The mountains, hands off the wheel. Town, into oncoming traffic.
It went up and down to Lares on roads where sometimes only one car fits, and I never had to take the wheel. But twice in the same day it was heading into oncoming traffic. That is a severe failure, and on its own it holds the score at 6.0.
The home test.
Two spots on the way out of my house, the same ones every version. The dip, which no version has taken at the right speed yet. And the crossing onto a four-lane avenue with constant traffic, split by a guardrail. This shows whether the new version handles them, without letting them sink the score for everything else.
v13.2.9 and v14.3.2 were measured as a success rate over many exits; from v14.3.3 each evaluated trip records how it went. “Partly” means it managed, but not under the condition that matters (for example, with no traffic).
The dip is a capability flag: it costs points within its category but does not cap the score. The exit also costs points if the car struggles to pull away or merge, but an unsafe left turn or stopping mid-crossing is a safety flag and does cap it. How it is calculated →
v···
This table repeats identically every review. Same criteria, version after version. For an interactive view with any pair of versions, open Compare →
Each criterion runs 1–10 unless otherwise noted. Through v14.3.5 the scores are editorial; from v14.3.6 (right of the dashed line) they come from the 65-question FSD Score. The two scales are not directly comparable. “—” means that criterion does not exist in that system.
Same exit. Same table. Next review.
With every new version I repeat the same format — the house exit, the same criteria, the same table — so progress stays verifiable.