A contractor in Phoenix installs a 4-ton heat pump in a new 2,200-square-foot home, runs ACCA Manual J through Wrightsoft, and lands a system that matches the calculated cooling load within half a ton. He did the math correctly.
On August 3, it hits 119°F. His system runs at maximum capacity from 11 a.m. to 9 p.m., and the thermostat never reaches its setpoint. By 4 p.m., the living room is 82 degrees and climbing, even though nothing is broken, the refrigerant charge is correct, the ductwork is sealed, and the contractor followed every step of the process exactly as ACCA specifies it.
His weather file lied.
A Ghost Climate Inside Your Load Calculation
Every residential HVAC load calculation in the United States feeds from a single data source: ASHRAE's climatic design conditions, published in the Handbook—Fundamentals. Its 2025 edition, released in June, covers 12,424 weather stations worldwide using data compiled under Research Project RP-1923, and the numbers are comprehensive, peer-reviewed, and accepted as the global standard for building design.
They are also, by construction, a portrait of the past.
ASHRAE's Typical Meteorological Year files, which serve as the standard weather inputs for energy simulation software, compile 25 to 30 years of historical weather observations into a single representative year. TMY3, the version still embedded in most residential HVAC tools, draws from data spanning 1991 to 2005, which means that when a contractor in 2026 runs Manual J with TMY3 data, the calculation is anchored to weather patterns that include observations from the Clinton administration. That is not an oversight or a bug in the software. TMY files were designed to smooth out year-to-year volatility and produce stable, reproducible design conditions, and for decades that approach worked because climate variables drifted slowly enough that a 30-year average remained a serviceable proxy for the near future.
Around 2010, the atmosphere stopped cooperating.
Measured, Not Hypothetical
Lawrence Berkeley National Laboratory now reports that actual peak cooling loads in high-density markets exceed modeled projections by 20% to 35%, a figure that accounts for grid electrification load and thermal envelope saturation during sustained heat events, both factors that compound when outdoor temperatures exceed the design-day assumptions baked into the weather file.
Separate academic studies reinforce the pattern. Hong et al., analyzing building performance across 17 ASHRAE climate zones, found that TMY3 underestimates long-term annual energy consumption by 9%, a discrepancy that compounds silently over a system's operating life because the error falls entirely on the cooling side of the ledger, where demand is growing, rather than the heating side, where demand is declining. A 55-year weather dataset study by Cui et al. across ten major Chinese cities found TMY-based cooling load calculations diverged from actual conditions by up to 17.8% in the worst cases. Cross-climatic building simulations presented at IBPSA 2025 projected cooling peak load increases of up to 48.77% under SSP5-8.5 scenarios, with heating peak loads declining by as much as 39%.
Researchers at Tsinghua University tested this directly by generating TMY files from data windows of 1, 5, 10, 15, 20, 25, and 30 years across 20 cities, then validating each against the subsequent decade of actual weather. Shorter spans, around 10 years, consistently outperformed the standard 30-year baseline, which doesn't just lag behind reality but actively dilutes the signal with conditions that no longer represent the operating environment.
A 15-Year Lifespan in a 30-Year Weather File
Consider what happens inside a typical residential HVAC replacement cycle when you overlay it on the climate trajectory the weather files are trying to describe. A homeowner in Dallas installs a new system in 2026, with the contractor sizing it using current ASHRAE design conditions derived from 25+ years of data, and the system will operate until roughly 2041.
Dallas-Fort Worth's cooling degree hours have been increasing at a rate that produced a net increase of 2,774 degree-hours over the 1990-2020 period, according to an assessment of Texas climate data published in Energies, while summer peak electricity demand in the region has grown at approximately 1.6% per year, driven largely by cooling loads that outrun every projection the utilities had modeled. A system installed today will spend its final five operating years in a climate meaningfully hotter than the one its sizing calculation described.
A Nature Scientific Data study made this concrete for a single-family home in Los Angeles by simulating performance under both historical TMY3 conditions and future TMY projections for 2050. Under historical conditions, the cooling system kept up; under the 2050 projection, it could not meet loads during the hottest periods. Researchers tracked thermal comfort using Predicted Mean Vote analysis and found the home accumulating substantial warm-discomfort exceedance hours, periods where the indoor temperature exceeded occupant comfort thresholds because the mechanical system simply couldn't remove heat fast enough, not because it was broken but because it was correctly sized for a climate projection that was already outdated when the researchers ran the simulation.
Available Tools, Absent Workflow
ASHRAE has published explicit guidance on incorporating future weather files into building design, and commercial tools exist to do it. WeatherShift generates climate-adjusted TMY files using downscaled global climate model projections, Meteonorm 8.x includes climate change scenario generation, and the CORDEX framework produces region-specific future weather datasets at resolutions suitable for building simulation.
Not one of these tools plugs into a residential HVAC contractor's workflow.
ACCA Manual J, the standard residential load calculation procedure, neither requires nor accommodates climate-adjusted weather inputs, and the International Residential Code references ASHRAE for design conditions without mandating any forward projection. Wrightsoft, CoolCalc, and every other ACCA-approved calculator accepts current ASHRAE data and nothing else, which means there is no dropdown for "size this for the climate in 2035," no checkbox for "account for a warming trend," and no pathway through which the future weather files that already exist in the commercial building world can enter the residential sizing pipeline at all. ASHRAE itself acknowledges this gap indirectly: the Environment+Energy Leader analysis noted that "most procurement and design processes don't require or incentivize" climate-adjusted weather inputs, even though the data has been available for years, because the additional upfront cost of oversizing for future conditions, typically $1,000 to $2,000 on a $10,000-$15,000 residential system, is difficult to justify against a lowest-initial-cost standard that governs 95% of residential HVAC installations.
