What's new

When we cannot stand behind an engine code, the vehicle page no longer shows one anyway

A vehicle whose engine code could not be verified was still having an engine named for it elsewhere on the page and in the API. A 2003 Nissan Almera displayed a 2020 engine. A 2012 Civic's data named a Japanese-market Fit engine while the correct code sat unused. Those claims are now withdrawn wherever they appear, and nothing is thrown away: each withdrawn value is kept in the response, marked unverified, with the reason it was held back.

• 3,641 unverified engine claims were removed across 486 vehicles in the test set. They appeared in the engine family line on the vehicle page, in the engine name, and in several provenance fields returned by the API.
• The engine family line is shown on the vehicle page, so this is visible: a 2003 Almera read 'Engine code: -' with 'Nissan HR10DET, 1.0L 3-cylinder turbo' printed beneath it. That engine is from 2020 cars.
• Nothing is deleted. Every withdrawn value is preserved in the response under an explicitly unverified field, together with which rule withheld it, so nothing that was known is lost.
• Sentences are no longer edited word by word. An earlier attempt cut engine codes out of prose and produced text like 'Camry Hybrid Hybrid'; the shipped version drops a whole clause or the whole field instead.
• This does not fill in missing engine codes. The number of vehicles showing no engine code is unchanged, by design: the codes being withheld were measured as correct only about 43% of the time, and showing them would make the decoder less reliable, not more.
• Known gap: one vehicle in the examples still reports 3 cylinders inherited from the wrong engine. A different rule owns that field and it is queued.

North American VINs now use NHTSA's own per-VIN record for the plant, the model name and the engine details we were holding back

Manufacturers must file their VIN decoding with NHTSA, and we already held its answer for 157,000 VINs without using it. Comparing 2,400 North American VINs showed the plant city wrong 46% of the time, model names cut short on one in eight, and engine details hidden on the majority of vehicles even when our own data had them. This release uses the NHTSA record to confirm and release those values, and repairs several wrong model names.

• Plant city was wrong on 46% of North American vehicles and is now taken from NHTSA. A 2015 Jeep Compass that read Detroit now reads Belvidere, Illinois.
• US-built Toyota Highlanders were being shown as Kluger, the Australian name. They now read HIGHLANDER, and their engine, cylinders and power are unchanged.
• Other model repairs: a Subaru shown as Crosstrek is an Impreza, a Nissan shown as Sentra is a Maxima, and 'NEON HIGHL' now reads Neon. Cut-short model names fell from 13.4% to 3.3%.
• Engine details we held but hid are now shown when NHTSA's record for that exact VIN agrees. Missing engine codes fell from 70.9% to 55.6%, missing displacement from 31.2% to 13.1%, missing cylinders from 12.5% to 4.0%, and missing fuel from 9.1% to 3.6%.
• A 1991 Ford Ranger was returned as a 2021 on the fast API. It now returns 1991.
• Trade-offs, stated plainly: 48 vehicles lose the assembly plant's name while gaining the correct city, one Highlander's engine code is now withheld rather than shown, and two vehicles' series code regressed.
• 1,000 vehicles built outside North America were checked and none changed.

Vehicles now show the assembly plant that built them, and a model year can no longer change between identical requests

NHTSA's per-VIN record names the plant a vehicle was built in, and we were dropping that name instead of serving it. It is now shown for 415 vehicles in our test set, checked against the vehicle's own city and state so a corporate label such as 'GMNA' is refused rather than displayed. Separately, a year-correction rule was switching itself off on a vehicle's first decode, so the first person to look up a VIN could see a different year from everyone after them.

• 415 vehicles gained the name of the plant that built them. Nothing was overwritten: that field was empty on every one of them.
• NHTSA's text is free-form and carries typos, so the name is repaired to the spelling the record itself uses elsewhere for the same plant. 'Michigan Assy' now reads 'Michigan Assembly', and 'Ohio Assebly Plant' reads 'Ohio Assembly Plant'.
• Names that identify a company rather than a plant, such as 'GMNA', are refused. The field stays empty rather than showing something misleading.
• Plants that share a city stay distinct: Toledo North Assembly and Toledo South Assembly are never merged.
• A model year could previously change between two identical requests. The first-ever lookup of a VIN served an uncorrected year and held it for an hour, while later lookups returned the corrected one. The year correction no longer switches itself off.
• Known gap, unchanged: four flex-fuel vehicles are still shown as petrol.

