cert-governance · severity medium

Certificate lifetime readiness

Share of newly issued certificates with a validity window longer than the 47-day maximum the CA/Browser Forum is phasing in — renewals that must be automated before the deadline.

Measured every Sunday · last updated

How it is measured

Read Certificate Transparency logs, parse each leaf, compute not_after − not_before. CT logs are sharded by a certificate's expiry year, so every shard that can hold a certificate issued today is read, each contributing in proportion to its share of the window's entries; within a shard, entries are drawn as evenly spaced clusters across the 24 hours before the build. Every CA is therefore represented in proportion to what it issued over the same period, at every validity length. Population unit: certificate; segmented by CA.

Why it matters

The same deadline seen from the supply side. It shows which CAs have already moved their customers to automated short-lived issuance and which are still writing certificates that will be non-compliant — a per-CA readiness table nobody else publishes.

Limits of this measurement

Measured over a sample of recent issuance, so it reflects current practice rather than the installed base. The 47-day threshold was chosen after a 100-day cut returned 0% everywhere — issuance had already moved below it, leaving no variance to observe. The sample is drawn in contiguous clusters rather than one certificate at a time, which widens the margin of error beyond what the sample size alone implies; the effect is small for a CA issuing steadily and largest for one that issues in bursts. A CA's figure covers only the sampled window, so a CA with low issuance in that window is suppressed rather than published thin. One further caveat on n, not on the rates: the sample is a fixed number of log entries however busy the logs are, so the sampling fraction moves with their throughput. Rates stay unbiased, but a segment's n tracks log volume as well as that issuer's own output, so week-to-week changes in n are not by themselves a change in issuance. A break in the series at 2026-10-13. Until then this read a single CT log, and because logs are sharded on expiry year that silently selected for short validity as 2026 wore on — by October the 2026 shard could not contain a 90-day certificate issued that month at all, and the figures it produced fell toward zero. Points before the break were drawn from one shard and understate long-lived issuance, increasingly so through 2026; from the break the frame spans every relevant shard. Sampling frames are recorded per snapshot, so which applies to a given point is checkable rather than inferred. Finally, this reads one operator's logs (Cloudflare Nimbus): a CA that does not submit there is absent from the population rather than measured at zero.

The base rate

Measured over the full population of certificates, from public bulk sources.

All certificates 90.71% fail · n=44,085
within 47 days: 9.29%longer than 47 days: 90.71%
within 47 days · 9.29%longer than 47 days · 90.71%

Every segment

Pick one to see the citable answer, with its trend and methodology.

From the newsletter

Where this base rate turned up as a real, fixable exposure.

Slices below 100 measured members are withheld rather than shown with a wide interval — an absent country or issuer means too little data, not zero failures. Licensed CC BY 4.0. Machine-readable data per segment: append .json or .csv to its URL.