so local time slides underneath a schedule it is still keeping perfectly: the hour moves by exactly one and the weekdays do not move at all. The meter sees that as a schedule change because, from the pipe's point of view, that is precisely what it is.

And the seasonal setting rides along with the clock. The winter program is not a different program — it is the same fifteen minutes a zone with the seasonal-adjust dial turned to 50%, which the controller rounds to seven. The meter catches both times it was moved, and neither lands on anything horticultural:

Cycles ran 90 gallons up to 31 October 2025 at 07:00, and 43 from 6 November 2025 at 20:00 — both changed inside the same 6-day gap, which is one trip to the controller.

Cycles ran 47 gallons up to 21 March 2026 at 21:00, and 114 from 24 March 2026 at 20:00 — both changed inside the same 3-day gap, which is one trip to the controller.

Which makes the whole thing one habit rather than two: the trip to fix a clock that cannot fix itself is also the only reliable prompt to change the watering. It works — but it binds the irrigation season to whenever the clock next goes wrong, and it is why summer began on 24 March this year rather than whenever the garden wanted it.

The cost of the August episode is not the water — the cycles were the same size. It is when they ran. Midday in Las Cruces in August is the worst hour of the worst month to water: most of it evaporates before it reaches a root. And with no working battery it will happen again at the next outage, which is a two-dollar fix.

A cycle that delivered 114 gallons now delivers 206

The drip controller opens two valves in series at 20:00 on Tuesday, Thursday and Saturday, and everything it delivers lands inside that one hour. That makes each cycle directly measurable — and through April and May it measured 114 gallons, cycle after cycle.

On Tuesday 2 June 2026 it stepped to 147, and it has climbed every week since — about +7 gallons a cycle per week. The most recent cycles run 206 gallons: 92 more than the clean figure, 81% above it. Across 27 cycles and 63 days that is 1,938 gallons, and it is currently running about 278 gallons a week.

The money is not the point, and the numbers make that plain: at $3.27 per thousand gallons the excess has cost about $6.33 so far — too little to notice on a bill, which is exactly why it ran for nine weeks. The point is the 1,938 gallons. This is the Chihuahuan Desert, the aquifer under it is not being refilled at the rate anyone is drawing on it, and putting that much drinking water into the ground through a split pipe is the kind of waste worth fixing whatever it costs.

Every irrigation cycle in the record

Gallons through the meter in the 20:00 hour, on each night the controller ran.

BeforeAfter

The winter block in the middle is the controller doing what it was told — 14 minutes of valve-open time against 30 in summer. The rise on the right is not that. Note where the record starts: the same season a year earlier sat at 110 gallons.

Table view

Date	Day	gallons per cycle
2025-07-01	Tue	109
2025-07-03	Thu	109
2025-07-05	Sat	113
2025-07-08	Tue	112
2025-07-12	Sat	115
2025-07-15	Tue	112
2025-07-17	Thu	115
2025-07-19	Sat	110
2025-07-22	Tue	109
2025-07-24	Thu	120
2025-07-26	Sat	111
2025-07-29	Tue	109
2025-07-31	Thu	110
2025-08-02	Sat	107
2025-08-05	Tue	109
2025-08-07	Thu	110
2025-08-09	Sat	109
2025-08-13	Wed	112
2025-08-15	Fri	113
2025-08-17	Sun	110
2025-08-20	Wed	113
2025-08-22	Fri	111
2025-08-24	Sun	121
2025-08-27	Wed	105
2025-08-29	Fri	106
2025-08-31	Sun	122
2025-09-04	Thu	55
2025-09-06	Sat	101
2025-09-09	Tue	99
2025-09-11	Thu	103
2025-09-13	Sat	98
2025-09-16	Tue	96
2025-09-18	Thu	96
2025-09-20	Sat	96
2025-09-23	Tue	97
2025-09-25	Thu	97
2025-09-30	Tue	96
2025-10-02	Thu	95
2025-10-04	Sat	95
2025-10-07	Tue	92
2025-10-09	Thu	93
2025-10-11	Sat	90
2025-10-14	Tue	90
2025-10-16	Thu	92
2025-10-18	Sat	92
2025-10-21	Tue	90
2025-10-26	Sun	90
2025-10-28	Tue	90
2025-10-31	Fri	109
2025-11-06	Thu	41
2025-11-08	Sat	43
2025-11-11	Tue	44
2025-11-13	Thu	43
2025-11-15	Sat	41
2025-11-18	Tue	41
2025-11-20	Thu	39
2025-11-22	Sat	40
2025-11-27	Thu	41
2025-11-29	Sat	41
2025-12-02	Tue	40
2025-12-04	Thu	44
2025-12-06	Sat	46
2025-12-09	Tue	47
2025-12-11	Thu	45
2025-12-13	Sat	47
2025-12-16	Tue	49
2025-12-18	Thu	48
2025-12-20	Sat	49
2025-12-23	Tue	49
2025-12-25	Thu	47
2025-12-27	Sat	50
2025-12-30	Tue	51
2026-01-01	Thu	54
2026-01-03	Sat	56
2026-01-06	Tue	48
2026-01-08	Thu	48
2026-01-10	Sat	70
2026-01-13	Tue	47
2026-01-15	Thu	47
2026-01-17	Sat	52
2026-01-20	Tue	48
2026-01-22	Thu	49
2026-01-24	Sat	47
2026-01-27	Tue	47
2026-01-29	Thu	46
2026-01-31	Sat	46
2026-02-03	Tue	49
2026-02-05	Thu	46
2026-02-07	Sat	48
2026-02-10	Tue	45
2026-02-12	Thu	47
2026-02-14	Sat	47
2026-02-17	Tue	48
2026-02-19	Thu	48
2026-02-21	Sat	48
2026-02-24	Tue	48
2026-02-26	Thu	49
2026-02-28	Sat	50
2026-03-03	Tue	49
2026-03-05	Thu	50
2026-03-07	Sat	64
2026-03-10	Tue	49
2026-03-12	Thu	46
2026-03-14	Sat	47
2026-03-17	Tue	47
2026-03-19	Thu	47
2026-03-21	Sat	46
2026-03-24	Tue	106
2026-03-26	Thu	108
2026-03-28	Sat	124
2026-03-31	Tue	119
2026-04-02	Thu	111
2026-04-04	Sat	154
2026-04-07	Tue	110
2026-04-09	Thu	110
2026-04-11	Sat	116
2026-04-14	Tue	160
2026-04-16	Thu	106
2026-04-18	Sat	107
2026-04-21	Tue	106
2026-04-23	Thu	107
2026-04-25	Sat	106
2026-04-28	Tue	106
2026-04-30	Thu	106
2026-05-02	Sat	106
2026-05-05	Tue	107
2026-05-07	Thu	107
2026-05-09	Sat	141
2026-05-12	Tue	108
2026-05-14	Thu	110
2026-05-16	Sat	109
2026-05-19	Tue	109
2026-05-21	Thu	108
2026-05-26	Tue	109
2026-05-28	Thu	110
2026-06-02	Tue	147
2026-06-04	Thu	149
2026-06-06	Sat	157
2026-06-09	Tue	157
2026-06-11	Thu	154
2026-06-13	Sat	165
2026-06-16	Tue	168
2026-06-18	Thu	167
2026-06-20	Sat	168
2026-06-23	Tue	173
2026-06-25	Thu	195
2026-06-27	Sat	180
2026-06-30	Tue	180
2026-07-02	Thu	186
2026-07-07	Tue	192
2026-07-09	Thu	191
2026-07-11	Sat	214
2026-07-14	Tue	199
2026-07-16	Thu	200
2026-07-18	Sat	199
2026-07-21	Tue	199
2026-07-23	Thu	233
2026-07-25	Sat	208
2026-07-28	Tue	206
2026-07-30	Thu	201
2026-08-01	Sat	206
2026-08-04	Tue	209

The climb, as a straight line through the cycles

V(t)=V₀ + k · t

V(t)gallons in one cycle, t weeks after the breakV₀152 gal — the fitted volume at the break itself, already 39 above the clean 114k+7.4 gal per week, fitted by least squares on all 27 cycles since the break

R² 0.84. The intercept is the useful part: the fault did not start at nothing and grow — it arrived as a step of 39 gallons and then began climbing. The line reads 219 gal at the latest cycle against 206 measured, so it slightly over-runs the recent weeks — it describes the climb rather than forecasting it.

Four things it is not

Each of these would leave a different fingerprint on the meter, and the record carries none of them.

Could it be…	Then the meter would show	What it actually shows
The season	the same weeks last year at the same level	110 → 206 gal a cycle, 1.9× the water on the same program
The timer	one step to a new level, then a hold	climbing for 9 weeks, and the 21:00 hour after it still only carries 4 gal, so nothing runs over
A leak elsewhere in the house	the quiet hours rising too	2.00 → 2.00 gal/h on the nights the controller sleeps — unchanged, so it only moves water while a valve is open
A measurement artifact	the billed total unmoved	July billed 4,365 → 6,097 gal

And it was there. What the meter could say was that a valve somewhere was losing 92 gallons a cycle and getting worse. Walking the zones during a run found it in Zone 1. That last step is the one the data could not take: the two zones run back-to-back inside the same hourly bucket, so the meter can say how much a cycle lost and never which valve was open when it did.

What the repair should look like on the meter

The same measurement that found this will confirm the fix, and it is worth writing the numbers down now rather than deciding afterwards what would have counted.

A repaired line puts the 20:00 cycle back to about 114 gallons — not lower. That is what April and May delivered on this same program, and what the previous summer delivered before that. Anything landing near 206 means the loss is still there; anything in between means part of it is.

Two cycles are enough to tell, so a single week's export answers it. The one thing not to read as success is a fall in the daily total — that moves for a dozen reasons, which is how this went unnoticed from June to August. Read the hour.

Worth carrying forward as a standing check rather than a one-off: this whole fault sat 81% above a level the meter had held for two seasons, and nothing else on this page noticed. The alarm is one number — the 20:00 hour on a watering night — and it is already in every export.

The pool explains (almost) everything else

How big it is, then what it explains. Three anomalies across three utilities and one piece of equipment behind all of them — the water-temperature probe turns out to be the most informative sensor in the house.

Four instruments, none of which share a wire

The pool is on no drawing anyone kept, so its volume has to be inferred. Four separate chains do it, and the useful part is that they fail differently: a mistake in one has no way to propagate into the others.

What measured it	How	Answer
Water meter, daily	the March drain and refill, net of a normal day's use	5,028 gal
Water meter, hourly	the same refill, hour by hour, against a measured overnight baseline	5,002 gal
Gas meter + pool thermometer	how fast 270,000 BTU/h raised the water on 4 October	~4,750 gal
Electricity meter + pool thermometer	the pump's own step, which fixes when circulation starts but not how much water it moves	turnover count only

The first three agree to 6%. This page uses 5,000 gallons for pool and spa together, which is the refill rounded — and the refill is only the system volume if the drain emptied it, which is the owner's account rather than anything measured. That caveat is the one real weakness in the chain.