A failed lookup no longer hides a vehicle's engine details until restart, and the year's confidence dot now reflects the evidence

Two quiet defects. A failed call to NHTSA was remembered forever by whichever server handled it, so a single two-second timeout could hide a vehicle's engine details for hours. And the confidence dot beside the model year showed maximum certainty whenever a network lookup had set that year, even though that lookup checked nothing at all.

• A failed NHTSA lookup is now retried after five minutes instead of being remembered until the server restarts. Before this, one timeout could leave a vehicle missing its engine details indefinitely, and two servers could disagree about the same VIN forever.
• Successful answers are still kept for the life of the server, so this does not increase traffic to NHTSA.
• The year's confidence dot no longer claims maximum certainty just because a network lookup set the year. That lookup applies no checks: no verification of the VIN's check digit, and it never even reads NHTSA's own error code.
• Where the VIN's year letter reads as the same year one 30-year cycle apart, that is now treated as agreement rather than a contradiction, so the dot does not drop for the wrong reason.
• A dead check meant a vehicle's model year could disagree with its year. It is fixed. No vehicle changed today, so this is a guard against future drift rather than a visible improvement.

Safety recalls now show on every decode, and Toyota engine codes are confirmed from the dealer catalogue

NHTSA safety recalls were only shown the first time a model was ever decoded; on every later decode a date-handling fault hid them. That is fixed, and failed NHTSA calls can no longer be recorded as 'no recalls'. Toyota and Lexus engine codes are now checked against Toyota's own dealer parts catalogue, which fills or corrects about 1,800 vehicles. Fast API requests no longer make outside network calls.

• Safety recalls: recall lists already on file are now shown on every decode, not only the first time a model was ever decoded. A date-handling fault had hidden them since at least June 2026.
• A failed call to NHTSA can no longer be stored as 'no recalls'. Old 'no recalls' entries written during outages are re-checked on the next decode, and an NHTSA outage no longer hides recalls we already hold.
• Toyota and Lexus: 1,677 vehicles that showed no engine code now show one, and 119 wrong engine codes were corrected, each confirmed by Toyota's dealer parts catalogue. Checked against NHTSA's per-VIN data, none of them disagree.
• Fast (audit) API requests no longer call NHTSA or other outside services. On 2,400 North American VINs this changed 7 fast answers (6 Tesla 'NHTSA confirmed' flags, 1 recall list). The website's normal decode still uses NHTSA.
• Known gap, fix in progress: some models still show no recalls because the lookup uses the model name before it is final (for example a 2019 Ford F-150 or a 2017 Jeep Grand Cherokee).

Safety recalls are now found under the name NHTSA actually files them under, and the site's VIN-pattern count is honest

Recalls were looked up while the model name was still provisional, so NHTSA rejected keys like GRANDCHEROKEE, F150 or SEDONA EX and the vehicle appeared to have no recalls. The lookup now runs once the vehicle is fully identified and uses NHTSA's own model name for North American VINs. On 500 test VINs the right recall list is now returned for 487, up from 304. Separately, the headline that claimed 930,524 decodable VDS codes was wrong by about 14 times and has been replaced.

• Recalls: a 2019 Ford F-150 now shows 9 recalls, a 2017 Jeep Grand Cherokee 6, a 2006 Kia Sedona 7 and a 2020 Chevrolet Silverado 1500 12. All four previously showed none.
• Every recall list now states which vehicle it was fetched for, so a list can never be shown silently for a different model.
• Lists shown for vehicles built outside North America are marked as US-market NHTSA recalls, with a note to check whether they apply.
• The recalls-by-VIN endpoint now returns exactly the same list as the decode, and reads NHTSA's park-it flags under their correct names.
• The site's headline figure now reads about 66,800 road-vehicle VIN patterns instead of 930,524 'decodable VDS'. The old number counted the check digit, ignored the maker code and included around 332,000 boat hull numbers.
• Removed unsupported claims of sub-millisecond and under-5 ms decoding; a cached fast decode measures about 23 ms at the median.
• Known gap: the recall panel is still not shown on the website. The data is in the API and the display is being wired up next.

Electric GM vehicles are no longer given a combustion engine

Battery-electric Chevrolet, GMC and Cadillac models were being served a petrol or diesel engine code inherited from the position-8 engine tables. When the record itself shows the vehicle is electric, that code is now withheld and the evidence is stated instead.

