Based on 5 years and ~2,000 HV battery repairs of EV Clinic field data.
We will say it out loud: State of Health is a lie.
Not because the number is fake. Because the name is fake. SOH is not “State of Health”. It is State of Capacity. It tells you one thing only: how many kWh are left compared to day one. That’s it. It tells you nothing about whether the battery is healthy, nothing about when it will fail, and nothing about how much usable life is left in the pack you are about to pay 30,000€ for.
Capacity is not health. Degradation is not a fault. And a single percentage on a screen is not a diagnosis.
Three years of looking for a correlation that doesn’t exist
For the past 3 years we tried to find the connection between SOH and battery failure across our repair database. The conclusion is simple: there is none.
Our statistics from ~2,000 HV battery repairs show:
- The majority of EV battery failures happen between 5% and 15% capacity degradation, exactly where SOH says “this battery is excellent”
- Once degradation passes 50%, the system declares the battery faulty on its own: internal resistance rises and an industry standard kicks in where the system limits itself and reports a fault, even though the cells are still perfectly capable of driving (Renault Fluence, Kangoo ZE, Smart, Twingo, eGolf and similar platforms from our records)
- A battery above 50% remaining capacity is a usable, functional battery, yet packs die long before that point, and packs survive long after it
In not a single case could we take SOH as a parameter of current health. Zero correlation with failure.
70% SOH, 600,000 km, and the battery was healthy
A Model S 85 taxi, DC fast charged every single day. SOH dropped for 30%. The owner panicked. Most workshops would say: wreck it, replace the pack.
We opened it and measured properly:
- Cell balancing: excellent
- Voltage delta between modules: minimal
- Internal resistance: homogeneous across the entire pack
- Pressure test: no leakage
The battery was functionally healthy. Degraded in capacity, yes. Failed, no. This taxi from Vienna arrived with a BMS_u029 fault, we repaired it, and the car is still driving on the same pack today. Why? Because every other parameter was moderate: low average speed, low load, low consumption. The pack lived an easy life, and the parameters that actually matter proved it, while SOH screamed “replace me”.
Now the opposite case. LG packs (Model 3/Y) failing completely and irreparably at only 8% degradation. Across six of our workshops, we have not managed to repair a single one. Internal resistance imbalance triggers a chain reaction. Capacity still looks fine. The pack is dead.
Same metric, opposite outcomes. That is not a health indicator. That is noise.
The LG vs Panasonic paradox: better SOH, worse battery
Here is the case that destroys SOH logic completely.
Panasonic cells in Model 3/Y often drop to around 80% capacity by 100,000 to 200,000 km. Relatively fast degradation, ugly SOH number. LG cells in the same models hold capacity noticeably better, with degradation comparable to a Model S.
By SOH logic, LG is the better battery. Reality says the opposite. The LG packs are the ones that fail completely and irreparably, even at only 8% degradation where cell internal resistance jumps to 120mOhm, while the “degraded” Panasonic packs keep driving. The buyer comparing two listings would pick the LG car because of the higher SOH, and pick the worse battery.
And one more detail nobody tells you: there are two different Panasonic cells in European Model 3 Performance vehicles, Nevada production and China production, with a noticeable difference in characteristics. Same model, same SOH display, different battery.
A metric where the better number points to the worse product is not a metric. It is a trap.
EOLHC: the metric we actually needed
Because the abbreviation “SOH” is permanently compromised, we will not even try to fix it. We formulated a new parameter as part of our internal research, and it is the core of our ongoing study:
END OF LIFE HEALTH CYCLE (EOLHC)
EOLHC is expressed in kilometers remaining until battery repair is no longer technically or economically viable. After that point, the HV battery is only usable for solar storage and other powerbank second-life applications.
SOH is misleading twice. First because it measures capacity and calls it health. Second because it only looks backwards: how much capacity was lost. It tells you about the past of the battery and nothing about its future.
EOLHC looks forward: how much repairable life is left. And that makes it the single most important parameter of trust in the product you are buying. When you spend 30,000€ on a used EV, you are not buying its history. You are buying its remaining future, and EOLHC is the only metric that measures it.