What that volume costs to heat, and to replace

Two figures follow directly from the number above, and they are the ones worth carrying away from this section.

One degree costs 58.2 cf of gas — $0.25 at January's rate, $0.15 at July's, for the 5,000 gallons of pool and spa together. Every heating question on this page reduces to that number multiplied by a temperature difference and a number of days, which is why it is worth stating once and plainly. The two rates differ only because gas is priced seasonally; the physics does not change.

A complete drain and refill costs about $21 — $16 of water for 5,028 gallons, plus $4.74 of gas to bring it back up to temperature from cold city supply.

One more measurement belongs here, because it sizes the spa rather than the whole system and does it without any energy accounting at all. The probe is dropped in at the bottom of a soak's heat-up, so the rise is clean and linear — 55, 57, 55°F an hour across three evenings. Inverting that against the heater's 270k BTU/h input — 216k of it reaching the water at the assumed 80% — gives 469 gallons, against the 550 this page uses. Two methods on different assumptions, agreeing to within 15%, and the rate method needs neither a starting temperature nor any accounting for gas that went to the pool. It is what 550 rests on.

Why one degree costs what it costs

V_cf=V · ρ · ΔTη · h

V_cfgas burned, cubic feet — 58.2 cf per °F hereVwater heated, 5,000 gal of pool and spa togetherρweight of water, 8.34 lb/gal. One BTU raises one pound one °F, so this is also the BTU per gallon per degree — the specific heat is 1 by definition and never appearsΔTtemperature rise, °F — one, hereηappliance efficiency, 80% — assumed, and the only term on this line that is not measured or a constant of naturehheat content of the gas, 896 BTU/cf, taken from the bill's own Mcf-to-Dth conversion rather than a table

Multiply by the gas rate to get money: $0.25 in January against $0.15 in July. The physics is identical; only the tariff moved.

The moment the pump starts, seen three ways

Every 5-minute sample of the year, averaged by time of day.

Water temperature, rate of changeHousehold electrical loadSolar radiation

Read the three panels down the same vertical line. The water spikes at the exact quarter-hour the load steps up, while the sun is already on its way down. Correlation this sharp, across sensors that share no wiring, is about as close to proof as a house gets.

Table view

Time	Water temperature, rate of change (°F / 5 min)	Household electrical load (kW)	Solar radiation (W/m²)
00:00	-0.01	0.92	0
00:15	-0.01	0.88	0
00:30	-0.01	0.88	0
00:45	-0.01	0.88	0
01:00	-0.01	0.88	0
01:15	-0.01	0.88	0
01:30	-0.01	0.88	0
01:45	-0.02	0.88	0
02:00	-0.02	0.88	0
02:15	-0.02	0.84	0
02:30	-0.01	0.88	0
02:45	-0.02	0.88	0
03:00	-0.02	0.84	0
03:15	-0.02	0.84	0
03:30	-0.01	0.88	0
03:45	-0.02	0.88	0
04:00	-0.02	0.84	0
04:15	-0.02	0.88	0
04:30	-0.01	0.88	0
04:45	-0.01	0.88	0
05:00	-0.01	0.88	0
05:15	-0.01	0.84	0
05:30	-0.01	0.84	0
05:45	-0.01	0.84	0.1
06:00	-0.02	0.88	0.71
06:15	-0.01	0.92	2.53
06:30	-0.06	0.92	6.13
06:45	-0.01	1	11.6
07:00	-0.01	1.2	18.8
07:15	0	1.04	26.9
07:30	0.01	1.04	37.2
07:45	0	1.16	54.4
08:00	0	1.16	85.5
08:15	0	1.16	144
08:30	-0.04	1.16	188
08:45	-0.02	1.16	235
09:00	-0	1.16	282
09:15	-0.01	1.16	330
09:30	0.01	1.16	374
09:45	0.01	1.2	416
10:00	-0.01	1.28	449
10:15	-0	1.24	485
10:30	-0.02	1.28	515
10:45	0.01	1.32	548
11:00	0.02	1.32	569
11:15	0.03	1.36	589
11:30	0.03	1.36	603
11:45	0.08	1.4	617
12:00	0.03	1.48	625
12:15	0	1.6	634
12:30	0.03	1.48	639
12:45	0.06	1.56	629
13:00	0.01	1.64	624
13:15	0.03	1.56	619
13:30	0.03	1.6	601
13:45	0.03	1.72	584
14:00	0.03	1.8	557
14:15	0.03	1.68	540
14:30	0.03	1.8	513
14:45	0.03	1.8	480
15:00	0.07	1.72	449
15:15	0.23	2.64	416
15:30	0.03	3	386
15:45	0.04	2.92	343
16:00	0.03	2.96	310
16:15	0.12	3.24	273
16:30	-0.04	3.36	241
16:45	0.01	3.4	206
17:00	-0.03	3.36	175
17:15	-0.04	3.44	149
17:30	-0.02	3.4	119
17:45	-0.01	3.52	97.2
18:00	0.01	3.48	78.3
18:15	0	3.48	59.9
18:30	0	3.4	45.2
18:45	-0.01	3.44	31.5
19:00	0.01	3.44	20.6
19:15	-0.01	3.52	12.2
19:30	-0.04	3.48	6.46
19:45	-0.03	3.44	3.23
20:00	-0.03	3.44	1.25
20:15	-0.02	3.44	0.25
20:30	-0.04	3.32	0.02
20:45	-0.03	3.28	0
21:00	-0.03	3	0
21:15	-0.03	2.92	0
21:30	-0.03	2.88	0
21:45	-0.02	2.8	0
22:00	-0.02	1.8	0
22:15	-0	1.24	0
22:30	-0	1.16	0
22:45	-0	1.12	0
23:00	-0	1.08	0
23:15	-0	1.08	0
23:30	-0.01	1	0
23:45	-0.01	0.96	0

Spa or dry probe — the sun tells them apart

The probe reads above 100°F on 10 days, which pool water never does. Three different things cause it, and they separate cleanly on when they happen and on what the gas meter was doing at the time.

3 are spa soaks. They peak after dark — zero solar — alongside a large gas draw, and they top out at 110.8, 110.8, 110.7°F. The heater is set to 110°F and the alert is set to the same number, so the ceiling is both the thermostat arriving and the phone buzzing.

The probe is lifted out of the pool and dropped into the spa while it heats, which the record shows plainly: a reading flat to a tenth of a degree for half an hour, then a single-sample step of +21, +6, +21°F respectively, then a clean linear climb. It stays in the spa overnight and has equalised with the pool by morning — which is what lets the morning-anchored energy balance elsewhere on this page still work.

6 are the probe out of the water and genuinely baking. They peak in the afternoon under 553–869 W/m² of sun while the gas meter stays idle — nothing is being heated, so the probe is out in the air rather than in the water.

The alert threshold shows up in how long each one lasts. The 2 that climbed past 110°F set the phone off and were back in the water within a mean of 32 minutes. The 4 that stopped short of it raised no alarm and sat out for 129 minutes on average — the worst reaching 109.9°F, a tenth of a degree below the trigger, and going unnoticed for 4.2 hours. A threshold visible in the data purely through what it failed to catch.

One of those 6 is not an accident. 29 March is the hottest reading of the year at 120.2°F, and the probe was where it had always been — the pool was not. That day is dealt with below.

One is the pool heater simply running. 24 August 2025 peaks at 106.7°F in the afternoon with 150 cf on the gas meter — sun and gas together, which is neither of the other two patterns. It is the one excursion where the reading is the water's real temperature.

Three ways for a pool probe to read above 100°F

Every excursion, placed by the sun at its peak and the gas burned that day.

A soak burns 17× the gas of a dry-probe day at none of the sun, which is what makes those two a reading rather than a judgement call. The point between them is the one that matters: 24 August 2025 has sun and gas, and is the one excursion where the probe was reporting the water's real temperature.

Table view

Day	Solar at the peak reading (W/m²)	Gas that day (cf)
Mon 19 Jan 2026: 110.8°F, 18:35–19:30	0	504
Sun 8 Feb 2026: 110.8°F, 19:00–20:25	0	416
Tue 31 Mar 2026: 110.7°F, 19:15–20:55	0	228
Sun 24 Aug 2025: 106.7°F, 15:50–16:15	242	150
Sun 29 Mar 2026: 120.2°F, 15:20–16:25	553	32.0
Fri 15 May 2026: 106.9°F, 11:30–12:10	849	24.0
Mon 18 May 2026: 109.9°F, 12:50–17:05	844	22.0
Sun 21 Jun 2026: 111.6°F, 12:55–12:55	852	22.0
Wed 6 May 2026: 102.2°F, 13:15–16:25	801	18.0
Thu 7 May 2026: 104.5°F, 12:05–12:35	869	18.0

29–30 March: the pool was drained and refilled

The 5,229-gallon event that opened this investigation is a service visit, and the three meters narrate it between them:

Sunday 29th — the probe hits 120.2°F in full afternoon sun starting at 15:15, the exact minute the pump switches on. The pump started and found no water. Pump-window consumption falls to 19.4 kWh, the lowest in weeks: you do not run a pump on a draining pool. 2,192 gallons go back in.

Monday 30th — 3,036 more gallons, and the probe is submerged again. The daily meter splits the same two days 1,922 and 3,307; the hourly one moves 270 gallons from Monday to Sunday and totals the same. Where the two granularities disagree the finer one is closer, and here it is also the one that can say the fill ran through midnight.

Tuesday 31st, then 4–6 April — 2,088 cf of gas, bringing cold fresh water up to temperature. That single stretch is 11% of the year's gas.

The heating events also size the equipment. On 4 October the water climbed 5.45°F/hour while the heater burned 879 cf over the ramp — 270,000 BTU/h of gas going in, against the 333,000 BTU/h input rating its label turned out to carry. Both are input rates, which is the only way the two are comparable; at the assumed 80% that ramp put about 216,000 BTU/h into the water. Working back from that rate gives a system volume near 4,750 gallons, consistent with the 5,229 gallons the meter billed across the two days — and with the 5,002 the hourly register later measured going in once the household's own use is taken back out.

5,229 gallons is the entire system. Pool and spa together hold about 5,000 gallons, so this was a complete drain and refill, not a top-up — routine maintenance on water as hard as Las Cruces'. That also explains the gas: 2,088 cf went into bringing a full body of cold city water up to temperature from scratch, which is why the reheat took three separate days.

The whole episode cost roughly $16 in water and $4.74 in gas: operationally dramatic, financially trivial. Worth knowing before chasing it.

