My Own Air Conditioner Was Undercharged for More Than a Year. Here Is What It Actually Cost.

The air conditioner on my own house was running undercharged. It had been for more than a year, through at least one professional service visit, and nobody in the house knew — including me, and I build the instrument that watches for exactly this.

I want that on the page before anything else, because the useful part of this story is not that our monitor caught a fault. It is what happened after the fix, when I pulled every byte of data my house generates and found that my own first estimate of what the fault had cost was off by about seven times.

The house

Three thousand square feet, two storeys, six bedrooms and a pool, in Denton, Texas. Gas heat, which means the electric meter goes quiet in the winter — that turns out to matter later. One condenser feeding two zones through a zone panel: an ecobee upstairs, an ecobee downstairs, dampers routing the air. Vetralis unit B001 sits on the condenser outside.

That is an ordinary North Texas house. It is also, by accident, an unusually well-instrumented one. Four independent witnesses were already running when this started, and none of what follows required a test rig:

• The Vetralis monitor on the condenser — suction and liquid-line pressure and temperature, and from those, superheat, subcooling and compression ratio.
• Two ecobee thermostats, one per zone — compressor calls, indoor temperature, humidity and setpoints in five-minute buckets. About 53,000 rows across the summer.
• The City of Denton electric meter — whole-house kilowatt-hours by day. The only instrument in the house that measures dollars, and the only one watching everything.
• Weather — daily highs and hourly outdoor temperature, to separate “the system got worse” from “Texas got hotter.”

What the monitor saw

The refrigerant circuit gives the cleanest version of the diagnosis. Through late July the system ran with superheat pinned near 36 °F — the expansion valve wide open and still unable to feed the evaporator — while subcooling scraped along at 3–7 °F. That is the textbook undercharge signature, and it does not appear on a thermostat. The ecobees saw only zones that ran, and ran, and lost ground anyway.

On 30 July a technician charged the system in three steps between 3:51 and 4:19 in the afternoon, using the monitor as the gauge set, and an evening check closed on target at 6.2 °F subcooling. Daily-average superheat while pumping stepped from 35.7 to 26.2 °F. Suction pressure rose from 112 to 122 psi. The compression ratio eased from 2.83 to 2.56. Twenty-eight minutes, no second truck roll, and no guessing by gauge-feel in a 99-degree driveway.

Then I priced it, and got it wrong

I wrote the refrigerant side up first, working only from the monitor, and concluded the recharge was worth about 1.4 compressor-hours a day — call it $35 a month. Reasonable. Defensible. I would have said it out loud to a customer.

What the electric meter saw

The city's meter does not know my house has zones. It adds up everything — air conditioning, pool equipment, kitchen — and reports one number a day. June averaged 138 kWh a day. The failure window, 18–30 July, averaged 210, with a summer record of 287 on 23 July. The eighteen days since the recharge average 158, in equal or hotter weather.

Weather-corrected against healthy operation, the failure window carried 723 kWh of excess — about 56 kWh every single day. That is an entire second household's worth of consumption stacked on top of ours. At my blended rate of roughly 15¢/kWh, about $8.40 a day.

The recharge finished at 4:19 pm on 30 July. 31 July was the hottest day of the entire window at 103 °F, and posted 137 kWh — a below-June pace. The meter did not taper back toward normal over a week. It snapped, within hours, on the hottest day.

One accidental advantage of a gas-heated house: spring days when the compressor never turned a blade read the non-air-conditioning floor directly. Thirteen such days average about 90 kWh, most of it the pool. Measured against that floor, cooling this house costs about 68 kWh a day in healthy August weather — and cost about 120 kWh a day while the undercharge was biting, in the same heat.

One machine, two zones — the part I had wrong

Here is my actual mistake, and it is the reason the $35 figure was so far off. I had been treating my two thermostats as two air conditioners. They are not. There is one condenser, and two zones calling it through a zone panel. The data settles it: the monitor's daily runtime matches the union of the two zones' calls, not either zone on its own.

That one machine can serve a single zone comfortably, or both zones at once expensively. The undercharge did not change how many hours it ran by much. It changed which kind of hours it ran.

When July's heat outran what the diminished charge could deliver, the upstairs zone stopped being satisfiable — its calls stretched toward around the clock. So every downstairs call now landed on top of an already-running system. Fitting the meter against the call pattern across all 91 days: an hour of one-zone cooling costs about 3.8 kWh, and an hour with both zones calling costs about 9.3.

This is the cascade, and it has a nasty property: it hides the failure. Downstairs comfort never moved. The main floor held 75.0 °F through the entire event, because the system always gave that zone exactly what it asked for — it just needed double the hours to do it. Nobody standing on my first floor would have called a technician. The symptom lived upstairs, the expensive hours lived in the overlap between the zones, and the bill lived at the meter.