• A 2024 Silverado EV was being shown with the L5P 6.6-litre Duramax diesel; Hummer EV pickups and SUVs with a 3.6-litre V6 or a 1990s 3.1 V6 label. Those are now withheld.
• The check only fires on positive evidence that the vehicle is electric - the manufacturer's own electrification data on the record, or a VIN line that is verified electric-only. Vehicles whose powertrain is unknown are left exactly as before.
• Plug-in hybrids keep their real engine codes: the Ampera, Volt and ELR were tested and are unchanged, as were roughly 3,500 petrol and diesel trucks and SUVs.
• When a code is withheld, the response carries it under engine_code_contested with the reason, and a warning that states the evidence on file rather than a verdict about the vehicle.

BMW engine codes from the dealer catalogue are now checked against the engine reference before being served

The BMW dealer-EPC path serves engine codes for about three quarters of BMW VINs. A small class of catalogue tokens that no engine reference recognises was being served at full confidence - and in some cases used to overrule and blank out a car's displacement and cylinder count. Those are now served at engine-family grain, or withheld with the evidence stated.

• 579 of 614 catalogue type-code entries name an engine the reference knows; those are unchanged. The 35 that do not (for example B32S15A, where the only known B32 family is a Mazda engine) now serve the family or abstain.
• One class of X5 M / X6 M / XM cars was being shown as a 3.0-litre six-cylinder because of an unrecognised S68B44A token; they now correctly show the S68 4.4-litre V8.
• Nothing that was served correctly changed: 0 of 984 keyed BMW VINs went from a value to nothing, and 150 North-American-WMI BMWs, 150 unkeyed BMWs, 600 non-BMWs and 45 BMWs with human overrides were byte-identical.
• This is the first step of the programme to put the dealer EPC catalogues to work across brands; Nissan/Infiniti is next.

Nissan and Infiniti engine codes now come from the dealer parts catalogue

We hold Nissan's own dealer catalogue and a per-VIN factory build table, but the decoder had never used them. It now looks up a Nissan or Infiniti VIN in the factory table, reads the chassis code, and takes the engine from the catalogue. On a 1,761-vehicle test, 776 cars that previously showed no engine code now show one.

• 776 vehicles went from no engine code to a real one. Nothing lost: no vehicle's year, make, model, trim, drivetrain or transmission changed, and confidence never went down.
• Where the catalogue lists several engines for a chassis, the vehicle's own displacement and cylinder count decide; if they cannot, we say nothing rather than guess.
• The catalogue also corrects wrong codes: 79 were replaced and 19 withdrawn, all from a structural rule measured wrong on 111 of 115 checks. An Infiniti Q45 showing a 5.6-litre V8 now correctly shows its 4.5.
• Checked against four sources the catalogue has no hand in: our own registration buckets agreed on all 289 comparable cars, factory geometry on 991 of 991, and human-verified records on 196 of 196.
• Human overrides and verified records were never overruled: 269, 29 and 12 of them respectively came through untouched.

When a VIN looks mistyped, we now offer a lead instead of asserting a correction

If your VIN's manufacturer code is one we have flagged as commonly mistyped, we used to tell you what your vehicle 'really' is. We now say what we actually know: that the code resembles another one, and that you should check it against the vehicle.

• The old message said a code was 'likely a typo' and named 'the legitimate WMI'. That is a per-code table entry applied to every vehicle sharing the prefix - it cannot tell you what YOUR car is.
• Measured on registration cards: the six vehicles whose VINs read as 'PH1' are a Toyota, a Honda, a Renault and BMWs. The single stored correction pointed all of them at Toyota.
• Nothing was hidden or removed. The suggested VIN is still returned exactly as before - we only stopped stating it as fact.
• We tested two designs that would have SUPPRESSED suggestions instead, and rejected both: one would have silenced correct suggestions for the newest cars, and the other would have rejected 5 of the 11 VIN corrections we had already verified and accepted.

Fewer wrong answers in every category

We found a flaw that let one vehicle's engine details be quietly copied onto unrelated vehicles - and fixing it reduced wrong answers across engines, displacement, year and fuel all at once.

• When the decoder had little to go on, it borrowed details from a 'typical' vehicle sharing the VIN's manufacturer prefix. That typical vehicle was picked from an unsorted sample, so it was often not typical at all - on one Mazda prefix it picked a model making up under 1% of the vehicles.
• The borrowed details were then presented as though measured on your vehicle. A 1990s Mazda was being reported with a diesel engine that did not exist until 2012.
• The decoder now abstains instead of borrowing, and says so. Ten wrong answers were removed from our regression set with no new ones introduced - the best scores we have recorded in every field.
• Vehicles whose VIN does not encode a year in the standard position, and vehicles with verified engine records, are explicitly left untouched.

