Data Sources & Methodology

How CVE Explore ingests, normalises, and enriches CVE records — sources, cadence, freshness guarantees, and the chain of trust for each enrichment.

Last updated: 2026-05-15

CVE Explore is a derivative aggregation. Every page is built from public upstream feeds; the source-of-record for each CVE remains NVD or MITRE. This page documents which feeds we ingest, how often we refresh them, and which source wins on conflicting fields.

1. Summary

We ingest seven upstream feeds on independent schedules, persist them to Postgres 17 as the source of truth, and project a derived search index to Meilisearch via a transactional outbox trigger. Every CVE row carries the timestamp of its most recent successful upstream synchronisation so callers can see exactly how fresh the page is.

2. Data sources

  • NVD — the National Vulnerability Database. The primary source for CVSS metrics, references, and CPE-style affected-product lists.
  • MITRE cvelistV5 — the JSON 5.x CVE feed published by MITRE. We use it to backfill records and to fill fields missing in NVD (descriptions, CWE references, vendor advisories).
  • CISA KEV — the Known Exploited Vulnerabilities catalog. Drives the "KEV" flag, the actively-exploited badge, the "known ransomware" signal, and the remediation-due-date field.
  • FIRST.org EPSS — the Exploit Prediction Scoring System. Provides the daily EPSS score and percentile for every CVE.
  • Metasploit modules — the open-source Rapid7 metasploit-framework repository. Module metadata is indexed so CVEs that already have a public, weaponised exploit are visible in search facets and on the detail page.
  • trickest CVE corpus — a curated, community-maintained corpus of CVE → PoC GitHub repo mappings.
  • nomi-sec / PoC-in-GitHub — a second curated CVE → PoC corpus we ingest in parallel. The two PoC corpora are deduplicated by repository URL.

3. Refresh cadence

  • NVD — every 15 minutes (recent window) plus a 3-hourly backfill pass.
  • MITRE cvelistV5 — every 30 minutes.
  • trickest, nomi-sec — every 2 hours.
  • Metasploit modules — every 2 hours, offset by 30 minutes.
  • CISA KEV — every 6 hours.
  • FIRST.org EPSS — daily at 02:15 UTC.

Each ingester runs in a bounded worker pool with per-CVE upsert transactions; a feed failure on one source does not block the others.

4. Source precedence

NVD beats cvelistV5 on score conflicts. cvelistV5 fills missing fields only — we never let a backfill source overwrite a field NVD has already populated. KEV, EPSS, Metasploit, trickest, and nomi-sec write to dedicated columns that never collide with the CVE-record columns NVD owns.

5. Enrichment & derivation

  • CVSS, CWE, and affected products — taken verbatim from upstream; we do not re-derive them.
  • Severity bucket (Critical / High / Medium / Low / None) — derived from the CVSS v3 base score using the FIRST scoring system's canonical thresholds.
  • PoC presence — a CVE has "public PoCs" if it has at least one row in the deduplicated PoC table (sources: trickest + nomi-sec + Metasploit).
  • Trending — the rolling 14-day surface, weighted by sum of PoC repository stars. Approximate by design: the trending list reads only the top-5,000 starred PoCs and filters in memory, so a 0-star PoC published an hour ago will not appear regardless of urgency.
  • AI-generated content — when an operator opts in (AI_PROVIDER is set), CVE Explore can additionally generate plain-English summaries and pre-validated proof-of-concept stubs. Both are clearly labelled and never replace the canonical description or upstream PoCs. Sandbox execution against a real target is operator-gated and disabled by default.

6. Freshness guarantees

Every CVE detail page surfaces the timestamp of the most recent upstream synchronisation in its provenance footer. We do not promise sub-minute freshness — NVD's own publication cadence is the lower bound — but the recent-window NVD ingester runs every 15 minutes, so a CVE typically appears within 15-30 minutes of NVD publishing it.

7. Deletions & rejections

When NVD or MITRE rejects a CVE, the record on CVE Explore is marked rejected and surfaced with a warning banner; the record is not removed. Public PoCs from third-party corpora may be delisted at the source maintainer's request — see the responsible-use policy for the takedown process.

8. Licensing & attribution

The aggregated data set is licensed under CC-BY 4.0. Attribute "CVE Explore" and link back to the canonical CVE Explore URL of any record you reproduce.

The upstream sources retain their own licensing terms — NVD and MITRE records are US Government works; CISA KEV is a US Government work; Metasploit module metadata is BSD-licensed; the curated PoC corpora carry their respective upstream licences.

9. Corrections

Spotted a record that disagrees with NVD or MITRE? Email corrections@cve-explore.com with the CVE id and a link to the upstream source. We do not edit CVE descriptions ourselves — corrections must flow upstream first and we re-ingest on the next refresh.

Terms Privacy Responsible use Questions? legal@cve-explore.com