CSV Feed

FIRST publishes a fresh CSV dump of every EPSS score every day at https://epss.cyentia.com/epss_scores-YYYY-MM-DD.csv.gz.

The file format is a single #-prefixed metadata line, a header row, and the score rows:

#model_version:v2025.03.14,score_date:2026-05-18T00:00:00Z
cve,epss,percentile
CVE-1999-0001,0.0046,0.7385
CVE-1999-0002,0.04525,0.92176
...

EPSS::CSV parses both the gzipped and uncompressed variants.

Stream-by-row (recommended for the full feed)

The current feed is ~240k rows. EPSS::CSV.each_score streams them without buffering:

require "epss"

File.open("epss_scores-2026-05-18.csv.gz") do |io|
  EPSS::CSV.each_score(io) do |score|
    db.upsert(score.cve, score.epss, score.percentile)
  end
end

The gzip magic bytes are auto-detected — pass either the compressed or the decompressed file. A String, an IO, or a Path all work as input.

Load the whole feed

If you want metadata + all rows in memory, use EPSS::CSV.parse:

feed = EPSS::CSV.parse(File.read("epss_scores-2026-05-18.csv.gz"))
feed.metadata.model_version  # => "v2025.03.14"
feed.metadata.score_date     # => Time
feed.scores.size             # => 240000+

Feed is Enumerable(Score), so iteration, filtering, and reduction work directly:

feed.select { |s| s.epss > 0.5 }
   .sort.reverse
   .first(20)
   .each { |s| puts s }

Date handling

Every row in Feed#scores is stamped with the score_date from the file's metadata header (rows in the feed itself don't carry per-row dates). If you parse a feed that omits the # header — for example, a hand-trimmed local copy — Score#date is nil and you'll need to set it yourself.

Errors

EPSS::CSV.parse raises EPSS::ParseError when:

• The header row is missing cve, epss, or percentile.
• A row has fewer columns than the header declared.
• A numeric value can't be coerced to Float64.
• A row's date field is non-empty but doesn't match %Y-%m-%d.

The streaming form (each_score) raises the same errors at the point of the bad row, so a corrupt mid-file row stops iteration.