Thirty-year misdates corrected on vehicles we know least about

A VIN's year character repeats every thirty years, so 1988 and 2018 look identical. When the decoder had to guess which, it was allowed to lean on evidence that was never about your vehicle - and sometimes chose thirty years wrong.

• A 2018 Land Rover was being dated 1988, and a 2011 Volkswagen 1981. Both now decode correctly.
• The decoder may no longer break that thirty-year tie using a range pooled from similar vehicles or a window covering a manufacturer's entire history - neither says anything about the specific car.
• Where a genuine per-vehicle record exists, nothing changed: the tie-breaking rule that works is correct 99.5% of the time across 700,000 vehicles and was left exactly as it was.

Cars are no longer turned away as boats

Some ordinary cars were being refused outright, with the decoder insisting they were boats. They now decode normally - and genuine boats and trailers are still correctly declined.

• A Kia Spectra and an Isuzu D-Max were among the vehicles rejected. The D-Max now correctly decodes as a pickup.
• The cause: almost half our reference records are American boat and trailer registrations, which are filed under a 12-character hull number rather than a 17-character VIN. When a car had no exact match, the decoder looked at vehicles with similar-looking codes and inherited 'boat' from those records.
• Because a boat's hull number can never be a 17-character VIN, we can tell that kind of coincidence apart from a genuine match with certainty - so real boats and trailers are still declined, as they always were.

We fixed how we measure ourselves

Our regression tests were quietly ignoring every vehicle the decoder refused to answer, and reporting a diagnostic column as zero when it was not. Both are fixed, and our scores are restated honestly.

• If the decoder refused a vehicle outright, our own scoring silently removed it from the test set rather than counting it against us - so refusing a car scored better than answering it wrong.
• A column meant to show how many vehicles we could not verify at all was hardwired to zero in every report we have ever produced. The real figure is between 52 and 81 vehicles per field.
• Restated honestly, the decoder's error counts are unchanged - the refusals were never inflating them. What changes is that the numbers now show their own denominator.
• No change to decoding. This affects only how we grade ourselves.

When we can't decode a VIN, we now say why - accurately

If we turn a VIN away, the message now tells you what we actually observed rather than asserting what your vehicle is. One customer's MINI was being told it was a trailer.

• Previously a refusal said 'this VIN appears to be a non-passenger vehicle' in two very different situations: when the vehicle's own record says so, and when only the manufacturer code suggests it. Those are now worded differently, because we know far less in the second case.
• A 2019 MINI Cooper S was refused as a trailer. Its VIN had been mis-transcribed by a single letter, and the code it landed on belongs to a trailer manufacturer - so the message now says the code, not the car, is what looks non-passenger, and that it may be a typo.
• Where we already hold a reviewed correction for a mistyped manufacturer code, the refusal now offers it - while making clear we have not decoded it and you should check it against the vehicle.
• We deliberately did not add automatic guessing of near-miss VINs: tested across our whole database it would have pointed roughly one time in thirty at a motorcycle, truck or bus instead of the car.

Verified answers replace blank fields

Verified engine codes now fill from factory-corroborated cells (Skoda Octavia BVZ, VW Jetta AWP/CBUA, Subaru FB25), a measured 98.5%-accurate cylinder fill drains 40% of the missing-cylinders backlog, Amarok/Crafter/Ranger diesels get their fuel from coherent engine evidence, and family-label power guesses (76.6% wrong when measured) now abstain while honest power ranges keep serving.

• Model-year 2027 vehicles decode correctly, and a live suppression-cascade bug was caught and structurally fixed before release: a marketing family label can never again pose as an engine code.

Build years for VINs that do not encode them

Serial-range ladders mined from 63 verified production cohorts now serve real build years for Toyota, Mazda, Nissan and Mercedes vehicles whose VINs encode no year at all - each cohort ships only after clearing a 95%-accuracy holdout gate, and vehicles outside a proven cohort keep honest year ranges instead of guesses.

