Who decides your battery is healthy

Somewhere around page 332 of the Tata Nexon.ev owner's manual, inside the battery warranty section, there is a sentence most owners will never read.

"after a maximum of four DC / opportunity charging cycles, you must use AC charging to 100% State of Charge for the optimum performance of the high voltage battery pack."

Read that as a condition rather than a tip, because that is where it sits. It appears in five of Tata's eight EV owner manuals: Nexon.ev, Harrier.ev, Curvv.ev, Punch.ev and Tiago.ev. It is absent from the Sierra.ev, the Tigor.ev and the XPRES-T EV. Three of the eight go further and add a limit on how often the car may complain about how you charge: "Vehicle shall not have more than 2 warnings related to EV charging pattern between two service intervals."

If you have a wall box at home, none of this will ever trouble you. If you live in a flat and charge where you can, which is most Indian EV owners, you are being held to a ratio nobody told you about.

And you have no record of your own ratio. Six of those seven manuals say who does. The warranty depends on you keeping the connected car subscription alive, because, in the manual's words, "IRA.ev subscription maintenance is crucial for telemetry data to be transferred back to TPEM to allow visibility of the last steady state of the HV Battery for application of the HV Battery Warranty."

So the company that pays the claim is the company that watches the car, and the owner is the only party to the contract who cannot see the evidence.

This is not about one manufacturer

We read them all. 67 models across 31 makers, four wheel and two and three wheel, at the primary document, across three days in September 2026. Not one press article, not one aggregator. Where a maker publishes nothing, we wrote down that it publishes nothing.

A correction, because this post originally said "around 45 models across 25 makers" while its own summary said 47. Neither figure had been counted from the dataset, and the maker count was in no file at any point. Counting it properly on 28 September turned up something worse than a wrong number: six makers the audit had read were missing from the published dataset entirely. Four were absence rows, which is the thing this piece argues absences are for. Two were not absences at all: Ultraviolette carries 8 years and 200,000 km on the Tesseract, and Hero Electric has a full warranty policy whose thresholds appear later in this piece.

They are all in it now. Dataset version 1.1.2: 67 models, 31 makers, 56 source documents, every one retrieved and checkable. The figures in this post are counted from warranty-terms.csv and nowhere else. A piece whose whole claim is that every cell can be checked should never have carried the word "around".

The pattern is the same everywhere, and it has four steps.

The maker sets a number. Usually 70 percent state of health. Tata, Mahindra, Toyota, Mercedes-Benz, Ather, Ola, Revolt and Altigreen all publish one. Kia publishes it for the Carens Clavis EV and the Syros EV, and for the EV6 and EV9 it publishes a clause titled a capacity warranty containing no percentage at all, until you open the owner's manual, where the 70 percent sits on page 592 of 605.

The maker measures it. By a method it does not set out. Two of the models we read name a condition under which state of health is measured. The Altigreen neEV TEZ, whose pack is warranted by Exponent, specifies a CC-CV profile at 54A. Hero Electric's lithium-ion Startup Guide, in a warranty policy printed in Hindi, gives replacement thresholds in amp-hours for a 30Ah battery at the C5 discharge rate.

Neither sets out a procedure. No temperature condition. No state of charge precondition. No rest period. No instrument. Nothing an owner or a third party could follow and repeat, and nothing that would produce the same answer twice in different hands. Every other maker publishing a threshold leaves the procedure that produces the number wholly undefined.

Ather says capacity is judged "as measured by Ather's standard testing procedure", and then never describes the procedure. Ola says "Battery Health certification will be done by the company only after the scooter has been driven for more than 10,000 kms."

The maker decides. Tata: repair or replace, "to be decided at sole discretion of Tata Passenger Electric Mobility Limited". MG: "JSW MG MOTOR's decision is final and binding on the Owner of the vehicle in all warranty matters." Hyundai: "HMIL's decision is final and binding on the Owner of the vehicle in all warranty matters." Kia: "KIN reserves the right for the final decision in all warranty matters." Ampere: "The right to determine whether the part needs repair ... rests solely with Ampere only." Mahindra reserves the right to appoint the inspecting engineer, including the one its own guide calls "an independent consultant engineer".