Real EOLHC values from our field data:
| Vehicle / Pack | EOLHC (km) | Capacity loss at EOL |
|---|---|---|
| Kia Soul EV | 80,000 | varies, often only 10-25% |
| Nissan Leaf | 120,000 | varies |
| Tesla Model 3 LG pack | 200,000 | as low as 8-10% |
| Tesla Model Y LG pack | 160,000 | as low as 8-10% |
| eGolf 2015 | 280,000 | ~30% |
| Tesla P85D (resealed) | 350,000 | only 14% |
| Renault Twingo 2022 | 390,000 | – |
| Tesla Model S 85 (resealed, serviced) | 640,000 | up to 70% and still functional |
Look at the P85D line carefully. 14% capacity loss, SOH says 86%, looks like a great buy. The pack is unrepairable and at end of life. Meanwhile a Model S 85 with 70% capacity loss is still driving a taxi. The “lying SOH” tells you exactly the wrong story in both cases.
[VISUAL 2: The EOLHC life scale. Horizontal timeline:
EV BAT LIFE START —–+————+—————-+—– EV BAT LIFE STOP (Model S 90D example, 550,000 km)
Marker 1 (first +): Contactor replacement, routine wear item. Above: “OLD SOH: 93%”. Below: “EOLHC: 61%” Marker 2 (second +): Broken voltage sense wire, repairable. Above: “OLD SOH: 90%”. Below: “EOLHC: 34%” Marker 3 (third +): Single cell failure. Above: “OLD SOH: 88%”. Below: “EOLHC: 9%, ACir degradation 42%”
Side panel with final pack parameters at LIFE STOP: ACir degradation: 42% CTCQ: Cylindrical (Repairability 90%, Quality 98%) Dsoc: 1.5% Dcac: 4.9% HUM: 2% (resealed, no leak) BTdelta: 1% AVGWS: +4.5%]
The parameters behind EOLHC
A serious health assessment requires at least 10 parameters. For a Model S we can pull up to 19,000 datapoints into our AI lifetime prediction model. The core parameters that define EOLHC:
CTCQ (Cell Type and Cell Quality) The single biggest factor. Cell format (pouch, prismatic, cylindrical), manufacturer (LG, Panasonic, Samsung, CATL) and the actual production quality of that cell generation. The worst offenders are pouch cells, which sometimes show “100% SOH” forever because manufacturers hide degradation behind a software buffer. We consider pouch cells a total failure for automotive use: 98% of all cells we send to recycling are pouch.
Dsoc (Delta SOC) Constant growth of charge-state deviation between cell groups. A rising Dsoc trend is an early failure signal that SOH will never show you.
Dcac (Delta Capacity) Capacity delta between cell groups, tracked over time. The trend matters, not the snapshot.
HUM (Battery Humidity Percentage) Humidity is always present inside a pack, but deviation means moisture ingress: bad seal or damaged breathing valves. Once a pack breathes moisture, cell shoulder isolation caps corrode through galvanic action, electrolyte leaks, and internal cell structure is damaged across all modules. From that point degradation accelerates permanently. This is why resealing is critical and why we have written about it before.
BTdelta (Temperature Delta) How much temperature deviated between modules over the pack’s lifetime. One single overheat event permanently damages cells and triggers degradation, sometimes only in one module, but that one module defines the EOLHC of the entire pack.
AVGWS (Average Watt Stress) If your vehicle is rated at 189 Wh/km and your real average is 220 Wh/km, you are running 15% above design load, charging and discharging alike, and your EOLHC shortens by roughly the same factor. Someone racing a car for 100,000 km on a pack with 200,000 km EOLHC will hit an unrepairable internal resistance state long before any capacity warning appears.
ACir (AC Internal Resistance) The deviation of internal resistance per cell, and the deviation versus the new-cell baseline. This is the closest and most accurate single value of true cell health that exists. Bonus: measuring ACir across the complete pack also reveals the quality of every busbar and connection, because deviation means a bad joint or a bad cell. From a safety standpoint this is one of the most important parameters in existence, and manufacturers never implemented it. It still requires removing the pack and measuring by hand.
