# CVSS v3.1 scores how severe a flaw is in itself, not how likely anyone is to exploit it

SCORING SYSTEM, read 10 Nov 2026

**The short answer.** CVSS v3.1 is the FIRST severity score from 0.0 to 10.0, built from eight base metrics and a published formula. It measures technical severity, not exploitation. In the KEV Lag Ledger, 313 of 318 listings carry a v3.1 score, and 10.4 percent of the 318 score below 7.0.

- **Maintained by:** FIRST, through the CVSS Special Interest Group
- **Scale:** 0.0 to 10.0; None 0.0, Low 0.1 to 3.9, Medium 4.0 to 6.9, High 7.0 to 8.9, Critical 9.0 to 10.0
- **Current version:** v3.1 (kept as an archive on first.org; v4.0 followed on 2023-11-01)
- **Updated:** Static specification; scores are set per CVE by the CNA or NVD
- **Answers:** The severity of a flaw if it is exploited, in the reasonable worst case

## What the score is

CVSS v3.1 is a published scoring method. The specification at first.org states that its outputs are "numerical scores indicating the severity of a vulnerability relative to other vulnerabilities" and that it has three metric groups: Base, Temporal and Environmental. The Base Score reflects intrinsic characteristics that stay constant over time and assumes the reasonable worst case impact across deployed environments.

A v3.1 score is therefore a statement about the flaw, not about an attacker or an organization. Two vulnerabilities with the same vector string score the same whether one is under mass exploitation and the other has never been seen outside a lab. That is the design, and it is also the reason a program cannot rank by this number alone.

## The formula

The Base Score depends on three sub-formulas, quoted here from section 7.1 of the specification. ISS = 1 - [(1 - Confidentiality) x (1 - Integrity) x (1 - Availability)]. Impact is 6.42 x ISS when Scope is Unchanged, and 7.52 x (ISS - 0.029) - 3.25 x (ISS - 0.02)^15 when Scope is Changed. Exploitability = 8.22 x AttackVector x AttackComplexity x PrivilegesRequired x UserInteraction.

BaseScore is 0 if Impact is 0 or less. Otherwise, with Scope Unchanged, it is Roundup(Minimum[(Impact + Exploitability), 10]); with Scope Changed, it is Roundup(Minimum[1.08 x (Impact + Exploitability), 10]). Roundup returns the smallest number, to one decimal place, that is equal to or higher than its input.

Worked example from the Table 16 constants: AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H gives ISS = 1 - 0.44 x 0.44 x 0.44 = 0.9148, Impact = 6.42 x 0.9148 = 5.873, Exploitability = 8.22 x 0.85 x 0.77 x 0.85 x 0.85 = 3.887, sum 9.76, rounded up to 9.8.

Appendix A of the specification addresses a practical trap. Roundup is defined as the smallest number, to one decimal place, that is equal to or higher than its input, and the appendix gives advice to implementers on avoiding small floating point inaccuracies that can push a result up by 0.1. An analyst who re-derives scores in a spreadsheet to answer an auditor should follow that advice, or the check will disagree with the scanner on a few vectors.

| Metric | Values and constants (Table 16) |
| --- | --- |
| Attack Vector | Network 0.85, Adjacent 0.62, Local 0.55, Physical 0.2 |
| Attack Complexity | Low 0.77, High 0.44 |
| Privileges Required | None 0.85, Low 0.62 (0.68 if Scope Changed), High 0.27 (0.5 if Scope Changed) |
| User Interaction | None 0.85, Required 0.62 |
| Confidentiality, Integrity, Availability | High 0.56, Low 0.22, None 0 |

## Temporal and environmental layers

The specification adds two optional layers. TemporalScore = Roundup(BaseScore x ExploitCodeMaturity x RemediationLevel x ReportConfidence), so a temporal score can only hold the base score or lower it. The Environmental Score recomputes the formula with Modified Base metrics and security requirements that the consumer sets.

The ledger on this site uses the NVD base score, so the layers are outside its figures. The exploit code maturity input is the only place where v3.1 touches exploitation at all, and it is a value someone has to set and keep current.

## What the KEV Lag Ledger shows about it

The KEV Lag Ledger covers the 318 CISA KEV additions from 2025-11-10 to 2026-11-09. Each has a CVSS score taken from the NVD record, v3.1 first. Of the 318, 313 are v3.1 and 5 are v3.0. The median score is 8.8, 45.9 percent (146) score 9.0 or above, and 285 (89.6 percent) score 7.0 or above.

That leaves 33 listings, 10.4 percent, with a score below 7.0. A program that treated CVSS High and Critical as the whole queue would have held back one in ten of the vulnerabilities CISA confirmed as exploited. It would also have queued far more than it needed: the same ledger shows that of the 123 listings scored 9.0 or above that had an EPSS value on listing day, 88 (71.5 percent) sat below 0.1.

The distribution is also narrow at the bottom. Only one of the 318 listings has a score below 4.0. The 33 below 7.0 are therefore almost all Medium. A policy that sets the remediation floor at High would have excluded those 33 from its fastest queue, while a policy that adds KEV membership as an override would have captured them on the day CISA listed them.

## Where it fits in a ranking

Use v3.1 for what it states: the technical consequence and the conditions of the attack. It is a sound input to a severity floor, to a service level by asset tier, and to the question an auditor asks first, which is how the score was derived. It is a weak input to the question of what attackers are doing this week.

FIRST publishes both specifications, so a program has to decide which one its policy cites. State the version in the policy text, because scores under the two methods are different measurements. See the CVSS v3.1 against v4.0 page for how they differ.

For the auditor, the record of a CVSS-based decision is short: the version, the vector string, the computed value, the source of the vector (NVD or a named CNA), and the date it was read. With those five fields, anyone can recompute the value from the Table 16 constants and confirm it.

## Limits to state in the policy

First, the score is assigned by people. The specification itself describes the qualitative rating bands as optional, and states that there is no requirement to publish them. Second, the base metrics assume the worst reasonable case, so they cannot account for a deployment where the vulnerable function is disabled.

Third, v3.1 is a frozen version. FIRST now documents v4.0 with a new Threat group and nomenclature (CVSS-B, CVSS-BT, CVSS-BE, CVSS-BTE) that v3.1 does not use. Labeling a number as v3.1 Base, with the vector string beside it, removes most arguments about what it means.

A fourth limit concerns change. A CNA or the NVD can rescore a CVE after publication, so the value read on one date can differ from the value read later. Store the read date, as this site does for every figure.

## Comparisons with this system

- [EPSS vs CVSS](/compare/epss-vs-cvss/)
- [CVSS vs KEV](/compare/cvss-vs-kev/)
- [SSVC vs CVSS](/compare/ssvc-vs-cvss/)
- [CVSS v3.1 vs v4.0](/compare/cvss-v3-1-vs-v4-0/)
- [CVSS vs EPSS vs KEV](/compare/cvss-vs-epss-vs-kev/)

## Sources

- FIRST, CVSS v3.1 specification document (sections 5, 7.1, 7.4, Table 16): https://www.first.org/cvss/v3.1/specification-document (read 10 Nov 2026)
- FIRST, CVSS v4.0 specification document (nomenclature, version history): https://www.first.org/cvss/v4-0/specification-document (read 10 Nov 2026)
- NVD news and updates: https://nvd.nist.gov/general/news (read 10 Nov 2026)

Last reviewed 10 Nov 2026. Canonical: https://epssvscvss.com/scores/cvss-v3-1/
