Methodology – how Caropen's figures are computed

Caropen's figures are computed from open technical-inspection data with the rules below. The same rules apply to every make and model, with no hidden weights and no manual corrections. When a rule changes, this page changes with the computation.

Common rules

• Vehicle – one registry pseudonym (tp_id), stable across years. Plates and VINs are hashed in the data, and we never reverse them.
• Main inspection – the scheduled technical inspection. Repeat inspections after a failure and document issues for a newly registered vehicle are not main inspections.
• Duplicates. The source sometimes publishes the same inspection twice. Rows equal on vehicle, week, inspection type, odometer reading and verdict count once.
• Plausible odometer reading – present and between 500 and 1,500,000 km inclusive. Other values are treated as typing errors and not used in the calculations.
• Age – the calendar year of the inspection minus the year of manufacture. A car made in 2014 and inspected in any week of 2022 is 8 years old. Ages below 0 or above 50 are treated as errors.
• Age bands: 2–3, 4–5, 6–7, 8–9, 10–11 and 12+ years. Cars under 2 years belong to no band: they have no comparable inspection history yet.
• Make and model are normalised names (e.g. "VW", "VOLKSWAGEN-VW" → Volkswagen). Registry spellings that map to no model count in national and municipality figures, but under no model.
• Municipality is where the inspection station is, not where the owner lives.

First-time pass rate

The first-time pass rate is the share of main inspections in a group recorded as passed. Every main inspection counts, including the rare case of a second full main inspection in the same year after a failure (2.7 % of vehicle-years; checked on a sample, it moves the rate by less than 0.1 percentage points).

Groups: model × year of manufacture, model, make, municipality × inspection year, country × inspection year, class × fuel × age, and model × age (the pass-by-age curve).

These are descriptive statistics: they show how cars of a model fared at inspections in Lithuania, not why – maintenance, mileage and use are not accounted for. That is why the sample size (n), the confidence interval and the class average at the same age are always shown next to the rate.

Confidence interval

Every pass rate comes with its 95 % Wilson score confidence interval. For a rate *p* over *n* inspections and z = 1.96:

• centre = (p + z²/2n) ÷ (1 + z²/n)
• half-width = z · √(p(1 − p)/n + z²/4n²) ÷ (1 + z²/n)
• interval = centre − half-width … centre + half-width

The true rate of the group lies in the interval with 95 % confidence. It is wide for small groups, which is why small groups are not ranked against each other on the rate alone.

Publication thresholds

A group below its threshold is computed but not published (or shown without search-engine indexing).

Class average

Every model page is compared with its class average: all main inspections of passenger cars (M1) and off-road passenger cars (M1G), each class separately, by age, for all fuels together and per fuel. A model-year page shows the M1 all-fuels average at the cohort's current age – the newest data year minus the year of manufacture.

Verdict and fuel mix

For the same groups, the number of main inspections per verdict: no defects, minor, major and dangerous defects. The fuel mix is the group's share of inspections per fuel type.

Mileage at age

What odometer reading is normal for a car of a given age:

1. For each vehicle and calendar year, take the last main inspection with a plausible reading.
2. Age = that calendar year − year of manufacture.
3. For model × year × age, model × age and class × fuel × age, compute the 10th, 25th, 50th (median), 75th and 90th percentiles with linear interpolation, rounded to a kilometre.
4. An age is published when it has at least 50 vehicles. When a model has too few, the mileage check uses the class × fuel × age figures and interpolates between ages.

Annual mileage

How far cars actually drive per year, from pairs of consecutive main inspections:

1. Take the vehicle's main inspections with a plausible reading, in order.
2. For each consecutive pair: annual mileage = (later reading − earlier reading) ÷ years between the inspection weeks (days ÷ 365.25).
3. Drop pairs less than 0.75 years apart and pairs whose difference is zero or negative.
4. Publish the median of the remaining pairs per model and year, model, make and municipality when there are at least 50 pairs. A pair counts towards the model, make and municipality of the later inspection.

Odometer inconsistencies

A vehicle is flagged when a later main inspection records a plausible reading lower than the highest plausible reading of any earlier main inspection by

• more than 1,000 km, and
• more than 2 % of that earlier highest reading.

Both conditions must hold: a drop of exactly 1,000 km or exactly 2 % does not flag. The comparison is with the running maximum, not just the previous inspection, so a reading that creeps down over several years is caught once the total drop exceeds the tolerance. Repeat inspections, duplicates and implausible readings are ignored.

Why the tolerance. A repeat visit can record a few kilometres less than the visit before it, and inspectors occasionally mistype a digit. 1,000 km and 2 % absorb this noise.

What the indicator cannot detect:

• An odometer or instrument-cluster replacement looks exactly like a rollback and cannot be told apart from one in this data.
• A rollback made before a vehicle's first inspection in Lithuania (for example before import) is invisible.
• A small rollback – within the tolerance – or one that has "grown back" past the earlier maximum by the next inspection goes unnoticed.