And nobody outside checks. Across the 67 models we read, not one gives the owner a right to dispute the reading. In most the question is not addressed at all: not refused, not conditioned, not appealable, simply not a thing the documents contemplate. Seven rows are coded as partial, and reading them is what makes the point rather than weakening it. Six are Mercedes-Benz, which gives you the baseline and no route to argue about the reading. The seventh is the Altigreen neEV TEZ, which requires a report within seven days and grants no independent right.

Mercedes-Benz is the only partial exception, and only on half the sum. It records rated capacity at production, prints it on a label on the pack, and will give it to you at any authorised dealer. So a Mercedes owner in India can learn the denominator of their own degradation calculation. No document in the audit sets out how the numerator is obtained.

Four makers, four different postures

Reading every manual from each maker, rather than one each, turned out to matter.

Tata has a number and varies the conditions around it. Eight manuals, and the DC charging cap is in five, the telematics requirement in six, the charging warning limit in three. The Sierra.ev, the newest car, keeps the 70 percent floor and the telematics requirement and drops the DC cap entirely. We searched its charging section for any equivalent wording and found only ordinary explanation of how fast charging works. Whether that is a deliberate change of policy is a question for Tata, not a conclusion for us.

Tata's warranty page carries the 70 percent threshold in plain sight, which is more than most makers do: "In the event the HV Battery falls below minimum 70% of its State of Health ("SOH") of nominal energy at any time during the warranty period". What it does not carry is how that number is produced. It also says the quiet part: "Unless such gradual loss results in a depreciation of the SOH value for the HV Battery below said threshold of 70%, the HV Battery Warranty being offered herein shall not apply." Ordinary ageing is outside the warranty until the day it crosses a line only the maker can see.

Three other things are on that page, each behind a plus sign a reader has to click.

"Lifetime" means fifteen years. The word appears on every Tata EV page and the definition is one level down: "The term 'Lifetime' refers to the period of fifteen years from the first date of registration of the vehicle."

"Unlimited kilometres" does not include working the car. The exclusion names display, demonstration, test drives, courtesy use, commercial operations, fleet management and taxi services. Those registrations get 8 or 10 years and a kilometre cap.

And the lifetime warranty does not follow the car to its next owner on its own: "The lifetime HV Battery Warranty shall not be available to the second or subsequent owner in absence of such notification." A used buyer who does not know to tell Tata about the transfer has no lifetime warranty, and nothing in the sale tells them.

The page lists six models. The Tigor.ev, the XPRES-T EV and the three smaller packs are not on it.

The XPRES-T EV is the one that changes the shape of this. It is the fleet and taxi variant, the vehicle in the range that will cycle its pack hardest, and its manual carries none of the three: no DC charging cap, no telematics requirement, no charging-warning limit. Searched in full, zero matches for each. It also carries no state of health figure anywhere. What it does carry, in clause 1 on page 145, is the whole warranty: "This warranty for TATA XPRES-T EV (Inclusive of Battery and Electric Powertrain) shall be for a period of 3 years from the date of sale or 1,25,000km, whichever occurs earlier." That is the shortest battery term of any Tata EV we read. The car worked hardest is given the fewest conditions and the least time.

Mahindra has a number and applies it consistently, to four of its five. The warranty guides for the XEV 9e, BE 6, XEV 9S and BE 6 FE carry the same tiered 85, 75 and 70 percent structure and the same charging exclusions. None of the four caps DC fast charging.

The XUV400 has no guide. Its warranty appears in one place, two rows at the foot of the specification table on page 8 of a marketing brochure: "BATTERY PACK AND MOTOR WARRANTY: 8 YEARS OR 1,60,000 KM" and "VEHICLE WARRANTY: 3 YEARS OR UNLIMITED KM", with a footnote reading "whichever is earlier". We read all eight pages. There is no state of health floor, nobody named as the party who measures, no method, and no charging exclusion.

So the same maker gives four models a tiered floor a buyer can point at, and the fifth a line item between the wheel size and the paint colours.

