Round-trip efficiency on a home battery, and when cheap overnight power still loses

The bid says the battery will buy cheap overnight power and run the house at 6 p.m. That only works if the hours on your tariff are far enough apart to cover the energy that disappears in the box.

Round-trip efficiency is that leak. You put energy in and you get less back out, because the missing kilowatt-hour left as heat in the inverter and the cells.

It does not decide whether the fridge stays cold in a storm. Start with battery vs standby generator if you are still picking a machine. Start with usable kWh vs the sticker if the argument is tank size versus how hard the tank can pour.

Two percentages on the same Tesla sheet

Tesla’s Powerwall 3 datasheet lists two efficiencies, and they are not the same job.

Solar to battery to home or grid: 89%. That is the round trip through the pack. Tesla footnotes it at 25°C, beginning of life, 3.3 kW charge and discharge, typical solar-shifting use.

Solar to home or grid: 97.5%. That is panels feeding the house without parking in the battery first. It is not round-trip. A bid that prints 97.5% as the storage number is reading the wrong line.

Tesla: Powerwall 3 specifications

EnergySage’s 2026 best-battery table still puts Powerwall 3 round-trip at 97.5%. That is Tesla’s solar-to-home line, not the path through the cells. If a quote copies that table, ask which test they mean.

EnergySage: best solar batteries for 2026

Enphase publishes both flavors on the IQ Battery 10C. AC round-trip is 90%. DC round-trip is 96%. You live on the AC number unless the installer can show a DC path you actually have.

Enphase: IQ Battery 10C

A cold garage, a ten-year-old pack, and a charge rate that is not the lab point all move you off the brochure. Tesla measured that 89% at 3.3 kW. A house that dumps 8 kW into the pack, or sips 500 W overnight, is not that test.

Tesla’s warranty language still talks about capacity at year 10, not efficiency at year 10. Budget the decade, not the datasheet afternoon.

If two datasheets both say “90%,” ask which conversion they counted. AC-coupled storage (the Enphase 10C) turns house AC into battery DC and back to AC, so you pay that hop twice. DC-coupled storage (Powerwall 3 with panels on its inverter) can skip a conversion when solar charges the pack, and Tesla still prints 89% for energy that went through the cells and came back out. The higher number on that sheet is the path that never used the battery.

The cheap hour has to beat the leak

Rate arbitrage means charging when the kilowatt-hour is cheap and discharging when it is expensive. Time-of-use is the utility plan that prices those hours differently.

The loss is not free. If you buy 10 kWh off-peak and the pack is 89% round-trip, you only avoid 8.9 kWh on-peak. The 1.1 kWh you paid for and never used has to come out of the spread.

A quick check: on-peak price has to be higher than off-peak price divided by the round-trip. At 89%, off-peak times about 1.12 is the floor. Below that floor, cycling the pack from the grid loses money even before you count the box.

Most of the country is not on a fat time-of-use plan. EIA’s Electric Power Monthly summary for July 2026 put the U.S. residential average at 18.31 cents per kWh. A flat 18-cent rate has no spread to harvest.

EIA: Electricity Monthly Update, July 2026

PG&E’s weekday peak, in the tariff

PG&E’s Electric Schedule E-TOU-D is a published bundled rate. Peak is 5 to 8 p.m. on non-holiday weekdays. Off-peak is everything else, including weekends. Effective March 1, 2026, summer peak is $0.47708 per kWh and summer off-peak is $0.34212. Winter peak is $0.38747 and winter off-peak is $0.34886.

PG&E: Electric Schedule E-TOU-D

Summer, 10 kWh into an 89% pack. You pay 10 × $0.34212 = $3.42 to charge. You get 8.9 kWh back and avoid 8.9 × $0.47708 = $4.25 on-peak. Gross, that cycle is about $0.83. It is not $1.35, which is what 10 kWh times the 13.5-cent spread would have been if the pack were 100%.

Winter on the same plan. You pay 10 × $0.34886 = $3.49. You avoid 8.9 × $0.38747 = $3.45. That cycle loses a few cents before the inverter even warms up.

The floor math matches. Summer off-peak divided by 0.89 is about 38.4 cents, and peak at 47.7 cents clears it. Winter off-peak divided by 0.89 is about 39.2 cents, and peak at 38.7 cents does not.

Your utility is not PG&E. Pull the tariff, not the sales slide. If winter spread is 4 cents and the pack is 89%, grid charging is a winter hobby.

A fatter EV plan still does not buy the box

Some California houses sit on PG&E’s EV2-A instead, the home-charging time-of-use plan. Peak is 4 to 9 p.m. every day. Off-peak is midnight to 3 p.m. Effective March 1, 2026, summer peak is $0.53809 per kWh and off-peak is $0.22558. Winter peak is $0.41099, with the same $0.22558 off-peak.