The data and formula nobody will give you
Three parameters would make EOLHC prediction almost exact, and every manufacturer has them:
- Maximum charge/discharge cycle count from aging tests. This is the most relevant datapoint of all, more relevant than anything else in this article. Every manufacturer determines it through aging simulation, available even at prototype stage. With it, EOLHC would give you the literal exact mileage at which the battery reaches end of life and must be replaced. Battery replacement would become a predictable, planned cost instead of a financial surprise. That is exactly why you will never get it.
- Simulated EOL mileage from aging simulators. Derived from the above, never published.
- Cell Failure Parts Per Million (PPM). The holy grail. The single most honest number about cell reliability that exists. Every cell manufacturer tracks it. Nobody outside will ever see it.
Here is how powerful the cycle count is. The first generation Tesla Model S 85 has roughly 2,000 allowed cycles. Multiply by a realistic average range of 320 km per cycle and you get an EOLHC of 640,000 km. And that is precisely what we confirmed in the field, in cases where the wire sense bonding was not destroyed by moisture and the pack was resealed at least twice along the way. Wire bonding corrosion is that pack’s only Achilles heel. The cells themselves were among the highest quality ever produced, and they still are, even today.
One number from a manufacturer’s drawer, one multiplication, and a used EV buyer would know exactly what they are buying. The industry knows exactly how long their batteries last and exactly how often their cells fail. They simply decided you don’t need to know.
When the BMS simply lies
Everything above assumes SOH is at least an honest capacity number. Sometimes it isn’t even that.
Korean manufacturers display 95% SOH on certain models while a real discharge test, driving from 100% to empty and measuring actual energy, shows around 80% real capacity. That is not measurement tolerance. That is a systemic BMS lie, and dealerships use exactly that number in warranty decisions. Without an independent discharge test, every warranty claim gets rejected because “SOH is 95%”.
The textbook case is a first generation Korean EV with pouch cells under the front seat, where temperature delta between cells kills the pack. When one row goes, everything goes. The result is vehicles with 40 to 50% real capacity that the manufacturer refuses to repair, while the dashboard still shows a healthy battery. In Norway these cars sell for 3,000€, and now you know why.
And it gets worse. These batteries were never globally recalled, even though the pouch cells swell, tear the housing open and potentially endanger the passengers sitting directly above them. This is exactly the type of technical defect for which the industry’s own guidelines mandate a recall. The recall was carried out only in Norway and the USA. The rest of the world got nothing. Same battery, same defect, same risk, but apparently your safety depends on your postal code.
So the situation is worse than “SOH is the wrong parameter”. The parameter itself can be manufactured. And this is the data foundation the Digital Battery Passport intends to build on.
The Digital Battery Passport: bureaucracy disguised as transparency
The Digital Battery Passport sounds like a step toward sustainability. In reality it is on track to become another bureaucratic document that gives buyers a feeling of safety with no technical substance behind it.
The entire system is built on parameters like SOH, which, as shown above, is just remaining capacity. A battery at 90% SOH can fail tomorrow. A battery at 70% SOH can drive hundreds of thousands of kilometers. We analyzed hundreds of battery systems from different manufacturers and found no reliable correlation between SOH and actual failure probability. The critical failure drivers are moisture, corrosion, cell internal resistance, thermal stress, cell quality, pack architecture and future repairability. The Battery Passport shows the buyer none of them.
A used EV buyer does not want to know the CO2 footprint of a battery produced ten years ago. The buyer wants to know one thing:
“How long, and to how many kilometers, will this battery remain technically flawless and economically repairable?”
