Methodology
How the census counts
Every figure on this site is computed from public data, the same way every day, and says which sources it uses. This page defines the words those figures rely on and where each date comes from.
The daily run
- Fetch. Each source is downloaded into a new snapshot with a conditional request (unchanged files are not downloaded again). The snapshot replaces the previous one only if the download finished and the number of records did not fall by more than 20% (core sources) or 50% (others). Otherwise the previous snapshot is kept and the source is marked out of date.
- Load. Snapshots are loaded into a DuckDB database, one table per source.
- Join. Every source is joined on the CVE ID into one row per published CVE and one row per event in its life (see events).
- Export. The figures, one file per CVE shard and per entity, and the open data are written; the site is rebuilt from them.
The data on this site is as of 16 Nov 2026, the latest CVE publication date in the CVE List snapshot.
Definitions
CVE
A record in the CVE Program's CVE List V5 with state PUBLISHED. Rejected records are counted only where a chart says so. The publication date is the record's datePublished; the reservation date is dateReserved. The year of a CVE is its publication year, not the year in its ID, unless a chart says “by ID year”.
Exploited
A CVE is exploited if it is on at least one of: CISA's Known Exploited Vulnerabilities catalog, ENISA's EU Vulnerability Database list of exploited vulnerabilities, VulnCheck's KEV (when configured), or Google Project Zero's “0day In the Wild” tracker. Nothing else counts: a public proof of concept, a high EPSS score or an SSVC “active” decision do not make a CVE exploited here. The first exploited date is the earliest date any of these lists gives (CISA: date added; ENISA: exploited since; Project Zero: date discovered). These lists learn of exploitation late, so the true first attack is often earlier.
CISA's catalog started on 3 November 2021 and took in hundreds of older CVEs at once. Time-to-exploitation figures therefore only use CVEs published from that day.
Zero-day
An exploited CVE whose first exploited date is on or before its publication date.
Public exploit
A CVE has a public exploit if Exploit-DB lists an exploit for it, Metasploit has a module that references it, or at least one GitHub repository in nomi-sec's PoC-in-GitHub index claims a proof of concept for it. The first public exploit date is the earliest Exploit-DB publication date or GitHub repository creation date (Metasploit modules carry no creation date). GitHub claims are not verified; some repositories are fakes or malware. Nuclei templates are detection, not exploitation, and are shown separately.
Severity
The first available CVSS base score in this order: the CNA's own, CISA's (ADP container), NVD's. The highest CVSS version given is used (4.0, then 3.1, 3.0, 2.0). The severity label is the one given with that score.
Weakness
The CWE IDs the CNA names; if none, CISA's; if none, NVD's. A CVE can name several CWEs and is counted once for each.
Vendors and researchers
Vendors are taken from the affected list the CNA writes. Names that differ only in letter case, spacing or a company suffix (Inc., Corp., Ltd., GmbH and similar) are merged; nothing else is. Researchers are the credits of the CVE record, exactly as written, merged only on case and spacing. A credited name can be a person, a team, a company or a tool; names are shown as the CNA wrote them.
EPSS
FIRST's daily probability of exploitation activity in the next 30 days. The census keeps the scores of the first day of every month since May 2021 and every day of the last 35 days, to test how scores behaved before CVEs were listed as exploited.
Events
Each CVE has a list of dated events, each from one source. Only dates a source states are used.
| Event | Date used |
|---|---|
| ID reserved, CVE published | CVE List V5 dateReserved, datePublished |
| In NVD | NVD published |
| First PoC on GitHub | Creation date of the earliest repository |
| Exploit-DB exploit | Exploit-DB date_published |
| Nuclei template | First commit adding the template (history starts April 2020) |
| CISA KEV, ENISA, VulnCheck | Date added / exploited since |
| Zero-day found, patched | Project Zero's discovery and patch dates |
| ZDI advisory | ZDI publication date |
| OSV advisory; Debian, Ubuntu, Red Hat fix | OSV published of the first advisory that has a fixed version |
| Named in a paper | arXiv submission date of the first cs.CR paper naming the CVE in its abstract |
| HackerOne disclosure | Disclosure date of the first public report naming the CVE |
Limits
- Only public data. Exploitation that no list records, private exploits and undisclosed bug bounties are invisible here.
- CVE records are written by hundreds of CNAs with different habits. Comparing CNAs compares their records as much as their software.
- NVD's status is as of the latest NVD data feed and can lag the API by a day.
- Matching papers and bounty reports to CVEs relies on the CVE ID appearing in the text, so many links are missed.
- “Last 12 months” always ends on the data date above, not on the day you read this.
Questions, corrections or a source we should add: contact@webcracy.ch.