• Genesis sedans get their correct engines per market (the same VIN letter means different engines in Korea vs America - now handled), classic Subarus decode their era-correct engines, model-year-2027 vehicles resolve properly, Ford Super Duty trucks shed a phantom NAVISTAR petrol engine for their real Power Stroke diesels, and 179 engine records damaged by an old data-link bug are restored.

Ford Super Duty engines and cleaner provenance

Ford Super Duty trucks decode their real engines per year-band (year-guessing default deleted - unknown codes now honestly abstain), the European VW/Audi engine tail joins the abstention framework with its downstream writers gated in the same release, Genesis 3.3/3.8 sedans gain 9,000 verified engine fills, Toyota HiLux and Nissan X-Trail/Patrol gain new verified year ladders, four statistically-impure inference rules are disabled with their weekly-rebuild resurrection path closed, and mixed-language fuel labels are normalized at serve time.

Tacoma and Sportage engines now fully shown

User-reported fix: Toyota Tacoma (3.5 V6, 2GR-FKS) and Kia Sportage (2.5 GDI, G4KN) VINs now serve their full engine identity - the decoder had resolved both codes internally but a safety gate hid them.

• Fixed with fully verified per-family rows (420/420 corpus agreement on the Sportage), and 373 corrupted reference windows repaired, improving engine and displacement accuracy across the board to campaign bests.
• Two broader gate relaxations were tested and rejected by measurement - correctness gates stay strict where the data still demands it.

Mazda CX-9 gets its real engine back

A user spotted a suspicious engine label on a 2014 CX-9 - and they were right.

• 103 database rows carried a scraped junk label, 'CAV California', including a misspelled twin, instead of the real engine code.
• The junk was contesting the correct code and hiding it; all rows now carry CAY1, the factory 3.7L V6, verified against Mazda's own parts catalog.
• First-generation 3.5L models are correctly left separate, and the affected Mazda3 rows were cleaned to honest-empty.

Ford engines un-hidden: Explorer, Mustang and truck data freed from a bad rule

A user's 2020 Explorer was missing its engine - the culprit was a legacy one-size-fits-all Ford letter table that contradicted correct data and made the decoder hide it.

• 2020-2025 Explorer now decodes its engines with verified specs: 2.3 EcoBoost (300 hp), ST 3.0 (400 hp), 3.3 V6 (285 hp) - each confirmed against Ford's own VIN guides.
• 2016-2019 Explorer 2.3 correctly shows 280 hp, and the stray 270 hp figure (which belongs to the Ranger) was cleaned out of the data.
• The retired rule had been silently blanking engine codes and horsepower on thousands of Ford, Lincoln and Mustang vehicle groups - those now decode from real data instead.
• Also fixed: US Explorers no longer show the European electric model's VW platform note.

BMW engines now verified against the manufacturer's own dealer catalog

BMW and BMW M engine identification is now cross-checked against the same OEM parts-catalog data BMW dealerships use - 12.6 million catalog rows now back the decoder.

• Over 600 BMW vehicle groups that previously returned no engine code now serve one, straight from the dealer catalog - including full factory designations like B48B20B and S85B50.
• Existing engine codes that agree with the OEM catalog are upgraded to dealer-catalog confidence.
• The catalog also quietly fixed real errors: an E39 M5 that was being served the wrong generation's V10 now correctly shows its S62 V8.
• Ambiguous catalog entries are deliberately excluded - the decoder only serves what the catalog is unambiguous about.

The dealer catalog now has the last word on BMW engines

We found the decoder second-guessing the manufacturer's own parts catalog - and sometimes replacing a correct engine with an impossible one. The catalog now wins.

• A date-checking rule was treating OEM catalog data as a guess. On a 2-Series it swapped the real 2.0 four-cylinder for a 4.4 V8, and on an M3 it fitted the M5's V10 - both triggered by a misread build year. Catalog engines are now protected from it.
• One M3 was even being reported as an M5, because a hardcoded lookup had its chassis wrong. The catalog's 7,160 records for that car say otherwise, and it now decodes as the M3 it is.
• Where the catalog contradicts our other data, the catalog now wins outright and is served at full confidence.
• The exception is your vehicle's own registration document: if it disagrees with the catalog we keep what the document says, flag the difference, and show you what the catalog reads - a swapped or replaced engine is a fact about that specific car.

Honda CR-V and Accord records cleaned against the factory catalog

We removed engine records that Honda's own dealer catalog shows are impossible for the cars they were attached to - including a third-generation CR-V engine recorded on second-generation vehicles.

