A complete Base vector always carries eight metrics. Four feed the Exploitability sub-score and four feed Impact (Scope sits with the impact side because it changes how impact is calculated).

Exploitability metrics

Attack Vector (AV) is how close the attacker must be: N Network (reachable from anywhere), A Adjacent (same local network segment), L Local (needs local access or a logged-in session), or P Physical (needs to touch the device). Network scores highest.

Attack Complexity (AC) is whether conditions outside the attacker's control must line up: L Low (repeatable at will) or H High (needs a race, a specific configuration, or reconnaissance). Low scores higher.

Privileges Required (PR) is what access the attacker must already have: N None, L Low (an ordinary user), or H High (administrative). None scores highest.

User Interaction (UI) is whether a victim must do something: N None or R Required. None scores higher.

Scope

Scope (S) is the most misunderstood metric and the only one that changes the shape of the formula. It is U Unchanged or C Changed. Scope is Changed when a successful exploit affects resources beyond the security authority of the vulnerable component, for example a flaw in a sandboxed process that lets the attacker act on the host, or an injection in one virtual host that reaches another. A scope change raises the score and is what allows a vector to reach 10.0. Importantly, when Scope is Changed the Privileges Required weights also increase, because crossing a trust boundary is more serious.

Impact metrics

Confidentiality (C), Integrity (I), and Availability (A) each take H High, L Low, or N None. High means total loss (all data disclosed, or the system fully unavailable); Low means limited or constrained loss; None means no effect on that property. These three are symmetric in the formula, and the impact sub-score rises steeply as more of them move to High.

Reading the balance

The Base score is a tension between the two families. A flaw can be trivially reachable but score low because it barely touches confidentiality, integrity, or availability, and a flaw with total impact can score moderately if it needs high privileges, physical access, or a difficult race. Decoding a vector metric by metric, rather than reading only the final number, is what makes a score defensible when someone challenges it.