Day	Water (gal)	Gas (cf)	Electric (kWh)	Probe
Fri 27 Mar	76	14	43
Sat 28 Mar	266	26	50
Sun 29 Mar	1,922	32	41	probe dry — pool drained
Mon 30 Mar	3,307	20	63
Tue 31 Mar	189	228	44	spa soak
Wed 1 Apr	79	24	40
Thu 2 Apr	191	24	49
Fri 3 Apr	133	36	44
Sat 4 Apr	309	1,122	52
Sun 5 Apr	185	244	55
Mon 6 Apr	195	494	55
Tue 7 Apr	252	34	—

How the house answers the weather

Balance points are fitted rather than assumed at the conventional 65°F. Where the line meets the axis matters as much as its slope: the intercept is load the weather never touches. The monsoon closes the section — Las Cruces takes most of its year's rain in a few violent afternoons between June and September, and the station catches all of it.

Degree-days, integrated over the day's samples rather than its extremes

CDD(T_b)=1N · ∑ max(0, T_i − T_b)

T_ieach five-minute outdoor reading, °FNsamples that day — 288 when the station misses nothingT_bbalance point, °F — fitted by scanning, not assumed at 65

The conventional (max + min) / 2 form assumes the day is a symmetric curve between its two extremes. Near the balance point — which is exactly where these fits live — that assumption is worth several percent, and the station already reports every five minutes. Heating degree-days are the same with the subtraction reversed.

The energy signature each scatter below is fitting

U=U₀ + k · DD(T_b)

Uthe day's metered use — kWh for electricity, cubic feet for gasU₀intercept: what the house uses when the weather asks for nothing. This is the number the fit exists to producekslope — the marginal cost of one more degree-dayT_bbalance point, scanned over a range and chosen by fit: 70°F cooling, 47°F heating

Fitting the balance point rather than assuming it is what makes the intercept trustworthy: a base temperature that is wrong pushes its error straight into the baseline, and the baseline is what gets called always-on load elsewhere on this page.

Electricity against cooling demand

Daily kWh vs cooling degree-days, balance point fitted at 70°F. R² = 0.74 over 359 days.

The line meets the axis at 38 kWh/day. That intercept is consumption no amount of mild weather removes — it is the floor plus the timer plus everything you plug in. Each cooling degree-day above 70°F adds 2.17 kWh. Cooling starts late, at 70°F, which is a well-behaved house, not a leaky one.

Table view

Day	Cooling degree-days (base 70°F)	kWh per day	Role
27 Jul 2026	20.6	102	in fit
26 Jul 2026	21.0	102	in fit
1 Aug 2026	14.6	95.7	in fit
8 Jul 2026	18.1	89.9	in fit
30 Jul 2026	22.8	88.6	in fit
9 Aug 2025	18.3	88.3	in fit
24 Jul 2026	20.2	87.6	in fit
21 Jun 2026	17.8	86.9	in fit
23 Jul 2026	20.2	86.8	in fit
2 Jun 2026	14.5	85.0	in fit
25 Jul 2026	20.4	84.7	in fit
7 Jun 2026	14.6	84.0	in fit
29 Jul 2026	20.1	83.7	in fit
4 Jul 2026	20.1	83.6	in fit
6 Jun 2026	11.8	83.2	in fit
10 Jul 2026	17.3	83.1	in fit
12 Jul 2026	16.6	82.8	in fit
28 Jul 2026	19.9	82.4	in fit
3 Jul 2026	19.2	81.9	in fit
13 Jun 2026	15.8	81.4	in fit
27 Jun 2026	16.9	81.1	in fit
17 Jun 2026	17.3	80.8	in fit
31 Jul 2026	20.2	80.2	in fit
5 Jul 2026	13.3	80.1	in fit
9 Jul 2026	17.8	79.9	in fit

Gas against heating demand

Daily cubic feet vs heating degree-days, balance point fitted at 47°F.

Explained by the fitHeld out of the fit (17)

Two R² values, because only reporting the flattering one would mislead. On the 347 routine days the fit keeps, R² = 0.83 — heating demand explains those days well. Across all 364 days it is only 0.13, because the 17 days held out of it carry so much volume. Your house starts calling for heat once the outside temperature dips below about 47°F, and burns 19 cf/day when it is not heating at all — that floor is the water heater.

Table view

Day	Heating degree-days (base 47°F)	Cubic feet per day	Role
4 Apr 2026	0	1,122	outlier
4 Oct 2025	0	898	outlier
13 Jan 2026	8.49	684	outlier
19 Jan 2026	5.96	504	outlier
6 Apr 2026	0.05	494	outlier
23 May 2026	0	444	outlier
8 Feb 2026	3.55	416	outlier
5 Apr 2026	0	244	outlier
17 Jan 2026	7.99	232	in fit
31 Mar 2026	0	228	outlier
27 Jan 2026	9.24	224	in fit
26 Jan 2026	12.5	212	in fit
10 Jan 2026	11.9	206	in fit
11 Jan 2026	11.1	204	in fit
12 Jan 2026	9.2	186	in fit
30 Dec 2025	8.99	182	in fit
8 Dec 2025	6.7	180	in fit
29 Dec 2025	5.94	172	in fit
28 Jan 2026	7.74	170	in fit
31 Jan 2026	6.93	162	in fit
16 Dec 2025	4.63	156	in fit
24 Aug 2025	0	150	outlier
18 Jan 2026	9.36	148	in fit
2 Dec 2025	3.95	146	in fit
9 Jan 2026	5.09	138	in fit

The line through those points

G=G₀ + k · HDD(T_b)

Ggas that day, cubic feetG₀19 cf/day — what the house burns with no heating demand at all, which is the water heater's standing drawk16.2 cf per degree-day — the envelope, and the only term here that describes the building rather than the weather or the appliancesT_bbalance point, 47°F. Fitted, not assumed: every candidate from 45 to 80°F was tried and this one leaves the straightest line

R² 0.83 on the 347 days the fit keeps.

And what a heating degree-day means here

HDD(T_b)=∑ max(0, T_b − T_i)n

T_ieach temperature sample that daynsamples in the day — 288 at five-minute spacing

Averaged across the day's samples rather than taken from its mean, and at this balance point the difference is not cosmetic: integrating finds 387 degree-days a year against 147 from daily means, 2.6× more. 103 days average above 47°F and still spend part of the night below it. A daily mean would score every one of them zero.

What the gas meter does across a day

Mean cubic feet in each hour, grouped by the day's mean outdoor temperature. Every panel on the same scale, each against the annual mean for reference.

Temperature bandEvery day

The hot-day panel is the water heater, alone and visible. It is nearly flat at 0.8 cf an hour and totals 19 cf a day — against the 19 cf/day the regression above puts at its intercept, reached by fitting 364 daily totals against degree-days. Two methods with nothing in common agreeing to within a cubic foot. And flat matters: a standing loss looks like this, whereas hot water actually being drawn would show morning and evening humps.

The cold-day panel is the furnace, and it is concentrated: 22 cf in the 06:00 hour alone, 12% of the whole day's gas, which is the recovery from a night setback.

Table view

	Every day	Cold day (22d)	Mild day (100d)	Hot day (85d)
00:00	1.31	2.64	0.48	0.81
01:00	1.86	4.64	0.48	0.47
02:00	1.96	5.18	0.38	0.33
03:00	1.98	5.18	0.4	0.36
04:00	1.99	5.09	0.5	0.36
05:00	4.64	12.5	0.72	0.66
06:00	8.49	22.5	2.12	0.89
07:00	5.31	13.6	1.8	0.49
08:00	4.52	11.2	1.42	0.96
09:00	3.81	7.36	3.44	0.64
10:00	4.43	8.18	4.56	0.54
11:00	2.94	4.91	3.24	0.66
12:00	2.17	3.55	2.3	0.66
13:00	1.66	2.91	1.36	0.71
14:00	1.97	3.27	1.94	0.71
15:00	5.84	11.4	3.72	2.42
16:00	6.7	16.0	3.24	0.87
17:00	2.81	5.73	1.38	1.32
18:00	2.69	4.64	2.64	0.8
19:00	4.02	8.36	2.8	0.89
20:00	2.6	4.09	2.7	1
21:00	4.21	9.45	1.94	1.24
22:00	2.35	5.36	0.92	0.78
23:00	1.32	2.82	0.72	0.44

How much of the weather gets inside

Indoor temperature swing against outdoor swing, on the 99 mild days when the thermostat is mostly idle.

The house passes 11% of each extra degree of outdoor swing through to the inside — the slope of the fit, which is what survives a thermometer reading a little high. In plain averages an outdoor day moves 30°F and the interior 4.1°F, a slightly larger fraction because the line does not pass through the origin. That is a genuinely good envelope, and it agrees with the late 70°F cooling balance point found above — two independent measurements of the same insulation. For contrast, the garage passes 41% of the same extra degree — 4× the house, fitted the same way on the same days — which is what an uninsulated space looks like. Its plain-average figure runs higher still for the same intercept reason, and the zones section ranks all four buildings on it.

Table view

Day	Outdoor daily temperature swing (°F)	Indoor daily swing (°F)
1 Mar 2026 · outdoor mean 67°F	43.9	7.2
25 Feb 2026 · outdoor mean 63°F	39.4	7
18 Mar 2026 · outdoor mean 69°F	41.3	6.6
26 Feb 2026 · outdoor mean 65°F	34.0	6.6
20 Dec 2025 · outdoor mean 61°F	25.3	6.5
2 Mar 2026 · outdoor mean 65°F	40.5	6.5
12 Mar 2026 · outdoor mean 60°F	32.2	6.5
15 Mar 2026 · outdoor mean 69°F	32.0	6.5
27 Feb 2026 · outdoor mean 65°F	38.7	6.3
15 Nov 2025 · outdoor mean 62°F	34.9	6.1
26 Oct 2025 · outdoor mean 63°F	36.8	6
11 Feb 2026 · outdoor mean 59°F	28.6	6
14 Mar 2026 · outdoor mean 66°F	36.4	5.9
5 Mar 2026 · outdoor mean 62°F	33.8	5.8
31 Oct 2025 · outdoor mean 61°F	41.6	5.8
28 Feb 2026 · outdoor mean 66°F	36.2	5.8
4 Mar 2026 · outdoor mean 61°F	34.2	5.8
10 Feb 2026 · outdoor mean 60°F	25.6	5.6
2 Nov 2025 · outdoor mean 59°F	38.5	5.6
14 Nov 2025 · outdoor mean 61°F	36.2	5.6
25 Dec 2025 · outdoor mean 60°F	24.1	5.6
11 Mar 2026 · outdoor mean 59°F	39.6	5.6
3 Apr 2026 · outdoor mean 68°F	30.4	5.4
19 Mar 2026 · outdoor mean 71°F	45.5	5.4
6 Nov 2025 · outdoor mean 63°F	38.0	5.2

What a day costs, against how warm it was

Every day's electricity and gas priced at the rate that applied to it, against that day's mean outdoor temperature.