PG&E: Electric Schedule EV2

Summer, same 10 kWh into an 89% pack. You pay $2.26 to charge. You avoid 8.9 × $0.53809 = $4.79 on-peak. Gross, that cycle is about $2.53.

Winter. You still pay $2.26. You avoid 8.9 × $0.41099 = $3.66. Gross, about $1.40.

That is a real spread. It is also not a reason to treat the battery as a trading desk. Even if you could shift 10 kWh every day of the year on those two numbers, you are in the neighborhood of $650 before a cold garage, a rate change, and the days you are not home to use the peak. EnergySage’s 2026 storage page puts a typical install around $15,647 before incentives for about 13.5 kWh. Hundreds a year against that sticker is extra. It is not a three-year payoff.

EnergySage: solar battery cost

A 13.5 kWh pack cannot arbitrage all day anyway. Tesla lists 13.5 kWh usable and charge power of 5 kW on a single Powerwall 3, or 8 kW with expansion units. After Tesla’s 89% round trip you have on the order of 12 kWh at the wall on a good day. A five-hour peak window at 4 kW of house load is 20 kWh. One pack does not cover that. The sales model that “shifts your whole evening” is often two packs, or a short list of circuits.

Export is a different napkin. If the utility pays you almost nothing for noon solar, storing it at 89% still beats giving it away. If they pay retail, the battery is competing with a clean export credit, and the leak looks worse. Ask what you actually get per exported kWh before you treat self-consumption as automatic savings.

Three envelopes that all say savings

Backup only. One 13.5 kWh pack, storm mode, no time-of-use story. Round-trip barely matters here because you are buying hours of fridge, not a trading desk. Ask for usable kWh, continuous kW, and which circuits sit on the critical-load panel. EnergySage’s $15,647 typical is this product.

Solar self-consumption. The roof makes power at noon. The house wants it at 6 p.m. Export credit is tiny or zero, so you store it. 89% of your own solar is still better than dumping it for a few cents. Compare that to the export rate on the bill, not to a made-up peak.

Grid arbitrage, no solar. “We’ll charge at night and cover 4 to 9 p.m.” On a flat rate, there is nothing to cover. On E-TOU-D winter, the napkin above goes negative. On E-TOU-D summer, 83 cents a cycle on weekdays does not pay off a $15,647 box. If they still subtract a federal 30% storage credit, send them to what still pays in 2026, because the Residential Clean Energy Credit ended for expenditures after December 31, 2025.

IRS: Residential Clean Energy Credit

EnergySage says the same thing in plain language: the federal tax credit is no longer available for residential batteries. State and utility programs may still exist. Name them on the bid.

Do not treat any of those envelopes as guaranteed bill savings. A closed export program, a rate change, and a cold garage all move the napkin. EnergySage’s own backup-without-solar writeup puts a standalone battery in the peace-of-mind column first, and says your bills will tell you fairly quickly whether time-of-use cycling is paying.

EnergySage: home battery backup without solar

What I’d ask in the driveway

Which efficiency line is on the paper, the trip through the pack or the path that never used it? You want the footnote too: temperature, power, beginning of life. 89% and 97.5% are different tests.

Is that an AC number or a DC number? Enphase 90% AC is the number you live with on an AC-coupled retrofit. A DC-coupled Powerwall 3 still quotes 89% for energy that went through the pack.

What is the actual tariff they modeled? Plan name, peak window, summer and winter prices. “Time of use” is not a tariff.

Did they cycle in winter, or only June through September? If the annual number is a summer story, the year is unfinished.

Is the gross price clean of a 2026 federal storage credit? 25D is done for new 2026 work. If a state or utility battery rebate is in the math, it should have a name.

Is backup scoped separately from the bill story? Circuits, gateway, and hours at your actual loads. A savings model is not a well-pump plan.

If two bids show backup plus a tariff by name, and the third is cheaper because “the battery pays for itself in three years,” the third is unfinished. I’d get a second bid that will write the rate.

What to do with this week’s envelope

Read the efficiency line on the datasheet, not the marketing page, and put 89% or 90% next to the model number.

Pull your actual rate plan. If peak and off-peak are the same, grid arbitrage is not a reason to buy.

Divide off-peak by 0.89. If on-peak is not higher than that floor, do not cycle from the grid on that plan.

Price backup as backup. EnergySage’s $15,647 typical 13.5 kWh install is the outage product. Anything the tariff returns is extra.

If you have solar and a weak export rate, store your own noon power before you store the utility’s midnight power.

Hire the job you would still want with the savings line set to zero.