# Is a Ferrari actually built better than a Toyota? — methodology

Every figure in the video, the caption and the article comes from the files in this kit.

## The dataset

**NHTSA's recall flat file**: every US safety recall since 1966, one row per campaign x make x
model x model year, with the potential number of units affected on every row. Keyless:
https://static.nhtsa.gov/odi/ffdd/rcl/FLAT_RCL_POST_2010.zip (+ `FLAT_RCL_PRE_2010.zip`, field
dictionary `RCL.txt`). Last record in the file used: 2026-10-01.

`supercar-recall-campaigns.csv` holds every campaign behind the numbers (144 rows), with
its model years, its units, how many of those fall inside 2016-2021, and NHTSA's own defect text.

## The two counts the video compares

**1. How many recalls.** Campaigns touching model years 2016-2021: Toyota + Lexus **116**,
Ferrari **11**, Lamborghini **17**.

**2. How far a typical recall reaches.** For each campaign, the units it covers inside the window,
divided by one year of that brand's sales; the video quotes the **median** campaign:

| brand | median recall | share of a year's cars | one in |
|---|---|---|---|
| Toyota + Lexus | 4,508 cars | 0.19% | 523 |
| Lamborghini | 224 cars | 12.4% | 8.0 |
| Ferrari | 893 cars | 32.4% | 3.1 |

The video rounds Toyota's 1 in 523 to "one in every five hundred". That overstates
Toyota's reach by 4.6%, which works against the video's own point, so it stands.

## The denominators

`supercar-sales.csv`: every row carries its source URL and the exact sentence quoted from it.
**Ferrari does not report US sales**, only "Americas" shipments (US + Canada + Latin America), so
Ferrari's sales figure (16,554 cars, 2016-2021) is an UPPER bound on its US sales, and its
reach and per-car rates are LOWER bounds. That direction is conservative for the video's claim.
Lamborghini: USA deliveries. Toyota + Lexus: US sales (incl. Scion).

## The hook: Ferrari's recalls by part, cars built since 2005

`supercar-ferrari-parts.csv`: 34 Ferrari campaigns on 2005+ cars, each assigned to the part it
fixes. One part in the video's piles = one recall of that part. The airbags (9) make the biggest pile.
Every part is drawn at the same size, so a pile's size follows its count.

## The brake fluid cap

Campaign **22V536000** (reported 2022-07-26): a leaking brake fluid
reservoir cap, **23,555 Ferraris**, model years 2005-2022 (18 model years),
19 models: 430, 488 PISTA, 612, 612 SCAGLIETTI, 812, CALIFORNIA, CALIFORNIA T, F12 BERLINETTA, F12 TDF, F60 AMERICA, F8 SPIDER, F8 TRIBUTO, FF, GTC4LUSSO, GTC4LUSSO T, LAFERRARI, LAFERRARI APERTA, PORTOFINO, ROMA.
It shows the mechanism: a small maker shares the same parts across models and years, so one bad part
reaches a big slice of everything it built. The field of 23,555 cars in the video is that
campaign's count, exactly.

## What the video does NOT claim

- **Recalls per car.** Ferrari's recall notices per car sold (1.41) are higher than
  Toyota's (0.63), but campaigns that span the window's edges have to be split
  across model years (NHTSA doesn't publish units per model year), and the strict lower bounds
  (0.25 vs 0.245) tie. So the video never says it.
- **That a Ferrari is built worse.** Recalls are what a maker FILES, not how often a car fails. The
  video's point is the shape: a few recalls that each reach a big share of a small maker's cars.
- The red car on screen is a stand-in 458 model, and the 458 is NOT on the brake-cap recall
  (it's on 21V833000); it's never lit as one of the 23,555.

## Credits (CC-BY: attribution required)

- Ferrari 458 Italia model by **vicent091036** (three.js example `ferrari.glb`), licence unverifiable, used
  by owner override.
- Generic passenger car pack by **Comrade1280** (CC-BY 4.0): the lots and the field.
- "Toyota Auris wagon" photogrammetry scan by **thesidekick** (CC-BY 4.0), via Objaverse; licence plates covered.
- HDRI, grass and concrete from **Poly Haven** (CC0). Fonts: Inter, Instrument Serif (SIL OFL).
- Sound: Sonniss GDC Game Audio Bundle and Mixkit (royalty-free).
- The prancing-horse mark is Ferrari S.p.A.'s trademark, shown to identify the brand under discussion.