Neither meter makes this shape on its own — gas falls as it warms and electricity climbs, so only the sum has a minimum. It sits at 58°F, which is roughly the outdoor temperature at which this house wants nothing from either utility.

Table view

Day	Outdoor mean temperature (°F)	Cost that day ($)
91°F · $13.42	90.6	13.4
91°F · $13.36	91.0	13.4
88°F · $12.89	88.3	12.9
85°F · $12.53	84.6	12.5
88°F · $11.79	88.1	11.8
90°F · $11.64	90.1	11.6
93°F · $11.59	92.8	11.6
90°F · $11.50	90.2	11.5
85°F · $11.40	84.5	11.4
88°F · $11.38	87.7	11.4
90°F · $11.37	90.2	11.4
84°F · $11.25	84.3	11.3
86°F · $11.23	85.5	11.2
80°F · $11.12	80.4	11.1
90°F · $11.08	90.4	11.1
90°F · $10.98	90.1	11.0
90°F · $10.95	90.1	10.9
87°F · $10.92	87.3	10.9
86°F · $10.89	85.8	10.9
87°F · $10.85	86.6	10.9
87°F · $10.80	87.3	10.8
90°F · $10.79	89.9	10.8
88°F · $10.76	87.9	10.8
89°F · $10.73	89.2	10.7
87°F · $10.62	86.9	10.6

Heat costs this house about 1.9× what equivalent cold does

The curve has a floor of $3.30 a day at 58°F, and it is not symmetric about it. The hottest band runs $11.08 a day — 3.4× the floor — while the coldest runs $5.68, only 1.7×.

Three things stack up on the warm side and none of them apply on the cold side. Cooling is electric and heating is mostly gas, and a marginal summer kilowatt-hour costs 1.55× a winter one before the fuel difference is even counted. The summer tier then bites on exactly the months that need the most. And the pool — pump and evaporation both — is a summer load that winter simply does not have.

The practical reading is about which direction to insure against. A degree of unexpected heat is worth roughly twice a degree of unexpected cold on this bill, so shade, envelope and cooling efficiency are where money goes furthest here — and it is the opposite of the advice that would suit the same house further north.

Rain and lightning through the year

Daily totals from the station. Each panel keeps its own scale.

Rain and lightning arrive together in a narrow season and are absent for the rest of the year. Wind is the exception: the year's strongest gust, 43 mph on 21 April 2026, came in spring with no rain at all. Those are the dust events, and they are a different phenomenon from the summer storms.

Table view

Date	Daily rainfall (in)	Lightning strikes (per day)	Peak gust (mph)
2025-08-07	0	0	15.9
2025-08-14	0	0	14.8
2025-08-21	0	0	11.4
2025-08-28	0	0	11.4
2025-09-04	0	0	15.9
2025-09-11	0	0	12.5
2025-09-18	0.21	18.0	21.7
2025-09-25	0	0	13.6
2025-10-02	0	0	9.2
2025-10-09	0	0	13.6
2025-10-16	0	0	15.9
2025-10-23	0.02	17.0	20.6
2025-10-30	0	0	8.1
2025-11-06	0	0	14.8
2025-11-13	0	0	13.6
2025-11-20	0.25	1	23.9
2025-11-27	0	0	11.4
2025-12-04	0	0	10.3
2025-12-11	0	0	11.4
2025-12-18	0	0	10.3
2025-12-25	0	0	12.5
2026-01-01	0	0	14.8
2026-01-08	0	0	28.6
2026-01-15	0	0	10.3
2026-01-22	0	0	9.2
2026-01-29	0	0	14.8
2026-02-05	0	0	11.4
2026-02-12	0	0	11.4
2026-02-19	0	0	20.6
2026-02-26	0	0	13.6
2026-03-05	0	0	21.7
2026-03-12	0	0	11.4
2026-03-19	0	0	11.4
2026-03-26	0	0	19.5
2026-04-02	0	0	15.9
2026-04-09	0	0	10.3
2026-04-16	0	0	17.2
2026-04-23	0	0	23.9
2026-04-30	0	0	21.7
2026-05-07	0	0	13.6
2026-05-14	0	0	17.2
2026-05-21	0	0	17.2
2026-05-28	0	0	14.8
2026-06-04	0.06	0	14.8
2026-06-11	0	0	12.5
2026-06-18	0	0	17.2
2026-06-25	0	0	19.5
2026-07-02	0	0	13.6
2026-07-09	0	0	27.3
2026-07-16	0	1	18.3
2026-07-23	0	0	11.4
2026-07-30	0	0	14.8

Rainfall by month

Every month in the record. The year starts and ends mid-August, so August appears twice — as a 25-day bar and a 6-day one.

Rainfall

September 2025, January 2026 and July 2026 carry the year — and note that they are not all monsoon months, which is the reminder that winter rain here arrives too. A drip system on a fixed weekly schedule waters straight through every one of them; a rain sensor, or a controller that skips after a storm, is the cheapest water saving on this page.

Table view

Month	Rainfall
08/25	0.42
09/25	2.74
10/25	0.59
11/25	0.67
12/25	0.25
01/26	1.83
02/26	0.04
03/26	0.04
04/26	0.01
05/26	0.84
06/26	1.12
07/26	1.73
08/26	0.24

Four boxes in the same weather

The house, patio, garage and shed stand in identical weather and are built to four different standards. That makes the weather an instrument: outdoors supplies the forcing, and each zone's response measures how well it is separated from it.

Shed

1.09×

of the outdoor swing · 1 h behind
+5°F on the daily low, +8°F on the high

Garage

0.58×

of the outdoor swing · 3 h behind
+16°F on the daily low, +3°F on the high

Patio

0.19×

of the outdoor swing · follows its thermostat
+17°F on the daily low, -7°F on the high

House

0.14×

of the outdoor swing · follows its thermostat
+20°F on the daily low, -6°F on the high

The shed is not a control — it is an amplifier

Four boxes, one weather. Ranked by how much of the outdoor daily swing each one lets through, they fall in the order you would guess — except at the top. The shed passes 1.09× the outdoor swing: it does not merely fail to insulate, it adds to the weather. Its daily high averages +7.6°F against the outdoor high and it still runs +4.9°F warm at dawn, peaking at 114°F over the year — hotter than the 111°F the outdoor sensor ever saw. That is a solar oven with enough mass to stay warm overnight, not a passive box.

The garage is the interesting one. It cuts the swing to 0.58× and runs 3 hours behind the weather, the longest lag of any zone — that delay is thermal mass, the slab and the shared wall. But its protection is lopsided: +15.6°F on the daily low against only +2.9°F on the high. It is very good at keeping cold out and barely does anything about heat. Anything stored there that minds freezing is safe; anything that minds 112°F is not.

One day, four buildings — the 30 hottest days of the year

Every panel on the same scale, each against the same outdoor curve.

ZoneOutdoors

The shed tracks the sun; the house ignores it. Outdoors peaks at 16:00 — the garage crests later, which is the thermal mass showing up as a delay rather than as a smaller number.

Table view

	Outdoors	Shed	Garage	Patio	House
00:00	82.9	88.1	91.9	73.7	75.3
01:00	81.3	86.3	90.4	73.7	75.9
02:00	79.8	84.6	89.2	73.7	76.3
03:00	78.4	83.0	88.0	73.8	76.6
04:00	77.1	81.6	87.0	73.8	76.8
05:00	75.4	80.3	86.1	73.9	76.9
06:00	74.6	79.0	85.3	74.0	77.0
07:00	77.0	78.4	84.5	73.5	75.5
08:00	82.4	79.2	84.2	74.1	75.3
09:00	87.6	83.1	84.9	74.1	75.2
10:00	91.3	88.7	86.9	74.0	75.2
11:00	94.7	94.7	89.8	74.2	75.3
12:00	97.7	99.3	93.1	74.6	75.4
13:00	100	102	96.5	75.1	75.5
14:00	102	105	99.9	75.5	75.5
15:00	103	107	102	75.7	75.5
16:00	103	109	104	76.0	75.7
17:00	101	108	105	76.0	76.1
18:00	99.4	106	105	75.5	76.0
19:00	95.8	102	103	74.7	75.7
20:00	91.9	98.0	101	74.0	75.4
21:00	88.2	94.9	98.6	73.8	75.1
22:00	86.7	92.4	96.2	73.7	75.0
23:00	85.3	90.4	94.1	73.6	75.0

The shared wall, measured

Each dot is one day. Garage minus shed, against house minus outdoors.

Slope 0.180, R²=0.649 over 363 days. The line runs through both halves of the year: when the house is cooler than outdoors, the garage goes cool relative to the shed by the same fraction.

Table view

House warmer than outdoors (°F)	Garage warmer than shed (°F)
-17.0	-1.62
-13.7	1.35
-11.3	1.43
-9.03	0.88
-7.05	1.61
-4.95	2.1
-2.47	4.15
-0.87	0.89
0.9	4.56
2.24	1.57
4.07	2.88
5.32	1.47
7.04	3.39
9.13	0.1
10.5	6.67
12.1	4.87
13.3	4.93
15.6	2.84
17.5	6.04
18.9	4.93
21.0	7.97
23.3	5.79
25.1	8.89
28.0	8.58
33.6	13.0

About 18% of the house reaches the garage, and 31% of its moisture

The garage is compared with the shed here, not with outdoors. Both are unconditioned boxes on the same property carrying the same model of sensor. The one difference between them is that the garage shares a wall with the house — so garage minus shed isolates that wall, and regressing it on house minus outdoors gives the fraction of the house's gradient that arrives next door.

18.0% of the temperature gradient (R²=0.649) and 31.1% of the moisture gradient (R²=0.686) get through. Moisture crosses more readily than heat, which fits a gap that leaks air: water vapour carried into the garage has nowhere to go, while the heat that rides in with it conducts straight back out through an uninsulated door.

The patio, by contrast, is not really an outbuilding at all. It follows the house's temperature gradient at a slope of 1.015 with R²=0.993 — one-to-one, sitting a steady 1.5°F off the house. Its moisture slope of 0.86 is the only sign it is not simply another room: about 14% more outdoor air gets in, which is what a wall of sliding doors would do.

That wall of sliding doors, measured

Two sliding doors face due south, dual pane with low-E, 76 sq ft of glass. Orientation had to be measured rather than guessed, and this is why: projected onto that surface, a due-south window here collects more energy in December than a horizontal one does (ratio 1.01) and barely a third as much in June (0.39). The summer sun is too high to get in; the winter sun strikes it nearly square.