Arithmetic Nobody Runs
Here is the calculation that would change the decision, if anyone performed it.
A correctly sized 4-ton system for a new home in a Sun Belt market costs roughly $12,000 installed, while upsizing to a 5-ton system at installation adds approximately $1,500 to $2,500 depending on equipment and duct modifications, buying capacity headroom for a climate that is measurably warmer than the one the TMY file describes.
Without that margin, the homeowner discovers in year 8 or 10 that the system can't hold setpoint on the 15 worst days of the year, and a mid-cycle replacement, factoring in equipment, labor, refrigerant line modifications, and potential ductwork changes, runs $13,000 to $18,000, at which point the homeowner has paid for 8 years of a system that now has zero residual value for the remaining 7 years of its design life.
| Scenario | Cost | Comfortable on peak days? |
|---|---|---|
| System sized to current TMY data | $12,000 | Years 1-7: yes. Years 8-15: increasingly no. |
| System upsized 15% at installation | $13,500-$14,500 | Years 1-15: yes. |
| Replace undersized system in year 8 | $12,000 + $15,000 | Years 1-7: no. Years 8-15: yes. |
Nobody publishes this table for residential buyers, because the contractor's Manual J printout says the load is 48,000 BTU/h, the system delivers 48,000 BTU/h, and the paperwork matches, which makes it look like a correctly sized system rather than one that was designed using data that describes a climate fifteen years in the past.
What AI Tools Could Do (and Don't, Yet)
AI-driven climate modeling has advanced faster than the residential HVAC industry has been willing to consume it, which is another way of saying that machine learning models trained on regional climate projections can now generate location-specific future weather files accounting for warming trends, changing humidity ratios, and shifting diurnal temperature swings, all variables that directly affect cooling load, but none of that capability reaches the truck-mounted laptop where a residential contractor runs Manual J. A domain-specific large language model developed at Michigan State University, trained on physics-based energy simulations of 536,416 U.S. residential building prototypes, can recommend optimal retrofit strategies from plain-language building descriptions with a 98.9% top-3 accuracy rate for CO₂ reduction and 93.3% for shortest payback period, which demonstrates that the computational infrastructure for intelligent, climate-aware residential energy advice exists and works.
It just doesn't connect to anything a homeowner or contractor actually touches when making sizing decisions. In 2016, the gap between what climate scientists know and what residential HVAC contractors use was already wide; in 2026, it is wider, because the science has accelerated while the industry's data pipeline has not moved at all.
What You Should Do About It
If you are building a new home or replacing an HVAC system in any market that has experienced record high temperatures in the past five years, which in 2026 means nearly every market in the country, start with this.
Ask your contractor which ASHRAE design temperature he used for your load calculation, and if the number comes from TMY3 data, which it almost certainly does, look up your city's actual peak temperatures over the past three summers. When the actual peaks exceed the design temperature by more than 3°F, your system is undersized for the climate you already live in, to say nothing of the one arriving in 2035.
A 15% capacity margin on a cooling system adds $1,500-$2,500 to installed cost, and that margin doesn't produce the efficiency penalties of traditional oversizing, the short-cycling and humidity control problems that Manual J is designed to prevent, if the system uses variable-speed compressor technology, which is now standard on most mid-tier and higher residential heat pumps. Variable-speed systems modulate output continuously, which means additional capacity doesn't force the system to run wastefully during mild conditions; it simply means the system has headroom when temperatures hit numbers that your weather file says shouldn't happen.
For existing homes with systems more than 10 years old in Sun Belt or transitional climate markets: pull ASHRAE's 99.6% cooling design dry-bulb temperature for your location and compare it against NOAA's actual observed high temperatures for the past three years. If observed highs consistently exceed the ASHRAE design value, your system is operating beyond its rated capacity during peak events, and you should plan for replacement sooner rather than later, sized to the climate you're measuring rather than the one the data file remembers.
What We Don't Know
LBNL's 20-35% gap figure comes primarily from commercial and high-density building data, and residential-specific measurement is thinner. Single-family homes have different envelope characteristics, occupancy patterns, and internal load profiles than office buildings, so the gap magnitude may differ, though the direction will not. Climate projections carry inherent uncertainty; SSP scenarios range from modest warming to extreme trajectories, and which pathway materializes determines how soon current weather files become functionally useless for equipment sizing.
What we do know, with confidence, is that the direction is unambiguous and has been for at least a decade. Every successive ASHRAE data update has shown higher cooling design temperatures and lower heating design conditions, which amounts to a technical admission that the data is chasing a warming signal it cannot catch because its averaging window is structurally too long. Tsinghua's recommendation to shorten the TMY generation window from 30 years to 10 represents the minimum viable response to a problem that has existed, quantifiably, since before most homeowners installed the systems they're currently running. Whether the industry adopts it before an entire generation of residential HVAC systems ages into premature undersizing is a question that the industry has not yet answered, or, based on the current trajectory of Manual J software updates, seriously asked.