NUMBERS GUIDE, read 10 Nov 2026
EPSS is a daily probability, so store the date and the percentile with every score
What the number is
FIRST states that EPSS estimates the probability a vulnerability will be exploited in the wild within the next 30 days, with a score between 0 and 1 for every publicly disclosed CVE, updated daily. The same FAQ states that CVSS and EPSS measure different things and that EPSS is not a complete risk score. It supplies the threat likelihood component only.
An EPSS of 0.02 means about two chances in a hundred over 30 days. It does not mean the flaw is minor, and it says nothing about what happens on the asset.
Probability and percentile are different statements
FIRST's article on bins puts it directly: probabilities convey a global sense of the threat of exploitation, while percentiles provide a relative, localized measure. A probability of 0.10 sits at about the 88th percentile, so 88% of scored CVEs are lower.
The same article recommends showing both, in a format such as 15.3% (92nd). Do the same in your tool and your reports. A percentile alone hides how small the underlying probability is, and a probability alone hides how rare it is.
| Statistic from the KEV Lag Ledger | Value |
|---|---|
| KEV additions with an EPSS score on the listing day | 260 of 318 |
| Median EPSS probability on listing day | 0.0066 (0.006555) |
| Median EPSS percentile on listing day | 0.55 |
| Share with probability below 0.10 | 75.8% (197 of 260) |
| Share with probability at or above 0.50 | 13.1% (34 of 260) |
Exploited CVEs often score low on the day they are listed
The ledger pairs each KEV addition with the EPSS value published on its dateAdded. Three quarters of those with a score were below 0.10 that day, and the median was 0.0066. A CVE can be exploited and still carry a low daily probability, especially when the exploitation is new, targeted or reported late.
This is not a defect to be fixed by choosing a lower threshold. It is the reason KEV sits above EPSS in the ranking: KEV records the event, and EPSS estimates the future. Use EPSS for CVEs with no confirmed exploitation, and expect it to miss some that will be listed next.
The daily change
Because scores are recalculated every day, the same CVE can move between runs. FIRST's FAQ says that in the CSV the first line is a comment that names the model the scores were produced under, so a model change is visible in the data you store.
Store the score, the percentile, the score date and the model line at the time of each decision. When an auditor asks why a CVE was in the 30-day queue on a given Tuesday, the answer is the value you saved that Tuesday, not today's value. Alert on large day-over-day jumps rather than on a fixed level alone.
A checklist before you trust a threshold
FIRST says there is no universal threshold and that 0.10 is commonly cited but carries no special authority from EPSS. Set yours from the number of CVEs your team can remediate in a cycle.
Check three things before adopting it. Count how many open CVEs it selects today. Count how many KEV additions in the ledger it would have selected on listing day. Record both counts next to the threshold.
- At 0.10, the ledger's listing-day scores select 63 of 260 additions (24.2%).
- At 0.01, they select 116 of 260 (44.6%).
- Neither number is a recommendation. They show what each cut costs in missed KEV entries.
Recording EPSS so it can be audited
Save the probability to at least five decimal places. The ledger stores values such as 0.00017 and 0.00297, and rounding to two places would turn most listing-day scores into 0.00 and erase the differences between them.
Write the fetch date beside the score date. FIRST publishes scores daily, and a score fetched late on a given day can differ from the file you pulled that morning. Both facts belong in the record so that the answer to an auditor is a lookup.
Common errors
Four errors recur. Reading the probability as a percentage chance of compromise on your asset. Comparing a score from one date with a score from another and calling the difference a trend without checking the model line. Sorting by percentile and treating the top 12% as 12% likely. Using EPSS as the only gate, which the ledger argues against because three in four KEV entries were below 0.10 on listing day. Write each as a line in your review checklist and tick it when you sign off the queue.
Sources
- FIRST, EPSS frequently asked questions, read 10 Nov 2026
- FIRST, EPSS probability and percentile bins, read 10 Nov 2026
Last reviewed 10 Nov 2026.