So the glass is a passive-solar asset, and a measurable one. Of the 969 kWh/m² landing on that face across the year, 358 arrives while the zone actually wants heat — which shows up as a gap between the two heat-loss coefficients. The envelope's is 89 BTU/h/°F; the heat pump only has to cover 57. 36% of the winter heating load is carried by the windows for free.

The low-E coating is close to a wash in energy — about 478 kWh of cooling saved against clear glass, 430 kWh of heating added — but it wins on money, because a summer kilowatt-hour costs 1.55× a winter one. And it keeps the room usable on a July afternoon, which no energy balance captures.

Water in the air, by month

Mixing ratio relative to outdoors — above the line is wetter than outside.

ZoneOutdoors

Relative humidity would mostly restate the thermometers here. Mixing ratio is absolute: equal readings mean equal water, whatever the temperature.

Table view

	Shed	Garage	Patio	House
Aug	-0.96	-0.95	-0.9	-0.73
Sep	-0.72	-1.18	-1.14	-1.09
Oct	-0.31	-0.06	0.92	1.42
Nov	-0.2	0.12	1.56	2.14
Dec	-0.08	0.3	1.56	2.13
Jan	-0.08	0.32	1.37	1.9
Feb	0.03	0.46	1.44	2
Mar	-0.01	0.86	2.07	2.8
Apr	-0.29	0.47	1.72	2.44
May	-0.28	0.33	1.19	1.9
Jun	-0.81	-0.48	-0.38	-0.05
Jul	-0.99	-1.41	-2.15	-2.02
Aug	-0.79	-1	-1.65	-1.39

Why every moisture comparison here is a mixing ratio, not a humidity

w=1000 · ε · ep − e with e = 6.112 · exp17.67 · T_dT_d + 243.5

wgrams of water vapour per kilogram of dry airεratio of molar masses, 0.622evapour pressure, hPa — the saturation pressure at the dew point, which is what a dew point meansT_ddew point, °C, as transmitted by each zone's sensorpstation absolute pressure, hPa — measured, not assumed: at 3,900 ft the air is about 12% thinner than sea level

Relative humidity divides by a temperature-dependent denominator, so comparing a 50°F garage with a 74°F house on RH would mostly restate their thermometers. Mixing ratio is absolute: equal readings mean equal water.

The house is wetter than outside in winter and drier in summer — the air conditioner is why

In winter the house holds +2.0 g/kg more water than the air outside — occupants, showers and cooking, with a tight envelope keeping it in. In July and August it holds -1.4 g/kg, drier than outdoors. Nothing in the house removes water on purpose; that is the air conditioner condensing it out of the air as a side effect of cooling, and dumping it on the ground.

The 3.4 g/kg reversal is the one moisture number here that no instrument error can manufacture. Comparing two sensors risks their calibration; comparing one sensor with itself in a different season does not. A hygrometer reading half a gram high all year shifts every month equally and leaves the swing exactly where it is.

The zones fall in the same order in both directions. The patio follows the house closely, the garage weakly, the shed not at all — and the shed sitting flat against outdoors all year is the useful negative: it says the ladder is measuring the building, not five sensors disagreeing.

Where the monsoon gets in, and how long it stays

Mixing ratio against the day before, averaged over 23 rain onsets.

ZoneOutdoors

Rain lands on day 0. Outdoor air peaks the next day and is most of the way back to normal by day 3.

Table view

	Outdoors	Shed	Garage	Patio	House
−1	0	0	0	0	0
0	1.8	1.16	0.6	0.54	0.33
+1	2.22	2	1.16	1.01	0.62
+2	1.15	1.14	0.92	0.66	0.51
+3	0.68	0.82	0.73	0.43	0.4
+4	0.68	0.75	0.71	0.35	0.33

The garage dries out last

Four days after the rain, outdoor air has shed 69% of the moisture it took on and the shed 62%. The garage has released only 38%. All four zones peak on the same day, +1; what separates them is what happens next. By day 2 the shed has given back 43% of its peak and the outside air 48%, while the garage has released just 21% — it is still sitting at its wettest well after the sky has cleared.

Less wet air gets into the garage than into the shed — the peak is 1.16 g/kg against the shed's 2.00 — but what does get in is slower to leave: on day 4 the garage still holds 62% of its peak while the shed holds 38% and the outside air 31%. That is what a slab and a stack of stored material do: they absorb water vapour on the way in and give it back slowly. The practical reading is that the garage spends several days a monsoon at its own private humidity, well after the weather has moved on.

“But it’s a dry heat” — except for $75 of it

Degree-days measure heat, not water, so an air conditioner's work condensing moisture is invisible to them. Adding outdoor mixing ratio as a second term to the cooling-season regression over 216 days puts a number on it: +0.73 kWh per g/kg per day (t = 4.5), lifting R² from 0.685 to 0.708.

The two terms are close to independent — VIF 1.15 — which is what makes the split meaningful rather than an artifact of humid days also being hot ones. Between a dry cooling day (3.7 g/kg, 70 of them) and a monsoon one (12.5 g/kg, 49), the humidity alone accounts for 6.3 kWh, or $0.83 a day at the summer marginal rate of $0.1304.

Across the season, measured against the driest quarter of cooling days rather than against an impossible zero, that is 578 kWh — $75. Not large next to the $400 the pool pump costs, but it is real, and it is the part of the summer bill that arrives with the clouds rather than with the heat.

So the local consolation is mostly true and precisely bounded. It is a dry heat, for 70 of the cooling days measured here — and on the 49 monsoon days it is not, at $0.83 a day. The saying is correct about the climate and wrong about the bill.

What the weather cannot explain

Every day is scored against what its own weather predicts. What survived that test then had to be explained from elsewhere in the record — and almost all of it was.

15 days flagged, 13 of them now accounted for — and every one by the pool

The detector's job is narrow: it says a day departs from what the weather predicts, and nothing more. Explaining those departures took the water-temperature probe, the calendar, and the other two meters. Keeping the two jobs apart matters — thresholds tuned until a story emerges are not evidence.

7 gas days: the pool heater. Together 3,766 cf — 21% of the year's gas. Most are identified by energy balance rather than by the calendar: the water's heat gain from one morning to the next accounts for a large share of the missing gas, which works in any season. 5 of 7 fell on a weekend, about 2.5× what chance would give.
3 gas days: spa soaks. Each peaked after sunset — zero solar — with the probe deliberately moved into the spa to watch it climb to 110°F. A furnace does not know it is evening.
2 water days: the drain and refill. 5,229 gallons across 29–30 March, the whole system going back in.
1 water day falls on the same date as a pool-heating event — a top-up alongside the heater.

So the year's anomalies across three separate utilities trace to one piece of equipment. That is the useful result: not that 15 days were odd, but that they were odd for a single reason, and the reason was findable from the data already being collected.

2 remain open, and are worth stating plainly rather than folded into the story:

10 May 2026 — water, 692 gal against 130 expected. Sunday, not an irrigation day. Electricity that day also ran above normal, which fits household activity rather than equipment.
23 November 2025 — water, 326 gal against 88.5 expected. Sunday, not an irrigation day. Electricity that day also ran above normal, which fits household activity rather than equipment.

This test found no leak, and there was one. It watches the quietest day of each week, which for this house sits at 54 gallons, and reports anything holding that floor above 135. Nothing did, all year — and that was correct, not a miss. A split irrigation line only loses water while its valve is open, so the quiet days stayed quiet and the floor never moved while 1,938 gallons went into the ground. The failure was in the question, not the arithmetic.

What it does still cover is a leak that runs continuously: a running toilet or a failed flapper would lift the floor well past the trigger. Below roughly 81 gallons a day even that hides inside the household's own variation, so a dripping tap is invisible at daily resolution. Two different faults, two different monitors — and the one that mattered here is the cycle volume.

The March refill left nothing running behind it: the weekly floor in the month afterwards sat at 65 gallons against 53 in the month before, which is ordinary seasonal drift rather than a new baseline. A drain and refill is a discrete event, and this one closed.

All 15 flagged days

Sat 23 May 2026Gas — Pool heater — the water absorbed the missing gas444 cf vs 18.8 expected

Sat 23 May 2026Water — Same day as a pool-heating event744 gal vs 131 expected

Sun 10 May 2026Water — Not accounted for692 gal vs 130 expected

Mon 6 Apr 2026Gas — Pool heater — the water absorbed the missing gas494 cf vs 19.6 expected

Sun 5 Apr 2026Gas — Pool heater — the water absorbed the missing gas244 cf vs 18.8 expected

Sat 4 Apr 2026Gas — Pool heater — the water absorbed the missing gas1,122 cf vs 18.8 expected

Tue 31 Mar 2026Gas — Spa soak — probe moved into the spa, reached 110°F after dark228 cf vs 18.8 expected

Mon 30 Mar 2026Water — Drain and refill of the whole pool and spa3,307 gal vs 91.5 expected

Sun 29 Mar 2026Water — Drain and refill of the whole pool and spa1,922 gal vs 86.0 expected

Sun 8 Feb 2026Gas — Spa soak — probe moved into the spa, reached 110°F after dark416 cf vs 76.1 expected

Mon 19 Jan 2026Gas — Spa soak — probe moved into the spa, reached 110°F after dark504 cf vs 115 expected

Tue 13 Jan 2026Gas — Pool heater — the water absorbed the missing gas684 cf vs 156 expected

Sun 23 Nov 2025Water — Not accounted for326 gal vs 88.5 expected

Sat 4 Oct 2025Gas — Pool heater — the water absorbed the missing gas898 cf vs 18.8 expected

Sun 24 Aug 2025Gas — Pool heater — the water absorbed the missing gas150 cf vs 18.8 expected

Water, day by day

Daily gallons across the year.

Water at this house is not weather-driven — with refrigerated air there is no evaporative cooling load, so what remains is irrigation and behavior. That is why water is judged against its own trailing baseline rather than against temperature.

Table view

Date	Day	Value	Unit
2026-03-30	Mon	3,307	gal
2026-03-29	Sun	1,922	gal
2026-05-23	Sat	744	gal
2026-05-10	Sun	692	gal
2025-09-27	Sat	455	gal
2026-07-04	Sat	402	gal
2026-07-26	Sun	395	gal
2026-07-02	Thu	375	gal
2026-07-23	Thu	354	gal
2026-04-14	Tue	347	gal
2026-06-18	Thu	343	gal
2026-07-07	Tue	340	gal
2025-11-23	Sun	326	gal
2026-04-11	Sat	322	gal
2026-07-11	Sat	315	gal
2026-08-04	Tue	314	gal
2026-04-04	Sat	309	gal
2026-06-27	Sat	304	gal
2026-06-02	Tue	300	gal
2026-08-01	Sat	295	gal
2026-07-28	Tue	293	gal
2026-07-30	Thu	292	gal
2026-06-15	Mon	289	gal
2026-06-04	Thu	287	gal
2025-09-25	Thu	285	gal

Gas, day by day

Daily cubic feet across the year.