The Battery Passport does not answer that question. Here is what else it doesn’t tell you:
- It doesn’t tell you that pouch cell vehicles are practically unrepairable and are the most common reason for total vehicle write-off
- It doesn’t tell you that with cell-to-pack and cell-to-body architectures, one failed cell condemns 400-17,000 healthy cells to recycling, and that in case of failure you will personally pay around 2,000€ for a scrapyard to even accept the car for recycling
- It doesn’t tell you whether basic consumable battery components are purchasable at all. Most manufacturers refuse to sell parts. We have seen cases where an externally mounted coolant temperature sensor forces a 16,000€ battery replacement
- It doesn’t tell you the warranty rejection rate for that manufacturer and model, so you can’t know whether to send the car straight to scrap before the warranty even expires because you didn’t read the small print, or after warranty expiry at the first minor fault
- It doesn’t tell you the price of a new battery out of warranty
The Passport was written by people who come from paper, not from practice. It doesn’t protect anyone. It exists to strengthen authority and shift blame. In its current form it doesn’t solve the problem, it makes it worse, by training the market to judge batteries by one marketing-friendly number while ignoring every real indicator of long-term viability. That is not circular economy. That is the opposite: it actively obstructs informed decisions and increases risk for used EV buyers.
If Europe genuinely wants sustainability, the focus must be repairability, parts availability, module replaceability, and remaining repairable life. Everything else is administration that looks good on paper and solves nothing on the road.
Design is destiny
After ~2,000 repairs, the verdict on architecture is final:
- Sustainable: Cell-to-module with cylindrical or prismatic cells (Tesla Model S/X/3, VW ID.3). In every single case so far we were able to swap a defective module or cell for a matched used one, laser-weld it, and return the vehicle to the road without replacing the entire battery
- Unsustainable: Pouch cells in any configuration (Kia, Nissan Leaf, Porsche Taycan, Renault). Impossible to repair, no replacement supply, the number one reason vehicles get written off
- Worst in class: Cell-to-pack and cell-to-body (structural Model Y, Cybertruck, BYD Blade). The manufacturer saved roughly 1,000€ in production cost. The customer inherits a battery where a single defective cell means the entire pack, and often the entire car, goes to recycling
The Battery Passport will not tell you any of this either.
Field notes: not all original packs are equal
A few concrete EOLHC datapoints from our workshops that no SOH display and no Passport will ever show you:
Model S 85 originals (2013-2016) are the champions. Taxi vehicles with 600,000 km on the original pack are not a myth, we service them. The problem: Tesla often replaced these packs under warranty with remanufactured packs that are worse than the originals, due to a mistake in fuse bonding during the remanufacturing process. Depending on stock, a warranty replacement gets you either a remanufactured 85 kWh pack with old degraded cells, or a new 90 kWh pack software-locked to 85, which is actually an upgrade. Tesla doesn’t tell you which one you will get.
Quick trick: if your Model S pack housing is black, it is one of the better originals. Do not replace it under warranty, because you will most likely get a worse one back.
Model S 75D: across six EV Clinic workshops, not a single original 75D pack failure. Zero.
Model S 100D: caution. Poor fuse soldering, problems typically appearing around 250,000 km.
This is what real battery intelligence looks like: pack generation, production batch, cell origin, known design defects. Not one percentage on a screen.
What this means for you
- Forget SOH as a health metric. It is a starting point, never an answer
- Demand a full diagnostic extract (Tesla Toolbox or equivalent), not a percentage
- Check voltage delta between modules: must be minimal
- Check internal resistance homogeneity across the pack
- Pressure test the housing: any leak is immediate disqualification
- Identify cell type, manufacturer and pack architecture before you buy. Avoid pouch. Avoid cell-to-pack
- For a Model S: check the pack color and whether it is original or remanufactured
- Unexplained high consumption (300+ Wh/km) is a red flag for hidden pack problems
The EV industry still sells cars as if batteries are a black box. They are not. The tools exist, the data exists, the experience exists. What’s missing is transparency and a standardized, consumer-grade assessment protocol.
Until then: trust the percentage on the screen less. Trust the people who have actually opened the pack.
EOLHC is part of our ongoing research elaborate, and we will publish deeper methodology and datasets as the study progresses.
EV Clinic operates Europe’s first independent EV and PHEV repair network, with ~2,000 HV battery repairs across 6 workshops. Component-level repair, not unit replacement.State of Health Is a Lie: Introducing EOLHC, the Metric the Industry Never Gave You
Based on 5 years and ~2,000 HV battery repairs of EV Clinic field data.