The indicator therefore describes readings that do not add up, never proven fraud. It is published only as the share of inconsistencies in a group – never for an individual car. Share = flagged vehicles ÷ vehicles with at least two plausible main readings (the only ones for which the indicator can fire). Groups: model × year, make, municipality of the last inspection and year of the last inspection (nationally).

Emissions

Per model × year and per model, from main inspections with a measurement:

• petrol (petrol, petrol + LPG, petrol hybrid): carbon monoxide at idle (CO, % vol) and at 2,000 rpm (CO 2000);
• diesel (diesel, diesel hybrid): smoke opacity (m⁻¹).

For each measure with at least 50 measurements: the median, the 90th percentile and the share above a limit. The limits – CO above 0.5 % vol and smoke above 1.5 m⁻¹ – are provisional: the legal limit depends on the engine's emission standard, which the data does not record, so these are conservative benchmarks, not the station's verdict. More in CO and smoke limits.

Municipality figures and age adjustment

Per municipality of the inspection station, per year and for all years:

• the raw first-time pass rate with its interval;
• the age-adjusted pass rate – what the municipality's rate would be with the national age mix, so municipalities with older fleets are not penalised for their age:

1. Split the municipality's main inspections into the age bands, plus one group for cars under 2 years or of unknown age.
2. Expected rate = Σ over groups (the municipality's share of its inspections in the group × the national rate of that group in the same year).
3. Adjusted rate = raw rate − expected rate + national raw rate, limited to 0 … 1.

Also shown: the number of stations, the median age of inspected cars, the most inspected models, the annual mileage median and the odometer-inconsistency share.

Registered fleet

From the quarterly file of every registered vehicle published by VĮ Regitra. It has no vehicle identifier, so fleet and inspection figures meet only on make, model family and municipality – never car by car.

• Edition – the register as of the first night of a quarter; the edition date is that quarter's first day (the file published on 2026-07-03 is edition 2026-07-01). Figures of earlier editions are kept as published.
• Population – passenger cars (M1 and M1G) allowed in public traffic. Deregistered cars and cars withdrawn from traffic are left out.
• Anonymised rows (make and model withheld because they would identify a person) count in the national and municipality figures but under no make or model.
• Municipality is the holder's registered address, not the inspection station, so a municipality's fleet and the cars inspected at its stations are different groups.
• Fleet share – the make's or model's registered cars as a share of all passenger cars of the same edition.
• Median age – years from the first registration anywhere to the edition date (the file's year of manufacture is almost always empty).
• Powertrain: battery electric (BEV – the electric-vehicle flag, or electricity as the only fuel), plug-in hybrid (PHEV – hybrid category OVC-HEV), hybrid (HEV – NOVC-HEV, or electricity as one of several fuels), fuel cell (FCEV) and combustion engine. Electric share is the BEV share.
• Published for a make, model family or municipality with at least 30 registered cars; national figures always.
• First registrations by quarter – cars first registered in Lithuania (new or imported) in each quarter since 2015, with their BEV, PHEV and HEV shares. They are counted from one edition, so cars already deregistered are missing from earlier quarters.

On make and model pages the registered fuel mix is compared with the inspected one: the registered fleet includes new cars not yet due for an inspection, the inspections only the cars that came to a station. More on Electric cars in Lithuania.

Rankings

The reliability ranking is an annual edition; earlier editions are kept as published.

• Passenger cars only (M1 and M1G).
• A cell is one model in one age band, over main inspections of all inspection years, for all fuels and separately for petrol and diesel.
• Only cells with at least 100 main inspections are ranked.
• Within each band, cells are ranked by first-time pass rate; equal rates share a place. Each row shows the rate, its interval and n.
• No composite score with hidden weights: each age band is ranked on its own.

A year's best and worst models are the ten M1/M1G models with the highest and lowest rate among those with at least 500 main inspections that year. Similar models are up to five models of the same class whose median year of manufacture is within 3 years and whose rate is closest.

Refresh and sanity checks

Data is loaded weekly. All figures are recomputed together and replace the previous ones only after the checks pass. A refresh is rejected (the previous figures stay) when:

• the national first-time pass rate is outside 40–90 %;
• any group has more inspections than the whole country;
• the national odometer-inconsistency share is outside 0.5–25 %;
• on a full dataset, fewer than 2,000 model × year groups are publishable.

Why stations are anonymous

Station codes are hashed in the source data. We do not reverse them and do not compare stations: a station's results depend not only on its work but on the age, models and owners of the cars that come in, and one station's sample is often too small for a sound conclusion. So we compare municipalities only – with the age-adjusted rate – and show just the number of stations.

How to cite

When citing, name Caropen.lt and the original source, for example:

• *Caropen.lt, based on State Data Agency / Transeksta technical inspection data (data.gov.lt, CC BY 4.0), data as of September 2026.*

Link to the specific page, give the sample size (n) with the figure and, where possible, the interval. Call odometer inconsistencies "the share of odometer inconsistencies", not fraud. Licence terms and the attribution text: Data and licence.