The winter block is heating. The isolated dark cells outside it are not.

Table view

Date	Day	Value	Unit
2026-04-04	Sat	1,122	cf
2025-10-04	Sat	898	cf
2026-01-13	Tue	684	cf
2026-01-19	Mon	504	cf
2026-04-06	Mon	494	cf
2026-05-23	Sat	444	cf
2026-02-08	Sun	416	cf
2026-04-05	Sun	244	cf
2026-01-17	Sat	232	cf
2026-03-31	Tue	228	cf
2026-01-27	Tue	224	cf
2026-01-26	Mon	212	cf
2026-01-10	Sat	206	cf
2026-01-11	Sun	204	cf
2026-01-12	Mon	186	cf
2025-12-30	Tue	182	cf
2025-12-08	Mon	180	cf
2025-12-29	Mon	172	cf
2026-01-28	Wed	170	cf
2026-01-31	Sat	162	cf
2025-12-16	Tue	156	cf
2025-08-24	Sun	150	cf
2026-01-18	Sun	148	cf
2025-12-02	Tue	146	cf
2026-01-09	Fri	138	cf

Where the gas goes

Annual cubic feet, split by the heating signature. 98% of metered gas is accounted for.

Cubic feet

Space heating is only 34% of the gas — less than the water heater's year-round standing draw, and the reason the whole gas bill loses to its own access fee in the cost section. The house starts calling for heat when the outside temperature dips below 47°F (rather than the conventional 65°F), which cuts the heating degree-days that matter from 2,215 to 387 — and only 28 nights all year fell below freezing.

Table view

End use	Cubic feet
Space heating	6,244
Water heater + cooking	6,845
Pool + spa	4,704

The long view

First the year as the meters recorded it — four measures on four separate scales, never two scales on one frame, because that manufactures correlations the data does not contain. Then the same span as the biller saw it: the only source here carrying cost, answering not what the house does but what it is charged for.

Daily consumption and temperature

Each panel keeps its own y-axis; only the dates are shared.

Electricity and temperature move together. Gas moves opposite to both. Water does neither, which is the finding — it is on a human schedule, not a thermal one.

Table view

Date	Electricity (kWh)	Water (gal)	Natural gas (cf)	Outdoor mean temperature (°F)
2025-08-07	79.8	202	10.0	90.1
2025-08-14	68.3	165	14.0	85.4
2025-08-21	63.5	77.0	10.0	81.7
2025-08-28	45.1	127	10.0	84.7
2025-09-04	54.9	180	14.0	75.0
2025-09-11	59.7	259	14.0	83.4
2025-09-18	30.9	115	10.0	72.4
2025-09-25	61.0	285	24.0	73.3
2025-10-02	50.4	154	14.0	74.7
2025-10-09	51.1	224	14.0	75.2
2025-10-16	50.3	154	14.0	70.3
2025-10-23	42.2	137	18.0	70.4
2025-10-30	41.3	90.0	26.0	54.3
2025-11-06	42.6	174	24.0	62.9
2025-11-13	38.9	102	22.0	57.2
2025-11-20	32.5	150	24.0	49.3
2025-11-27	35.0	86.0	80.0	49.6
2025-12-04	42.2	160	108	45.3
2025-12-11	41.6	97.0	56.0	49.5
2025-12-18	39.1	176	54.0	52.4
2025-12-25	43.0	113	96.0	60.0
2026-01-01	42.0	207	82.0	55.3
2026-01-08	37.9	96.0	40.0	47.0
2026-01-15	47.2	179	52.0	45.4
2026-01-22	36.7	106	116	48.9
2026-01-29	50.1	173	128	44.5
2026-02-05	40.9	106	58.0	45.0
2026-02-12	36.8	169	36.0	59.5
2026-02-19	38.8	114	32.0	51.0
2026-02-26	37.0	182	20.0	65.3
2026-03-05	40.4	123	20.0	61.7
2026-03-12	40.7	183	16.0	60.4
2026-03-19	41.8	113	20.0	71.1
2026-03-26	51.7	262	14.0	75.8
2026-04-02	48.9	191	24.0	64.2
2026-04-09	41.0	279	28.0	70.1
2026-04-16	44.7	188	18.0	64.8
2026-04-23	46.3	267	14.0	69.4
2026-04-30	38.0	182	14.0	64.9
2026-05-07	49.2	255	18.0	66.7
2026-05-14	57.7	177	20.0	80.7
2026-05-21	54.5	264	14.0	74.2
2026-05-28	40.3	175	14.0	77.6
2026-06-04	56.2	287	12.0	71.8
2026-06-11	71.3	244	14.0	87.4
2026-06-18	58.4	343	14.0	83.5
2026-06-25	71.8	285	18.0	90.6
2026-07-02	72.4	375	14.0	86.6
2026-07-09	79.9	263	14.0	87.8
2026-07-16	74.5	280	10.0	83.8
2026-07-23	86.8	354	16.0	90.2
2026-07-30	88.6	292	14.0	92.8

Monthly totals

Complete calendar months only.

Electricity (kWh)

Table view

Month	Electricity (kWh)
09/25	1,608
10/25	1,422
11/25	1,179
12/25	1,351
01/26	1,490
02/26	1,142
03/26	1,344
04/26	1,324
05/26	1,537
06/26	2,132
07/26	2,404

The summer tier is the whole story

EPE prices summer in two steps: the first 600 kWh at $0.07035, everything above at $0.10876 — 55% more. Winter is flat at $0.05816 with no tier at all. Applying that formula reproduces the energy charge on 30 of your 35 bills to the cent; the 5 it misses are all from 2023 and sit 0.5% off, which is a rate change rather than a broken model.

Your summer bills run 1,600–2,400 kWh, so roughly three-quarters of every summer kWh is priced in the upper tier. That is why the marginal cost of summer electricity — $0.1304/kWh once riders, franchise fee and tax are added — is nearly 1.6× the winter figure, and well above the $0.1071 blended average these bills imply.

Decisions should use the marginal number, not the average. Removing a kWh of summer load saves $0.1304, not the average.

Actually billed, 3 years

$5,754

53,713 kWh across 35 bills, Jul 2023 – Jul 2026

Energy charges only

$4,013

what the export's COST column records — $1,741 short of the real total

Summer marginal rate

$0.1304/kWh

all-in, upper tier · winter is $0.0841

Summer share of spend

51%

from 11 of 35 bills

Why each bill differed from the same month a year earlier

Split into the part you caused by using more or less, and the part the rate schedule caused on its own.

Usage effect — you used more or lessRate effect — the price moved

Across 22 paired months, usage decisions account for $361 of movement and rate changes for $16. This is the one view the interval export cannot produce — it needs cost, and three years of it.

Table view

Period	Usage effect $	Rate effect $	Net $
2024-08	+16.15	+2.59	+18.74
2024-09	-8.12	-0.12	-8.24
2024-10	+9.02	+0.61	+9.63
2024-11	+12.55	+0.48	+13.03
2024-12	+3.18	+0.43	+3.61
2025-01	-26.58	+0.00	-26.58
2025-02	-10.12	+0.00	-10.12
2025-03	-14.37	+0.00	-14.36
2025-04	+21.64	-0.01	+21.63
2025-05	-11.52	-0.00	-11.52
2025-07	+11.27	+1.34	+12.61
2025-08	-20.10	-1.98	-22.08
2025-09	-44.36	-5.02	-49.38
2025-10	-11.40	-0.00	-11.40
2025-11	-13.90	+0.00	-13.90
2025-12	-3.55	-0.00	-3.55
2026-01	+38.50	-0.00	+38.50
2026-02	+24.66	-0.00	+24.66
2026-03	+24.49	-0.01	+24.48
2026-04	+5.52	+0.01	+5.53
2026-05	-4.07	+0.00	-4.07
2026-07	+25.61	+2.89	+28.50

Weather-normalized, three years of bills are flat

The bills reach back three years; the weather station covers one. NOAA's record for USW00093041 fills the gap, scaled to reproduce the backyard station's degree-days over the 353 days they overlap (cooling R² 0.82, heating R² 0.56). The scaling is forced through the origin, so no degree-days in still means none out.

Stripping the weather-driven part from each bill leaves the consumption that weather cannot explain:

Year	Baseline
2023	39.5 kWh/day	5 periods
2024	38.1 kWh/day	10 periods
2025	36.4 kWh/day	10 periods
2026	37.9 kWh/day	7 periods

The trend across all 32 periods is -0.04 kWh/day per year, with an R² of 0.000 — no trend at all. And it has to clear a noise floor. Periods the backyard station measured scatter 2.3 kWh/day about their own mean; proxy periods scatter 8.5. Where the two sources overlap, the proxy misplaces the weather term by only ±2.3 kWh/day — which in quadrature with the station's own scatter would give about 3.3, not 8.5. So the proxy's weather error does not account for the extra spread on its own: either it performs worse outside the single year where it could be calibrated, or 2023–24 genuinely varied more. Both readings say the same thing about how to use it — the multi-year average is solid and individual pre-2025 periods are not.

So your consumption is not drifting. Bills move because summers differ and because the summer tier is expensive, not because the house is quietly using more each year.

Every bill, three years

Consumption per billing period. Periods vary from 28 to 34 days.

kWh billed

The seasonal shape repeats almost exactly year to year, which is what makes the weather-independent baseline visible: the troughs never approach zero.

Table view

Month	kWh billed
Aug 23	2,195
Sep 23	2,170
Oct 23	1,648
Nov 23	1,186
Dec 23	1,210
Jan 24	1,355
Feb 24	1,024
Mar 24	1,026
Apr 24	1,028
May 24	1,720
Jun 24	1,811
Jul 24	1,980
Aug 24	2,360
Sep 24	2,087
Oct 24	1,804
Nov 24	1,403
Dec 24	1,265
Jan 25	898
Feb 25	850
Mar 25	779
Apr 25	1,400
May 25	1,522
Jul 25	2,096
Aug 25	2,157
Sep 25	1,633
Oct 25	1,608
Nov 25	1,164
Dec 25	1,204
Jan 26	1,560
Feb 26	1,274
Mar 26	1,200
Apr 26	1,495
May 26	1,452
Jun 26	1,791
Jul 26	2,358

What to check the next export against

One number here is worth carrying forward as an alarm — and one real failure that neither it nor anything else on this page would have caught. Both are worth writing down.