MG has no number. We read the full text of the ZS EV, Comet EV and Windsor EV Service and Warranty Manuals, and there is no percentage figure in any of them. Cover triggers where the pack "has suffered a capacity drop below acceptable degradation according to the use and ageing of the vehicle". The exclusion removes cover "If the degree of degradation of the high-voltage battery is within the normal aging level". Neither phrase is defined anywhere in the document.

Those two clauses are the same clause facing in opposite directions, and the only party who can tell you which one applies is the party whose decision the same manual calls final and binding.

BYD has a period and excludes the thing the period is about. The Seal and ATTO 3 policies give the traction battery 8 years or 160,000 kilometres, and then, in the exclusions, put "the normal attenuation of battery capacity" outside the scope of the warranty. We searched both documents in full for a state of health figure, a retention percentage, any reference to the term. There is none. So the period covers the pack against failure and says in terms that it does not cover the pack getting older, which is the only thing most owners are worried about.

The comparison that makes this worth printing is BYD's own. Its warranty manual for Europe and Israel carries the same exclusion and then adds an exception to it: capacity loss is not covered, except to the extent specified in this Power Battery Coverage. That coverage is 96 months or 200,000 kilometres with minimum 70 percent retention, and state of health appears in the specification table as its own column with a footnote defining it.

Same manufacturer. Same packs. 96 months in both markets. Europe gets 200,000 kilometres and a floor the buyer can hold BYD to. India gets 160,000 and an exclusion with no exception after it.

The eMAX 7 and Sealion 7 documents are different again: they carry the warranty period table and stop, with no limitations or exclusions section at all, and refer the reader to "the further policies from BYD in its official website". We have not attributed the attenuation exclusion to those two models, because their documents do not contain it.

There is no standard, and the Government has written that down

We went looking for the Indian standard that defines how state of health is measured. AIS-038 Revision 2, AIS-156, AIS-040 Revision 1 and AIS-049 Revision 1, read at ARAI's own server. All four are type approval instruments. They test a new vehicle or a new pack for electrical safety, thermal and mechanical abuse, fire resistance, overcharge and over-discharge protection, and, in AIS-040, a laboratory range figure on a new car. None of them measures a battery already in service. IS 17855:2022 is bench performance testing of a new pack.

No instrument we could find requires a maker to publish its warranty terms, declare a floor, or disclose a method. The Consumer Protection Act 2019 makes it an unfair trade practice to give a warranty of "length of life" that is "not based on an adequate or proper test", which is a duty to substantiate rather than a duty to tell you, and there is no prescribed test to substantiate against.

The clearest statement of the gap is the Government's own. The draft Battery Pack Aadhaar guidelines from the Ministry of Road Transport and Highways put State of Health in the battery's dynamic data set, held on a central server, and then say this:

"The detailed methodology for validating SOH needs to be separately developed by the Government."

A field, with no method.

For contrast, the EU battery passport becomes mandatory on 18 February 2027, and Article 14 of Regulation 2023/1542 requires state of health and expected lifetime parameters to be held in the battery management system with read only access granted to the owner, to independent operators and to waste operators, expressly including for evaluating residual value. Europe prescribes the parameters and mandates the access. India has named the field and deferred the method.

What happens when somebody argues

We looked for a decided case. There is one that is directly on point, and it went the way you would expect.

A man bought a battery operated e-rickshaw on a representation that it would run 80 to 100 km per charge. It delivered about 20. The battery was replaced four times in three months and nothing improved. The District Commission ordered a refund with interest.

The Madhya Pradesh State Commission set that aside and dismissed the complaint. First Appeal No. 358 of 2021, order of 19 September 2024. The reason, in the Commission's words: "In absence of expert evidence it cannot be concluded that the subject e-rickshaw was having manufacturing defect."

He lost because he could not prove it. There is no prescribed Indian method by which he could have proved it.

And if he had wanted to try, the queue is 5,74,333 cases pending as of January 2026, with 169 President and 379 Member posts vacant at district level, in a system whose statute asks for disposal within five months where testing is required and uses the words "endeavour shall be made".

The shape of the problem

Put the pieces next to each other.