• Around 3,900 second-generation CR-V records carried the K24Z1, an engine Honda did not build until the next generation. They now carry the K24A1 those cars actually use.
• For 1997-2001 CR-Vs we removed the engine code entirely rather than guess: the record key those vehicles share cannot tell a 1998 car from a 1999 one, and Honda changed the engine exactly at that line.
• A small group of 1990s Accord records listed a four-cylinder engine on V6 cars; those now carry the correct V6.
• Every change was checked against Honda's dealer catalog and independently against NHTSA's own VIN records before being applied.

The three biggest error families, fixed

Wave D targets the three biggest remaining error families found by our 937,000-VIN test campaign: Nissan/Infiniti export VINs now decode their factory nameplate from the chassis code inside the VIN (Teana no longer masquerades as Sentra), Genesis and Hyundai luxury models get their true engine codes from the VIN engine letter (verified against factory catalogs - including 5.0 V8s that decoded as 3.3 V6s), and the Genesis-to-G80 2017 rebrand is now era-aware in both directions.

• Also fixed: Nissan Thailand Navara years, pre-2000 Skoda years, Audi A6/A7 platform confusion, and VW family-label displacement guessing on Touaregs and Tiguans.

A structural honesty layer for engine specs

Marketing-family engine labels (like VW EA211 or Ford ECOBLUE) can no longer smuggle a family-default displacement or cylinder count onto a specific car from ANY data layer - a structural rail catches all of them at serve time.

• Mixed-fleet fuel guesses below 94% certainty now abstain on every make, US registry body-type codes can no longer masquerade as model names, and a dozen verified per-VIN fixes land: Jeep Grand Cherokee CRD diesels, MG and Volvo EVs, Toyota Yaris Cross identity, Kluger and Corolla Cross hybrids, and four Mercedes C/B-Class trim corrections verified against parts catalogs.

Model names cleaned up across six brands

Wave F cleans up the model-name long tail across six brands: Dodge Grand Caravans no longer decode as Journeys, Nissan Muranos/Pathfinders/Xterras get their real names back (full letter-table audit), Audi Q5s stop reading as Q3s, Ford S-MAX identity restored, Renault Scenic/Kadjar/Express serve their own nameplates instead of platform siblings, VW Taigo and Skoda Kodiaq/Superb decode from the correct VIN positions, SEAT Arona/Ateca land right, and two internal self-contradictions (BMW model-vs-series, Mercedes W204-vs-W203) are fixed at the source.

• Mercedes trim rows verified against parts catalogs; market-name aliases (Kluger/Highlander, Kuga/Escape) now recognized.

VW-group fuel accuracy overhauled

Fuel-type accuracy overhaul for European VW-group vehicles (VW, Audi, Skoda, SEAT): the decoder no longer guesses petrol vs diesel from thin family statistics.

• Fuel now requires real per-vehicle evidence - registration records, verified engine codes, or a clean fleet consensus - and honestly reports unknown instead of guessing.
• Measured on our ground-truth corpus: wrong fuel labels cut by a third globally, and 4 out of 5 wrong VW-group fuel labels eliminated.

Nine fixes from the million-VIN test campaign

Nine-class accuracy wave from the full-corpus test campaign: Subaru Forester/Legacy engines now decode from the chassis code in the VIN itself (petrol cars no longer served a phantom diesel), Ford Ranger/Everest V6 models keep their real 6-cylinder identity, Nissan Patrol/X-Trail export VINs get honest production-window years instead of a wrong decade, Toyota bZ4X decodes as the EV it is, Volvo V70s are no longer mislabeled V90 (and a fictional Polestar rule is gone), Mercedes Vito vans keep their name, and Chery models abstain rather than guess.

• Verified against registries and factory catalogs across markets.

Skoda fuel read straight from the VIN

Skoda fuel now decodes from the engine letter inside the VIN itself (75 verified cells across Fabia/Octavia/Kamiq/Karoq/Enyaq - petrol, diesel, hybrid and electric all identified structurally, with unknown letters honestly abstaining).

• Audi S/RS badges are no longer guessed: a European Audi only wears an S3-style badge when real power evidence backs it.
• And the decoder stopped passing off statistical guesses as measurements for VW-group displacement - wrong displacement on our ground-truth corpus dropped by a third this wave, wrong fuel labels fell further, with honest unknowns instead of confident errors.

