Skip to content
  1. Conglomerado
  2. Calaa Mind
  3. Consulting
  4. Codex

01.4 · Calaa Mind

Industry knowledge, verified and cross-referenced in a single corpus

Codex is the corpus built by Calaa Mind. It brings together the technical, regulatory and service history of the vehicles on the road, and checks it record by record. Calaa reasons on it and answers inside Cal-1.

Ask Calaa, in Cal-1

The corpus today

More than 35 million records and more than 275 million data points

Measured on 10 November 2026
35,182,693
records
277,604,017
data points
7
official registries in the US, the EU and the UK

What it contains

The technical, safety and regulatory history of the vehicle fleet, in a single corpus

Use

  • 681 MThe results of 681 million roadworthiness inspections · United Kingdom, 2005-2023
  • 150 M150 million registrations · 32 European countries, 2010-2022

Safety

  • 1.6 M1.6 million driver complaints · US, since 1995
  • 46,76646,766 safety alerts · EU, since 2005
  • 30,43830,438 recall campaigns · US, since 1966
  • 17,31317,313 crash tests
  • 5,3445,344 safety investigations

Vehicles and their standards

  • 50,24250,242 versions with technical specifications · 1984-2027
  • 26,12926,129 vehicles from 2,582 brands, since 1900
  • 1,5421,542 US and EU standards

Bars on a logarithmic scale: each step multiplies by ten.

Years covered

From 1900 to the 2027 models

Each dataset enters with the period it has. Those still open run to today.

127
years of automotive history
7
datasets, each with its own period
Vehicles and brands1900 →
Recall campaigns1966 →
Technical specifications1984–2027
Driver complaints1995 →
Safety alerts2005 →
Roadworthiness inspections2005–2023
Registrations2010–2022

Which vehicles

  • Cars
  • Motorcycles
  • Trucks
  • Buses
  • Tractors
  • Racing

Sources: 7 official registries in the US, the EU and the UK.

Findings

What only appears when the data is cross-referenced

Measured on 10 November 2026
Finding 0110 Nov 2026

Complaints do not announce recalls: they follow them

×1.6

In the months after a recall campaign, complaints about that part surge: 1.6 times more than for the rest of the car. Before the campaign they move no more than the others.

Measured on 10,910 campaigns (2012-2024) and more than 2 million complaints.
Finding 0210 Nov 2026

No early warning

4.9 % vs 6.0 %

We tried raising the alarm when complaints about a part doubled. It ended in a campaign 4.9 % of the time; without an alarm, 6.0 %. Complaints do not serve to get ahead: we checked before promising it.

Measured on 10,910 campaigns (2012-2024).
Finding 0310 Nov 2026

Most recalls are not requested by the regulator

1 in 18

Only 1 in 18 recall campaigns stems from an official investigation. The rest are announced by the manufacturer itself.

Measured on 10,910 campaigns (2012-2024).
Finding 0410 Nov 2026

From investigation to recall

110 days

When the regulator opens an investigation that ends in a campaign, the campaign arrives after 110 days, at the median. In half of the cases, between 39 and 235 days.

Measured on 605 campaigns.
Finding 0510 Nov 2026

One part, eight manufacturers

8 groups

A single supplier appears in recall campaigns from 8 different manufacturer groups in the last year. A fault in one part does not stay in one brand.

12 months to September 2026.
Finding 0610 Nov 2026

Shared suppliers

32

In the last year, 32 suppliers appeared in campaigns from two or more manufacturer groups, and 14 in three or more. We have measured it for ten years: between 20 and 33 each year.

12 months to September 2026; series since 2016.
Finding 0710 Nov 2026

An ever wider supply chain

+56 %

The number of distinct suppliers involved in recall campaigns rose from 93 in 2017 to 145 in 2024.

12-month windows, 2017 and 2024.
Finding 0810 Nov 2026

The end of an era

From 17 to 2

The supplier behind the largest wave of campaigns of the decade went through 17 manufacturer groups. In the last year, it appears in only 2.

Campaigns from 2015-2024 against the 12 months to September 2026.
Finding 0910 Nov 2026

The battery enters the scene

Cells

Battery cell manufacturers now appear as suppliers in recall campaigns from several manufacturer groups. The electric car is starting to leave its mark on safety records.

12 months to September 2026.
Finding 1010 Nov 2026

Who makes the part

6 in 10

In 6 out of 10 recall campaigns, the defective part was made by an outside supplier, not by the carmaker.

Measured on campaigns from 2015 to 2026.
Finding 1110 Nov 2026

One company, fifty names

More than 50

In official reports, the same supplier company is written in more than 50 different ways. Unifying them is what shows who is behind each fault.

Campaign reports from 2015 to 2026.
Finding 1210 Nov 2026

One in five fails

21.5 %

In the United Kingdom, more than 1 in 5 roadworthiness inspections ends in a fail: 146.7 million out of 681.7 million.

Inspections from 2005 to 2023.

01

The order of preference

Top to bottom. A figure from a lower rung only goes in when it does not exist on a higher one, and then it is marked.

  1. 01Own measurementWhat the group's product measures. It is the only source whose method is known in full, and also the one with the least coverage.
  2. 02Official bodyRegistration records, public statistics, sector bodies and regulators. Slow, but anyone can check it.
  3. 03Academic studyPublished method and peer review. The method is solid and the figure tends to be old.
  4. 04Research firmBroad coverage and a proprietary method. It is accepted by citing who publishes it, because it cannot be audited.
  5. 05Trade publicationFast and useful for spotting a movement. It is never used as the only source of a figure.

02

The five checks

Every figure goes through five checks and they are applied in that order. The first one to fail stops the rest.

  1. 01AuthorityWho publishes it, and with what standing on this particular matter. A manufacturer is an authority on its own registrations and is not one on its customers' satisfaction. Authority is not inherited across subjects.
  2. 02PrecisionThat the figure can be verified and comes with its unit and its period. A percentage with no base is the most frequent case. "Up 30%" means nothing unless it says more than what.
  3. 03ObjectivityWhat interests sit behind the publication. A study funded by whoever sells the solution it recommends is not discarded, but it goes in declared and cross-checked.
  4. 04CurrencyWhether it still holds today. In this sector shelf life is short, and an aftersales figure from before electric arrived describes a market that no longer exists.
  5. 05CoverageWhether it covers the whole matter or only the convenient part. A figure from one country presented as European, or from new vehicles presented as the parc, are the same error twice.

03

All three rules can be checked from outside

What is done when the material does not fit.

A disagreement between two sourcesIt is usually about definition before it is about the figure.Both are kept and which one was used for the calculation is stated. First it is checked whether the disagreement is real or one of definition, for instance a market with or without light commercial vehicles.
A source discarded wholeThree causes, and all three are about method.When it publishes a figure without saying how it obtained it and does not answer when asked. When whoever funds it has a direct interest in the result and does not declare it. And when on two occasions it has corrected a figure without notice.
A check that failsThe figure does not go in, and the question is marked as open.If there is no substitute, what it would take to close it is stated, rather than answering it with whatever figure was at hand.

04

Without the order in sight, every figure looks equally backed

A written rung can be argued with. A fifth-rung figure stops weighing the same as a first-rung one, and whoever receives it decides whether it serves what they are about to do. Without that mark, the worst figure in the database ends up holding up the conclusion.

The rung

A source with no published method goes in one rung below the one it would otherwise get.

The mark reaches the report; it does not stay in the database.
01ConsultingFour stretches and the method that holds them.