How F1 changed · Chapter 2 of 3

Why cars don't finish

Every race entry in F1 history, bucketed into the five stories that actually matter: finished, something broke, something hit, retired anyway, or a rule kept you out. The shape of the last seventy years is engineering catching up with ambition — punctuated by the eras where it didn't.

Data: race-result status strings from Jolpica-F1, normalised into a consistent finish taxonomy.

The long arc: reliability wins

Sport-wide finish rate (anything classified as a finisher, lead-lap or lapped) rising from the mid-40s in the 1950s-60s to above 90% in the 2020s. Reliability, not aero, is the defining engineering story of modern F1.

The 1995-2010 plateau around 70% is the V10-era peak of "big engine, big aero, medium reliability". The step change into the hybrid era (>90% finish rate) is not that cars got safer — it's that power units became absurdly well-engineered.

Composition by decade

The same data decomposed into five narrative buckets: what fraction of entries in each decade ended how? The 1960s and 70s are a wall of mechanical brown — two out of every three cars failed to finish, almost half of those because something broke. The modern era is overwhelmingly green.

Stacked to 100% so the macro shifts are legible: mechanical failure collapsing from ~35% in the 1960s–80s to <5% today, and the steady background rate of accidents holding around 10% across every era. Other spikes in the 1980s capture the brief pre-qualifying "DNQ" regime — a rule that thinned fields by denying slow cars a grid slot. The 2020s bar reflects 2020–2022 only; see the coverage note below.

Each failure mode, on its own

Stacked views hide the individual arc of each cause — so we split them out. Each panel is one category's share of every race entry, season by season, on the same 0–60% scale. The colour is the mode's narrative role (brown mechanical, red accident, slate administrative); the trend line is the tl;dr of the panel.

Faceted 3-season rolling mean, 1950–2022. Mechanical shows the defining engineering story: a steady ~35% through the 1950s-80s, collapsing by the 2010s to almost nothing. Accident holds a stubborn 8-12% floor — the physical cost of wheel-to-wheel racing, largely invariant to era. Retired and Other are the long tails of bookkeeping: pre-qualifying in the late-80s, short-lived DNQ regimes, and the occasional mis-classified result.

A coverage break in 2023. From the 2023 season onwards, the source feed (Jolpica, and the FastF1 session-results mirror that proxies it) stopped coding specific retirement causes. Every non-finisher is now labelled as a bare "Retired", so "Engine", "Gearbox", "Collision", "Hydraulics" — the whole specific vocabulary — simply stop appearing in the data. That's why the two affected charts cap at 2022: the mechanical- and accident-share series from 2023+ would read as near-zero by artefact, not because the cars got better. Finish rates themselves are unaffected ("Finished" and "Lapped" are still coded correctly), so the long-arc chart at the top of the page runs to the present.

Methodology in brief. Jolpica's status field is free text — "Engine failure", "Collision damage", "+1 Lap" — and the vocabulary changes across decades. The pipeline normalises those strings into ten enumerated buckets via the dnf_category dbt macro, then this page rolls those into five reader-facing stories. The underlying mart stays one row per (season, category) with both raw counts and shares, so the visual layer is reusing published data rather than reclassifying results on the fly.