From theory to tool: ToolboxPro and EOLHC certificates
We are not waiting for the industry to fix this. We built the tool ourselves.
As a starting point we developed our own ToolboxPro diagnostic tool, working over LAN and CAN bus directly with the vehicle. In the first series we are producing detailed EOLHC certificates for Tesla Model S, 3, X and Y, for two reasons: to speed up our own troubleshooting, and to give buyers and owners trust in vehicle parameters that can predictably tell you what the battery’s future actually is, not what its past was.
The tool reads what the dashboard never shows: per-brick CAC values across all 96 bricks, brick-to-brick spread, pack and BMB statistics, serial and part-level battery identity, and full raw histogram data straight from the HV controller backup. This is the raw material EOLHC is built from.
ToolboxPro is already available in demo, and we have just completed Model 3 and Y extraction of Dsoc, Dcac and the other key parameters described in this article.
EOLHC is part of our ongoing research elaborate, and we will publish deeper methodology and datasets as the study progresses.
More in video:
You say that pouch cells are unsustainable and used by Renault, yet then you show Twingo having EOLHC of 390k km. AFAIK, Twingo has a very small battery and range. How come that it has so big EOLHC then?
Also, do you have data for newer Renault EVs (Megane, Scenic, 5)?
Pretty stuppind conlucusion. Smart is not 2500kg vehicle, it is 1000kg vehicle. Factor to take into account.
Same for pouch and cylndircar, what is mentioned, WATT load and quality of the cells takes into account.
Renualt EOL is usually 200000km.
Same way you didnt see conlusion that LG cylindrical cell for Model 3 are bad, but i stated that cylindrical are most sustainabale.
So please read 5-6 times what we wrote.
All parameters are live depending on may factors, and SOH is not one of them.
This EOL Health Cycle is a fantastic initiative: image in this became the standard for assessing EV Batteries life expectancy and the reassurance it would provide customers considering purchasing a Used EV (and there are many more Used EV purchases than New EV purchases). This could have all sorts on interesting consequences: for example, imagine if this initiative was incorporated into the assessment of economicial / non-economical body repairs following an accident with an EV?
Solid field work, and I don’t dispute the engineering points (capacity ≠ failure mode, architecture drives repairability). One genuine question about the statistics: did the analysis account for the base rate of the fleet, or is it drawn purely from the repair database?
The reason I ask: as written, the dataset is batteries that already failed and came to you, there’s no “did not fail” group in the sample. With only the numerator and no denominator, Figure 1 shows P(degradation | failure), not P(failure | degradation). Those two are equal only if every SOH band has the same number of cars on the road, which isn’t the case: a young EV fleet spends most of its life at 5–15% degradation, so most failures will land there even if SOH is a strong predictor. The chart can’t distinguish “SOH doesn’t matter” from “SOH matters, but most cars are simply healthy.”
For the same reason, “zero correlation” isn’t derivable from this data, correlation needs variation in the outcome, and there are no non-failure cases to correlate against. The same applies to model-level claims like “LG packs die at 8%”: without how many LG packs are running in total, that’s a count, not a rate.
So: were these cross-referenced against fleet/registration totals per SOH band and cell type? If not, the failure-share conclusions may be a selection-bias artifact rather than evidence against SOH, even though the underlying point that SOH alone is insufficient might still stand, I don’t deny that.
Fair question, and you formulated the limitation correctly: our dataset is a repair population, so Figure 1 shows the degradation distribution of failures, not failure probability per SOH band. We don’t dispute that.
Here is the problem with the study you’re describing: the denominator does not exist outside the manufacturers. Fleet SOH distributions, warranty failure databases, aging-test cycle counts and cell failure PPM are all proprietary and locked. No independent organization on this planet can compute P(failure | SOH), and that is one of the central points of the article. The data needed to validate or refute SOH as a predictor is exactly the data the industry refuses to release.