Show the threshold, and what it cannot see

Floor, month by month

0.74kW

median of 12 monthly medians — the same floor as above, aggregated for watching rather than for costing

Month-to-month wobble

±0.074kW

full range 0.24 kW across the year

Smallest new load it would catch

148W

1,297 kWh/yr — about $134 a year left running

Watch the floor: 0.74 kW, ±0.074

Most numbers on this page describe a year that has already happened. This one is for the next export. The always-on floor is equipment that never switches off, so unlike everything weather-driven it has no business moving at all — which makes its ordinary wobble a usable alarm threshold.

Over 12 months it held 0.74 kW with a month-to-month standard deviation of 0.074 kW and no trend (t = +0.17). Two standard deviations is 148 W. Anything new that runs continuously and draws more than that — a second refrigerator, a failing pump seal, a device that stopped sleeping — would clear the noise and show up as a step in this series.

Put in money, 148 W left running is $134 a year. So the floor catches any new always-on load costing more than roughly that, and cannot see one costing much less.

Where the alarm threshold comes from

P_min=2 · σ(F_m)

F_meach month's median always-on floor, kWσits month-to-month standard deviation, 0.074 kWP_minsmallest new continuous load that would clear the noise — 148 W

Two standard deviations rather than three: this is a prompt to go and look, not a claim that something is wrong.

The blind spot between the two alarms

There is a real failure in this house's recent history that neither the floor above nor anything else on this page would have caught. Shortly before this record starts, the air conditioner's outdoor fan motor was found drawing too much current and was replaced. A technician with a clamp meter found it; the whole-house meter could not have.

The arithmetic says why. That fan is rated 1.1 A, or about 228 W. A fan drawing double its rating therefore adds only another 228 W, and only while the compressor runs — which works out near 6% of the cooling slope. Comparing cooling efficiency across the two summers in this record — matched on how hot the days were, since an air conditioner is less efficient the hotter it gets — resolves a step of 30% (0.66 kWh per degree-day). The fault is five times smaller than the smallest thing the method can see.

So the two alarms bracket a gap. The always-on floor catches anything new that runs continuously above 148 W. Cooling efficiency catches degradation above 30%. A part that is small, intermittent and only draws when the weather says so falls between them. The pool pump was found here precisely because it is the opposite of that — large, long-running, and on a fixed schedule.

Worth recording that the naive version of this test fails badly. An unmatched changepoint scan over the cooling season reports a large, confident break at the start of June — and in the wrong direction, cooling apparently getting worse. It is not an event at all, but the curve of efficiency against outdoor temperature being fitted as a step. Matching on degree-days makes it vanish: across the 4–13 degree-day range the two seasons agree to within 0.10 kWh per degree-day.

A 14 gal/day overnight residue, recorded and not chased

Fitting the pool's make-up water against the weather leaves a constant behind: 14 gallons a day in the small hours that does not vary with evaporation. Splitting the record in half puts it between 10 and 14, so it is a range.

It is stable — the first half of the record and the second give the same figure, so nothing started during the year — and it is not the Zone 1 leak, which only flows on watering nights. Beyond that the meter cannot help: at one-gallon resolution a steady 0.6 gal/h and a couple of gallons every three hours are the same reading, and only 72 overnight hours in the year exceed ten gallons, so no appliance cycle stands out to be identified.

Which is why it is written down rather than investigated. A softener regenerating, an RO filter, an ice maker, a fill valve settling — any of them produces this, and the whole thing is $17 a year. The number is here so a future export has something to compare against: if it doubles, that is worth a look, and a single night with the pool's float valve shut would settle what it is. Neither is worth doing today.

What the labels say

7 nameplates, read off the machines and typed in by hand — the one source here with no file behind it. They settle one assumption, correct another outright, and give the meters something to be checked against.

Show the seven nameplates and what they settle

Central air conditioning

3.5 ton

Carrier 24ABB342A300 — 13 SEER, R-410A, cooling only
built Oct 2009, 17 years old

Central heating

88,000 BTU/h

Carrier 58DLA090-16 — 80.7% efficient, 71,000 BTU/h out
built Nov 2009, 17 years old

Water heater

50 gal

Rheem XG50T12HE40U0 — 40,000 BTU/h, natural draft
built Mar 2024, 2 years old
installed 28 Mar 2024

Pool heater

333,000 BTU/h

Sta-Rite SR333NA — natural gas, input rating
built Dec 2009, 17 years old

Pool filter

36 ft²

Pentair FNS Plus 36 — DE, rated 90 GPM for a private pool
built Jul 2022, 4 years old

Pool pump

1 HP

Pentair P6E6E-206L — Max-E-Pro EE, single speed
built Nov 2009, 17 years old

Patio mini-split

12,000 BTU/h

Premium Levella PIAW121800B — heat pump, both ways · R-410A · the only one on the property
date encoded in the serial, not decoded

How the house actually heats and cools

The condenser is a 24ABB342A300 — Carrier numbers cooling-only condensers 24… and heat pumps 25…, and its UL block reads central cooling air conditioner. So the split is clean: all central heat is the 88,000 BTU/h gas furnace, and all central cooling is the 3.5-ton air conditioner. The only heat pump on the property is the 12,000 BTU/h mini-split serving the patio.

That matters for reading the winter electric line. The electricity that tracks cold is not a heat pump warming the house — it is the furnace's own 373 W blower, the mini-split heating one room, and whatever else a dark cold evening turns on. It also explains why the two fuels could not be ranked against each other elsewhere on this page: they were never alternatives for the same space.

The furnace's efficiency is printed on it: 80.7%, within 0.7 of a point of the figure assumed for the pool heater

Every gas-to-heat conversion here runs through an assumed 80% appliance efficiency, and the caveats have said all year that a real appliance between 70% and 85% moves those figures by about a fifth either way. The furnace label prints both sides of that ratio: 88,000 BTU/h in, 71,000 out. Its efficiency is therefore 80.7% — read, not guessed.

The furnace's own conversions use that figure, and the mini-split model takes its blower wattage and input rating from the same label — 373 W and 88,000 BTU/h.

The pool heater and water heater labels are both here too, and neither prints an efficiency — only an input rating (333,000 and 40,000 BTU/h). So the assumed figure now has exactly one job left: converting the pool heater's gas into heat in the water. The water heater never needed it — its share of the meter is counted as cubic feet and priced directly, with no conversion in between.

That the one appliance with a printed efficiency was guessed correctly is the only independent reason to think 80% is fair for the pool heater as well. It is weak evidence, but it is more than the figure had before.

A nameplate that checks its own transcription (NEC 440.33)

MCA=1.25 · RLA + FLA

RLAcompressor rated-load amps, 17.9 AFLAcondenser fan full-load amps, 1.1 AMCAcomputes to 23.48 A against the 23.5 A printed

The largest motor is taken at 125%, the rest at nameplate. Reproducing the printed figure from the printed currents tests the typing, not the equipment — and the build refuses to run if it stops agreeing.

The air conditioner runs at 88% of what its label allows

The same trick that identified the pool pump works on the air conditioner: take the median load profile of the 20 hottest days of the year (mean high 106°F) and subtract the median profile of 60 mild ones. Everything that ignores the weather cancels — the pool pump runs the same 15:15 block on both kinds of day — and what is left is cooling.

It comes to 3.46 kW sustained through the afternoon, peaking at 15:00. The label allows 3.93 kW — compressor at its rated 17.9 A plus the condenser fan, at 230 V, at an assumed power factor of 0.90. That last figure is the only one here not printed on the label, and it matters: at unity the ceiling would be 4.37 kW and the ratio below would read 79% instead. Measured against rated is 88%, which is where a compressor should sit on the hottest days of its year: rated load amps is a ceiling, not an operating point.

Two further things fall out. On those same days the unit ran about 51% of the twenty-four hours — so at 106°F it is still cycling rather than running flat out, and 3.5 tons is enough for this house. And overnight the hot-day premium collapses to 0.16 kW: once the sun is down the house coasts, which is the same envelope the zone section measured from the other side.

As a check on the transcription rather than the equipment, the minimum circuit ampacity computes from the currents printed beside it — 1.25 × 17.9 + 1.1 = 23.5 A against the 23.5 A printed. The build refuses to run if that stops being true.

The pool heater's rating checks a number this page worked out, rather than one it assumed

Most of what a label settles here is an assumption. This one tests a derivation. The spa's volume was inferred from how fast the heater raised it, which needed the heater's burn rate — and that was reverse-engineered from the October ramp at 270,000 BTU/h, with no way to check it. The label reads 333,000 BTU/h.

The derived figure is 81% of the rated one, and that is the right shape of answer. A burner cannot beat its own input rating, so the test is one-sided: anything from 0 to 100% is consistent and only a figure above 100% would mean the ramp analysis was broken. What the remaining 19% represents is the part of the window the burner was not lit — purge and ignition at the start, and cycling as the water approaches setpoint.

The spa volume keeps the derived rate rather than the nameplate, and deliberately so: the ramp needs the average rate actually delivered across the window, and substituting the nameplate would assume continuous firing and inflate the spa by about 23%. The label's job here was to bound the derivation, and it does.

Note what the label does not give: an efficiency. Only the input rating is printed, so the pool heater stays on the assumed 80% along with the water heater.

The filter puts a ceiling on the pump's flow, and the pool conclusion survives it

Pump flow has been the other load-bearing guess on this page — 60 GPM, assumed, with turnover counts scaling inversely with it. The filter is a Pentair FNS Plus 36, 36 ft² of DE rated at 90 GPM for a private pool. That is a design ceiling, not a measurement — but a ceiling is exactly what an open-ended assumption was missing.

At the assumed 60 GPM the pump turns the water over 4.9 times a day. Push it all the way to the filter's rating and it becomes 7.3. Every value the plumbing admits leaves the pump running several times the one-to-two turnovers that is standard, so the recommendation to cut the runtime does not depend on the guess at all — only the precise saving does.

The pump carries two labels because it is two parts. The wet end is a Pentair P6E6E-206L, a 1 HP Max-E-Pro; the A.O. Smith SQ1102 whose nameplate predicted the measured 1.65 kW is the square-flange motor bolted to it. One pump, two labels — and the power prediction stands either way, because it was the motor that did the predicting.

Data & provenance

What each source can support, and where the numbers stop being measurements.

Show what each source can and cannot support

How much to trust each number

Weather — your WS-2000, 364 complete days at 5-minute resolution. Degree-days are integrated across every sample of the day — 288 on average, exactly 288 on 347 of 364 days — rather than derived from (max+min)/2, which matters near the balance point where the fits live. The station agrees closely with the weather UtilityHawk reports independently: daily highs r = 0.980 (bias +1.0°F), lows r = 0.988 (bias -0.7°F) over 364 days. The station's authority here is earned, not assumed.

Electricity — 364 days of 15-minute interval data, 359 of which overlap the weather station. Essentially complete: one day carries 92 intervals rather than 96, which is the spring-forward hour behaving correctly.