Golf-family precision and five more fixes

VW Golf-family VINs now distinguish Golf, Golf Plus, Jetta, Beetle, Sportsvan and e-Golf using plant and serial evidence inside the VIN.

• The Subaru chassis-cell engine table extends to nine more families (older Foresters, Legacy, Impreza/WRX) with every cell verified against factory catalogs; Mercedes V-Class/Vito Japan-market keys return at precise 4-digit grain; Lada serves proper VAZ model numbers the way registries do.
• A Toyota-UK year bug (1986 on 2021 cars) is defused.
• And Volvo XC70/V90 Cross Country mislabels are fixed with the last guess-prone label rows removed.

Nissan export VINs get their factory model names

Nissan JN1/JN8 Japan-export VINs now decode the factory model from the chassis code in the VIN itself (50 chassis families, e.g.

• Teana, Q50, X-Trail, Almera, Urvan) instead of falling back to a wrong lookup guess.

The decoder now understands VIN construction logic

A breakthrough in how our decoder works: instead of only recognizing VINs it has seen before, it has now extracted the underlying construction grammar manufacturers use - which positions encode the model, chassis and engine family for Honda, BMW, Renault, Mazda, Nissan and Subaru. Proven on thousands of held-out combinations at up to 100% accuracy and stress-tested against real-world registrations before going live, this means the decoder can correctly identify vehicles it has literally never encountered - because it understands the logic, not just the examples.

Honda Odyssey engines now exact across all four generations

A generation-blind rule was assigning the newest Odyssey engine to every North American Odyssey, and our own safety rails were catching the contradiction and hiding the engine rather than serving a wrong one. The chassis code inside the VIN itself now selects the correct engine for each generation - from the 1999 original through the current model - with the mid-2000s generation honestly reporting the engine family where the exact variant depends on trim.

Generation-aware engines across every brand

The bug class behind the Odyssey fix is now closed everywhere: we audited every rule that maps a model name to an engine and found dozens written for one generation but applied to all of them. The worst offenders - Jeep Cherokee, VW Jetta, Chevy Tahoe and Suburban among others - now carry verified generation ladders or honestly say nothing, and a new global safeguard automatically catches any engine claim that contradicts the engine's own production years, for every make, forever.

Full security and reliability audit of the decoder and website

Nine independent audit passes hammered every page, endpoint and decoding layer with over 1,500 adversarial requests. The core held up impressively - zero crashes, authentication solid, sub-second decode times under load. What the audit did find is now fixed: hardened input handling and new browser security headers, a fuel-type edge case where two data layers could disagree, two slow public endpoints made instant, stronger privacy for decode history, and smarter year handling for brands that don't encode the year in the VIN - eliminating seven wrong years from our benchmark in one change.

373 modern engines freed from a hidden date trap

A registration-card investigation on a Malaysian-assembled Renault uncovered a subtle data bug: engines documented as 'in production since year X' had been stored as if they stopped that same year, silently blocking 373 modern engines from ever being shown on newer cars. All 373 are now correctly open-ended, year arbitration got smarter about which brands truly encode a year in the VIN, and the reported vehicle now decodes perfectly from its official registration record.

Infrastructure upgrade: 10x faster data backbone

The decoder's entire knowledge base - over 550 collections including 300 million manufacturer records - now runs on dedicated high-performance infrastructure co-located with the decoder itself. Data lookups that previously crossed the internet now happen in microseconds on the same machine, making decodes faster and immune to third-party cloud slowdowns.

The decoder now learns directly from manufacturer factory records

We connected millions of official factory build records (Nissan, Subaru, Mazda, Volkswagen Group) straight into the decoder's brain. It learned over 10,000 new VIN patterns from the people who built the cars.

• Nissan & Infiniti: chassis and platform identified from 78 million factory records
• Volkswagen, Audi and Skoda: factory engine codes now resolvable for many cars
• Subaru: engine size and fuel type from the official parts catalogue
• Mazda: build-year ranges straight from production data

Five more manufacturers learned: BMW, Honda, Renault, BYD, Mitsubishi

The decoder's factory knowledge grew from 10,000 to 56,000 learned patterns in one day. BMW type codes, Honda's per-market engine ladders, Renault's own decoding rules, BYD prefixes and Mitsubishi production records now resolve directly from the VIN.