One more thing about our sample. We are the destination packs arrive at before recycling, regardless of brand, architecture or mileage. When a battery fails anywhere in our region and the dealer answer is “replace the pack”, that pack ends up on our bench. So while our database is by definition a failure population, it is the most complete failure population that exists outside OEM warranty systems, and that is precisely why we analyze it: the source of this information physically comes to us, and to nobody else.
There is a second layer. Even the numerator side is only measurable because we physically open packs and measure by hand. No existing consumer or workshop tool reads per-cell internal resistance deviation, humidity ingress or wire bond condition. So the “did not fail” control group you ask for is not only unpublished, it is currently unmeasurable at scale. Building exactly that reference database, pack by pack, model by model, is what our EOLHC certificate program exists for.
What we can defensibly say from ~2,000 repairs: every failure mode that actually kills packs (internal resistance imbalance, moisture ingress, bond corrosion, thermal damage) is routinely observed at 5-15% degradation, and packs at 50-70% degradation routinely pass every functional test. So in no individual case can SOH clear a pack or condemn it. That is the operational meaning of “no correlation” in the article: SOH is not usable as a diagnosis on a specific vehicle, which is the only question a buyer or a workshop ever faces. If a manufacturer opens its telemetry tomorrow, we will gladly run the fleet-level study together.
Thanks for the fair and honest answer! Kudos!
SUV’s of up to 3 000Kg, with up to 800kg cell to body batteries, are currently being marketed (green-washed) as environmentally friendly.
Sustainability and the environment are not the concern of corporate greed. The more companies move from mechanics to electronics, the faster they can produces goods, while reducing reparability to stimulate consumption and improve profits.
In 2050 when we are up to our eyeballs in defective EV’s, and there are only 5 brands left (VAG-Stellantis, BMW-Mercedes, Tesla-otherUSbrands, and 2 Chinese groups), some smarty-pants in the European Commission will come up with the term ‘EV-gate’. That is, if individual car ownership is still a thing by then (“Do plebeians really need their own private transportation?”.
I’m very much for electromobility, so it really pains me to see how the politicians fuck things up as usual. So what is it (again) this time: Ideological blindness, corruption, or (probably) both?
The day when we’ll have commercially viable nuclear fusion reactors (2050 also?), the term ‘solar panel gate’, ‘windmill gate’ can probably be added to the *gate list.
Hello, congratulations on this article or the findings behind it! An important step so that interested people understand what to expect when they buy an electric car of the brand yxz! I am amazed that the legislator allows non-repairable batteries to be built and installed! But well, Apple Lightning was banned for this and the new smartphone no longer includes power supplies …. 😅
I understand you repaired a lot of Teslas.
What about the other brands that you didn’t mentioned? Peugeot, Mercedes, … Do they have no or less problems? One can think that these are better cars with better batteries, since not mentioned.
Zanima me da li preporuka za VW ID3 i ID4 vrijedi samo ako su prismatic celije?
Koje to godiste i modele obuhvata preporuka?
Konkretno sam vidio da pisu da su prve generacije bile pouch, a vidio sam vw id3 2020 1st sa 170 000km predjenih , za 14k eura, pa me zanima da li se isplati ili ne i da li je to pouch?
Takodje me zanima misljenje o novim jeftinim kinezima, koji nisu BYD(posto ste rekli blade baterije da nisu za preporuku), znaci tipa dongfeng box nami, jmev ev3 i slicno. Da li mi je bolje kupiti vw id3 ili id4 sa ozbiljnom kilometrazom ili nov jeftin kinez sa garancijom?
Lijep pozdrav.
[…] uppfattas ofta som ett säkert köp. Men enligt den kända kroatiska batteriverkstaden EV Clinic, som uppger att de reparerat omkring 2 000 högvoltsbatterier säger siffran betydligt mindre än […]
Great article. I appreciate the logic (and science) behind it. As a soon to be EV buyer I am interested how to use your findings in my decision making in the real world in Australia. For example – one of the higher weighted Degration factors is ACIR which you state requires pack removal and tesing by hand which I suspect is not a trivial nor cheap excercise. Please advise.
Also will you be publishing any further ‘reports’ eg repairability rankings and failure frequency (count?) for each EV brand/model.
Thanks