Runtime arithmetic prices the fault you can see. The meter prices the whole system's response to it — zones stacking up, the panel staging the compressor, hours migrating into peak heat, and people reaching for the thermostat.

Upstairs, at 84 degrees

Upstairs was a different house. The system was not dead — every night that zone clawed its way back to the 72-degree sleep setpoint by the small hours. It simply could not climb the afternoon hill any more. Day after day it fell as much as nine degrees behind, topping out at 84.1 °F, running pinned at 94 % duty through the noon-to-eight block.

My family did what every family does. They dropped the setpoints to 70 and 72 and fought the thermostat. In thermostat data that reads like preference rather than failure, which is exactly the problem.

I track this as comfort debt — degree-hours spent above setpoint. June ran about 18 °F·h a day, which is what an undersized upstairs zone does on 95-degree days in Texas. The failure window averaged 65, peaking at 126. Since the fix: 21. The house did not only get cheaper on 30 July. It got livable.

It had already been low for more than a year

I assumed a slow leak. The thermostat archive says otherwise. In August 2025 — five degrees cooler on average than this August — the downstairs zone was already calling 8.2 hours a day against 4.7 now, simultaneous hours already ran higher than they do today, and the upstairs zone was already losing its afternoons to 82–83 °F at outdoor temperatures the recharged system now shrugs off.

Note what that means about the start date. The archive does not show a system that was healthy and then went low — it was already low when the record begins. Whatever happened, happened before my data does. So the honest answer to “how long?” is: more than a year, and I cannot tell you how much more. July did not break it. July exposed it. Cooler summers would have collected less than this July's $250 a month — but they would have collected it every month, and never once tripped a complaint.

I do have a suspicion about how much more, and I want to be plain that a suspicion is all it is. This system went in in 2012. Somebody had hands on it again in 2022. Neither of those is a measurement of refrigerant charge — one is an install date, and the other is a service call I remember rather than a subcooling reading I can produce. It is entirely possible this equipment was never right: that it left the installers low, ran that way for a decade, passed through a checkpoint in 2022 without anyone putting a gauge on the liquid line, and only ran out of margin when a July got hot enough to expose it. The Department of Energy's field review makes that unremarkable rather than unlucky — charge faults are the ordinary condition of installed equipment, not the exception. But I cannot get there from a thermostat archive that begins in 2025. So it stays a suspicion, and the number in this post stays at more than a year.

“Recently serviced” and “correctly charged” are independent claims. One comes from an invoice. The other only ever comes from measurement. I know that now from my own driveway.

What I took back to work

My house is a one-unit portfolio. Our customers run thousands. Scaled up, these two weeks taught me three things worth repeating:

• Hours are not energy. While it was broken, my compressor ran barely two hours a day more than it does healthy — and the cooling bill nearly doubled. The cost lived in how it ran: zones stacked, the panel staging up, hours landing in peak heat. Fleet analytics that rank units by their own runtime would have ranked B001 as boring.
• A comfort complaint is a broken smoke alarm. Downstairs never suffered. Upstairs recovered every night. The household adapted by turning thermostats down. Waiting for a complaint means waiting for the cascade to saturate — and in a cooler year it never would have.
• The refrigerant circuit is where the diagnosis lives. The meter knew something was wrong. The thermostats knew where it was uncomfortable. Only superheat and subcooling said what it was and how to fix it — and then guided the fix, live, and verified it that evening.

What this is, and what it is not

This is one house, measured carefully. It is not a study, and I am not going to dress it up as one. The things I cannot prove from this data, stated plainly:

• The 723 kWh is measured. Its attribution to the undercharge rests on the timing — the meter normalized within hours of the recharge, on the window's hottest day — not on per-unit submetering.
• Dollar figures assume my roughly 15¢/kWh blended City of Denton rate. Your utility will differ.
• The weather correction uses daily cooling-degree-days. Humidity is not separately modelled.
• The 2.4× cost of simultaneous-zone hours is fitted from the meter, not submetered. The likely mechanism is the zone panel staging the condenser up when both zones call — the thermostats only ever see their own first-stage wire, which is why they cannot report it. A current clamp would price it exactly.
• Since the fix, our own monitor reads 2–3 hours a day above the thermostats while the system cycles more gently. That is a detection-threshold calibration item on our side, and I would rather write it down here than leave it out.
• How the charge got low — a slow leak, or an under-filled service — cannot be determined from this data. The archive shows the system already low at the point the record begins, which puts the fault at more than a year but gives no start date — it predates the data entirely. If it is a leak, the signature will creep back, and the monitor is watching for exactly that.

I did not plan to run this experiment on my own family, and I would not have picked a 107-degree July for it. But I have spent two years telling people that the refrigerant side is where the early warning lives, and it is a strange kind of privilege to have my own house argue the case better than I can. If you want the method — the regression, the excluded days, all of it — write to me and I will walk you through it on your own building.