• BMW: 5,861 factory type codes map the VIN to series, model and engine family
• Honda: factory engine and gearbox per VIN prefix, worldwide markets
• Renault: the manufacturer's own decode rules, learned verbatim
• Four more cars in our verification library now decode their exact engine correctly

Fourteen more engines right - learned from 160 million parts records

We distilled the world's parts catalogues into an engine dictionary: which engine belongs to which chassis code, straight from manufacturer data. Fourteen more cars in our verification library now show their exact engine - with zero new errors.

• BMW: over 1,000 factory type codes now name their engine family
• Nissan chassis codes resolve their engines (S13 to CA18DT, Note to HR12DE)
• Malaysian registration cards added as document-grade evidence
• Everything re-learns automatically every week

Japanese homologation codes now power fuel and engine detection

Vehicles imported from Japan carry government type-designation codes, and our decoder now reads their official grammar: it can tell petrol, diesel, hybrid, plug-in hybrid and electric apart directly from the designation prefix, and for Renault models it reads the exact engine code embedded in the designation itself. Every new mapping was verified against official Japanese market specifications before going live.

Model-year logic validated against 60 million factory records

We tested our model-year decoding against 60 million manufacturer build records. The industry-standard year encoding was confirmed for 92 manufacturer codes with over 98% agreement, and we mapped exactly which manufacturers do not encode a year in the VIN, so the decoder never guesses where the data is not there.

Factory records now correct mislabeled Renault models

When a Renault VIN arrives with a wrong model name attached from a polluted data source, the decoder now recognizes the conflict and serves the model that Renault's own factory grammar proves for that VIN - while carefully preserving trim variants and verified data.

Independent audit hardens our accuracy benchmark

We ran a full independent audit of how the decoder learns, and raised the bar: our internal accuracy benchmark is now strictly separated from all training data, every learning pipeline enforces that separation automatically, and a handful of under-evidenced mappings were retired until stronger independent proof arrives. Where retiring a mapping could have exposed stale third-party data, new coherence safeguards now catch it - so a petrol car can never be served a diesel engine. Accuracy numbers we report are now provably earned, never memorized.

Volkswagen Group VINs now decode from the factory's own build records

For 1.5 million Volkswagen, Audi, SKODA, SEAT and Porsche vehicles, the decoder now consults the manufacturer's own production record for that exact car - returning the precise factory engine code, model year and even the build date, instead of an educated guess. When your VIN is in the factory books, you get the factory answer.

Renault years now bounded by the factory's production calendar

Renault encodes its model type in every VIN, and we now pair that with the manufacturer's own production calendar for each type - nearly 5,800 model variants gained factory build-year windows. When a VIN's year letter claims a year the factory provably never built that model, the decoder now says so honestly instead of passing the error through.

Every data source re-audited field by field

We audited every data source the decoder learns from, field by field, and recovered value that earlier pipelines had silently dropped: factory build-year windows now cover all Nissan and Subaru learned keys, Honda coverage quadrupled by handling the factory's wildcard notation, BYD coverage doubled, Australian displacement and body-style evidence turned back on across 200,000 listings, and several subtle data-reading bugs were fixed - including one that could label hybrid vehicles as petrol.

Engine codes deduced from the specs

When the specs on file (size, cylinders, fuel, power) match exactly one engine ever made for that brand, the decoder now names it - verified against real registration documents before release.

• More engine codes shown, with zero new wrong answers on our 659-document verification library
• Applies everywhere: from a Lamborghini's V10 to everyday hatchbacks

Rolls-Royce and Mercedes-Maybach: full coverage

Ghost, Phantom, Cullinan, and every modern Maybach (S550 to S680, GLS600) now decode with the exact factory engine, power and torque.

• A Cullinan that used to show as a 'Phantom convertible' now decodes perfectly
• Maybach S680's 6.0 V12 identified with 603 hp and 900 Nm

We'd rather show a blank than a wrong answer

New safeguards stop one mislabeled foreign record from contaminating a whole model family - the error that once showed petrol cars as diesels.

• A single record from abroad can no longer overwrite what we know about your car
• Engines that are impossible for a car's year are automatically rejected
• When a VIN genuinely covers several engines, we say so instead of guessing

Fifteen brand families upgraded in one month

Infiniti, Acura, classic Mercedes and AMG, Opel, European and Thai-built Fords, Land Rover, Lada and more - each fixed for the whole model family, permanently.

• Renamed models (QX56 to QX80, G37 to Q60) now resolve to the one correct badge for your exact car
• Classic AMG and W124 Mercedes decode from the factory model code
• Every fix is locked by automatic tests so it can never silently break