Water and gas — UtilityHawk supplied 366 daily readings; 364 of them fall inside the weather station's coverage and are the ones used here. Daily resolution is the binding constraint on leak detection. A leak has to lift the quietest day of the week before it registers at all, which sets that sensitivity at roughly 81 gallons a day — enough for a running toilet, nowhere near enough for a dripping tap, and blind to the one fault this house actually had. A split irrigation line leaks only while its valve is open, so it never lifts a quiet day at all. The hourly series answers a different question and answers it far better: it reads each irrigation cycle directly, and a cycle that changes size is visible within a week. See the leak and Water, by the hour.

The two electric exports agree — but they are not two measurements. Summing the 15-minute intervals over each of the 10 billing periods they fully cover gives 15,258 kWh against 15,244 kWh billed — a difference of +0.09%, with no single period off by more than 0.14%. The interval file reads consistently a hair high, which is what rounding to whole kWh on the bill would produce. Both files are the utility rendering the same smart meter register two ways, so this is a consistency check on the pipeline, not corroboration of the measurement — a miscalibrated meter would be miscalibrated in both. What it rules out is a misaligned billing boundary, intervals dropped from the export, a scaling error in parsing, or mishandled daylight saving. Those are the failure modes actually in play, and nothing in this dataset can test the meter itself.

What is genuinely independent is elsewhere: the backyard weather station against UtilityHawk's own weather source (different sensors, different operators); the pool pump's electrical signature against a thermometer in the water; and that pump's measured draw against its nameplate. Those are separate instruments agreeing, and they carry the weight this one cannot.

Money comes from 28 PDF bills, transcribed rate by rate. Nothing here divides a total by a usage figure to guess a price. Every tariff is typed from a bill and checked against it on each build — all 39 checks reproduce their source to within two cents, and the build aborts if any drifts. The worst disagreement across all of them is 2 cents, on Gas 2025-07, 1 Mcf @ $2.11/Dth. Listing all 39 passes would say nothing the sentence before it does not; the build failing is what carries the claim.

What 28 bills still cannot pin down. A complete year of both utilities — 14 EPE and 14 Las Cruces, one per meter read. Every rate below is now read off a bill for the month it applies to rather than carried forward from a neighbouring one, so these are the remaining edges:

Water has no second tier boundary anywhere in the record: 3,000 gallons free, then a flat $2.85 per 1,000 of commodity charge across every bill — $3.27 all-in once the franchise fee and tax that ride on it are added, which is the figure the cost section prices with — up to the largest ever billed at 9,000. Whether an escalating tier exists above that is unknown — no month has ever reached it, so no bill would have shown it.
The gas commodity rate is a monthly pass-through and is known for every month of the record, spanning $0.40 to $3.37/Dth. Only days between two meter reads are interpolated, and never across more than one month.
Wastewater is billed on a fixed allowance re-set annually, not on the month's own water use. A full year shows 4,000 gal holding through the February 2026 read and 2,000 from the March one. The date it re-sets is observed; the formula that chooses the number is not published.

Meter quantisation. The gas meter reads in whole Mcf and the bill charges whole dekatherms, so a month of 552 cf is billed as a full 1,000 cf. The water meter reads in whole 1,000-gallon units. UtilityHawk's daily figures are finer than either meter, which is why a month's daily sum need not land exactly on the billed volume — over the June–July 2026 period the daily water figures totalled 5,874 gal against 6,000 billed, a 2.1% difference entirely explained by rounding.

Caveats worth keeping in view. What follows is load-bearing for numbers on this page and is not measured:

Pool heater efficiency at 80%. Its label prints an input rating and no efficiency, and it is the only appliance still relying on the assumption — the pool and spa heating costs and the per-degree figures all run through it. A real heater between 70% and 85% moves those by roughly a fifth either way. The furnace, which prints both 88,000 BTU/h in and 71,000 out, comes to 80.7% — within 0.7 of a point of the assumed figure, and the only independent support the pool heater's number has.
Pump flow at 60 GPM. The turnover counts scale inversely with it: at 45 GPM today's runtime is 3.6 turnovers rather than 4.9. Assumed, but bounded: the filter's private-pool rating of 90 GPM caps it, and even at that ceiling the runtime is 7.3 turnovers. The conclusion that it runs well past one or two survives the whole admissible range; the exact figure does not.
That the March drain was complete. The refill itself is now measured rather than inferred — the hourly register puts 5,002 gallons into the system over 28 hours, against the 5,000 this page uses. But that is how much water went in, which equals the system volume only if the drain emptied it. That it did is the owner's account, not a measurement; a partial drain would mean a larger pool and a proportionally larger per-degree cost.
The patio's glazing — the coefficient, not the area. The glass is measured: 140 × 78 inches over both doors, 76 sq ft. The solar heat gain coefficient is assumed at 0.30, so the conduction half of the mini-split's cooling (291 kWh of electricity) is derived from measurement and the solar half (358 kWh of electricity) is half-measured — right area, assumed transmittance. Not to be confused with the similar-looking insolation figures in that section, which are kWh per square meter falling on the glass rather than kilowatt-hours drawn by the heat pump. One knock-on from the solar section belongs here too: this window is projected from the same station whose eastern sky is blocked by the roof ridge, so its mornings are under-read. A due-south window should collect near-symmetrically about noon and this one comes out 1.09 afternoon-to-morning, which puts the annual figure roughly 2% low. That is small next to the assumed transmittance above, and in the opposite direction to it, but it is a floor rather than a correction — nothing here has been adjusted for it.
Weather before Aug 2025 is a NOAA proxy. Not because the station was not there — it was, and it was recording — but because AmbientWeather only serves a rolling year, so the earlier record exists and cannot be fetched. The proxy carries ±2.3 kWh/day of error over the one year where it could be checked — and proxy-era periods scatter 8.5 kWh/day against the station era's 2.3, more than that error alone explains. Individual pre-2025 periods should not be read closely; only the multi-year average is solid.
The solar sensor is not used for solar. The station's pyranometer reads 26% low against NSRDB, and only 20% of that is the instrument — the rest is an eastern horizon standing about 21° up. Once the mount height is measured rather than estimated, the roof can account for at most 18° of that and possibly as little as 13°; the balance is most likely the trees due east, whose height is unmeasured, so this page does not claim to have divided them. That is invisible on a south plane and disqualifying on an east one, so plane-of-array here comes from PVWatts. It is the only modeled input on this page, and it brings typical-year weather with it: the monthly figures answer what an array would do in a normal year, not what it would have done in this one. Where the station is still used it is scaled by ×1.35, which a flat PVWatts plane independently confirms to about 2%.
The roof pitch is measured to about a degree. 22.5° ± 1° by phone inclinometer. The PVWatts runs were commissioned at 22.6°, inside that band, and are kept: plane-of-array moves about 4 kWh/m² per degree here, so the whole band is worth about 0.2%. Pitch matters four to five times more on an east-west roof than on a south one — the span of common pitches covers about 9% of production rather than 2% — which is why a degree of resolution was enough and a guess was not.
The two irrigation runtimes are the owner's account, not a measurement. 14 minutes in winter and 30 in summer, read off the Hunter X-Core. An hourly bucket records volume and never duration, so nothing here can check them directly — the corroboration is indirect and worth stating as such: dividing every clean month's cycle by the runtime it was set to yields 2.97–3.73 GPM across both programs, and a wrong pair of runtimes would have produced two different flow rates rather than one. That is evidence, not proof.
The hourly record does not reach back before July 2025. It covers 400 days, which is enough to hold one summer against the previous one — the comparison the leak rests on. It is not enough to say whether the 107-gallon summer cycle is itself normal for this system, only that it held for two seasons before it stopped. A second clean summer would settle that, and there is no way to fetch one retrospectively.
The condenser's power factor at 0.90. A nameplate prints amps, not watts, so turning 19.0 A at 230 V into the 3.93 kW ceiling needs one. It is typical for a single-phase hermetic compressor under load but is not on the label, and it is load-bearing twice over: the unit's 88% load factor would be 79% at unity, and the open-door section's headroom argument moves with it. The measured 3.46 kW does not — that is a meter reading.
The five hygrometers were never cross-calibrated. Nothing here establishes that they agree, so between-zone moisture is reported as regression slopes, which a constant offset cannot move, rather than as differences of means, which it can. The zone comparisons are built to survive an error of about a gram per kilogram in any one sensor; the garage and shed happen to agree to 0.04 g/kg once the house is accounted for, but that is a result rather than an assumption.

What is identified, and on what evidence. The 15:15 timer block is the pool pump, on three independent grounds — the 1.65 kW electrical step, the water temperature stepping at the same quarter-hour, and an A.O. Smith SQ1102 nameplate that predicts 1,641 W against 1,650 W measured. And all 10 gas anomalies are identified rather than merely flagged: 7 are pool heating on days with no heating demand, and the other 3 are evening spa soaks, pinned by the absence of sun at their peak. 7 appliance nameplates have since been transcribed. They put the furnace's efficiency at 80.7% from its printed input and output, and establish that the central system cools only, so central heat is the gas furnace and the patio mini-split is the property's only heat pump.

🏠Overview

What this is built from

💵What things cost

Where the money goes

⚡Where the electricity goes

Every interval of the year

The shape of a day, by how hot it was

Every interval of the year, sorted by size

🌞What solar would do here

An 8 kW array against the months it has to cover

What each array size returns

Every size, in full

💦The sprinkler saga

Irrigation versus everything else

💧Water, by the hour — and the one thing that is broken

The shape of a day, season by season

What the weather asks for, and what the meter delivered

When the controller actually fired

Every irrigation cycle in the record

🏊The pool explains (almost) everything else

The moment the pump starts, seen three ways

Three ways for a pool probe to read above 100°F

🌡️How the house answers the weather

Electricity against cooling demand

Gas against heating demand

What the gas meter does across a day

How much of the weather gets inside

What a day costs, against how warm it was

Rain and lightning through the year

Rainfall by month

📦Four boxes in the same weather

One day, four buildings — the 30 hottest days of the year

The shared wall, measured

Water in the air, by month

Where the monsoon gets in, and how long it stays

🚩What the weather cannot explain

Water, day by day

Gas, day by day

Where the gas goes

🧾The long view

Daily consumption and temperature

Monthly totals

Why each bill differed from the same month a year earlier

Every bill, three years

👀What to check the next export against

🏷️What the labels say

🗂️Data & provenance

Overview

What things cost

Where the electricity goes

What solar would do here

The sprinkler saga

Water, by the hour — and the one thing that is broken

The pool explains (almost) everything else

How the house answers the weather

Four boxes in the same weather

What the weather cannot explain

The long view

What to check the next export against

What the labels say

Data & provenance