You are held to a standard of charging conduct, on equipment and an electricity supply you do not own or control. The threshold that decides your claim is set by the maker, where it is published at all. The measurement is taken by the maker, by a method no document describes. The decision is the maker's, at its sole discretion, final and binding. No Indian standard defines the measurement. The regulator has written down that the method still needs to be developed. And the one owner who took it to a commission lost for want of evidence that nothing in the country is set up to produce.

None of that is an accusation of bad faith. Every maker here is operating inside a vacuum that nobody created on purpose. A warranty needs a threshold, a threshold needs a measurement, and in the absence of a standard the only measurement available is the maker's own. That is not a conspiracy. It is what happens when the instrument does not exist yet.

But it does mean something concrete for anyone buying an EV in India this year. The most valuable component in the car is warranted against a number you cannot see, measured by a method you cannot read, assessed by the party that pays if the answer goes your way.

Where we come into this

We are building an independent measurement of charging reliability in India, from the driver's side. The first Mumbai index publishes on 5 November 2026.

This piece is not that. We have measured nothing here. Everything above is what 31 makers and the Government of India have published about themselves, read at source between 22 and 29 September 2026, with the absences recorded as carefully as the presences.

We publish it because the same gap runs through both. A charging network reports its own uptime. A manufacturer measures its own battery. In each case the number that matters most is produced by the party with the most at stake in the answer, and in each case the reason is not malice, it is that nobody else is counting.

One of those two we intend to fix.

The table

The audit is published as an open dataset, under CC BY 4.0, at github.com/voltu-energy/india-ev-battery-warranty.

Three files. One row per model with the structured terms. One row per clause, quoted verbatim, each carrying the exact models whose documents we read it in. And one row per source document, with its URL, the date we read it, and how we reached it, including which ones were blocked to automated retrieval and had to be opened in a browser.

There are no blank cells, because a blank would hide the most interesting thing in it. Eight result codes instead, and the distinctions matter. NOT_FOUND means we could not find the document. NOT_STATED means we read it and it is silent. NOT_DISCLOSED means a threshold is set with no method described. NO_FIGURE_IN_DOCUMENT means we searched the full text and there is no percentage in it at all, and that one exists because of MG. NOT_ADDRESSED means the document does not raise the question, which is what almost every row says about the owner's right to dispute. PARTIAL means the document goes part of the way and stops, and it is the code carrying Mercedes-Benz's obtainable baseline and the neEV TEZ's seven-day report. DISCLOSED is one row. And EXCLUDED is the opposite of an absence, which exists because of BYD: the document addresses the thing and puts it outside cover, a stronger fact than silence that would have been flattened into a shrug as NOT_STATED.

A script in the repository enforces the rule the dataset stands on: any value that is not an explicit absence must reference a source document, and any quoted clause must carry the models it applies to. It runs on every change. That second check is what stops "five of Tata's eight owner manuals" becoming "Tata" the next time somebody edits in a hurry.

Source rows also record what the document was, not only where it was. Every PDF source but one carries a SHA-256 of the file we read, and 34 of the 54 carry an Internet Archive capture.

That is there because of BYD. On 29 September 2026, checking from outside India, four of five BYD India asset URLs that search engines had indexed did not serve, including all three warranty policies we tested. One brochure on the same host did serve, so we are not saying anything was taken down and we have not established why. What we can say is the narrow thing, with a date on it: documents that had been publicly retrievable did not answer when we went back.

A hash does not bring a document back. It lets anyone who obtains a copy by any route prove it is the one we read, which is the only version of "check it yourself" that survives the page going away.

One scope note we would rather state than have pointed out. Tata's smaller packs, the Nexon.ev 30, the Tiago.ev 19.2 and the Punch.ev 35, are not listed on the warranty page and have no manual of their own, so we have recorded them against the larger variant's manual rather than as separate terms. It is in the dataset with the reason attached.

A third gap closed while this piece was being corrected. The XPRES-T EV manual was obtained and read in full, and what it says is above.

If you work for one of the manufacturers listed and we have a cell wrong, that is the most useful message this project can receive. Open an issue with a link to the document, or write to hello@voltu.energy and we will file it for you. We will correct it in public, with the date, crediting whoever caught it, exactly as we correct our own.