Research/Software
How long internet standards last before they are replaced
Only 14% of the 9,851 IETF RFCs have ever been obsoleted, and survival analysis says most never will be. Since 2000 the IETF patches its documents more often than it replaces them, and it revises them less often overall.
Oct 29, 2026 · 6 min read

The internet runs on RFCs, the numbered documents the IETF publishes to define protocols such as TCP, DNS and TLS. We treated each RFC's replacement as a time-to-event outcome and ran a survival analysis on the RFC Editor's full index. The answer is that most RFCs are never formally replaced. Of 9,851 RFCs, 1,384 (14.05%) have ever been obsoleted, and the Kaplan-Meier survival curve never falls below one half. Those that are replaced go quickly: the median wait is 4.0 years. The pattern matters to anyone who cites a standard, because a document that is never obsoleted can still be quietly patched many times over.
14.05%
of 9,851 RFCs have ever been obsoleted
4.0 years
median time to obsolescence, among those obsoleted
0.872
Kaplan-Meier share still not obsoleted 10 years after publication
40.9%
of 1,113 RFCs with a held erratum have ever been revised
The question
How long does an RFC live before a later RFC obsoletes it? Existing RFC statistics measure the journey to publication or how widely a protocol is deployed. A prior-art search on 2026-10-28 found no published work that applies survival analysis to what happens after publication. We also asked a narrower question in a companion study: when an error report is marked "Held for Document Update", meaning it should be considered in any future revision, does that revision ever come?
What we did
We used one public file, the RFC Editor's rfc-index.xml. The date an RFC was obsoleted is the publication date of the earliest RFC that obsoletes it. RFCs not yet obsoleted are right-censored at the snapshot, the latest publication date in the index (2026.79). Four predictions were written down in a pre-registration before the data were analysed. A follow-up, also pre-registered, treated "Updates" and "Obsoletes" as competing risks. The companion study joined the 8,116 records in the RFC Editor's errata database to the same index.
What we found
Survival is high and flat. Kaplan-Meier estimates put the share of RFCs not yet obsoleted at 0.917 after 5 years, 0.872 after 10 and 0.833 after 30. Most of the risk comes early, and after two decades almost nothing more happens. The longest-lived replaced RFC in the data is RFC 793, TCP, published in 1981 and obsoleted by RFC 9293 after about 41 years.
Share of RFCs still standing, by years elapsed
We predicted that the chance of being obsoleted within 10 years would fall steadily from one decade to the next. That prediction failed. The rate is an inverted U: low for 1970s RFCs, about 30% for the 1980s and 1990s, then 10.4% for the 2000s and 6.2% for 2010 to 2016. Two other predictions held: the median time among obsoleted RFCs fell in the 3 to 12 year range, and Standards Track RFCs were obsoleted within 10 years far more often than Informational ones (0.171 against 0.069).
Obsoleted within 10 years of publication, by decade
Patching instead of replacing
Why did obsolescence fall after 2000? Part of the answer is that the IETF changed how it revises. For RFCs from the 1980s and 1990s, about four in five first revisions were full replacements. For RFCs from the 2000s on, about three in five are Updates. The switch is a step, not a trend, so our prediction that the Update share would rise strictly across cohorts failed. A second prediction also failed: Updates are not quick patches. The median wait for a first Update is 4.75 years, longer than the 4.0 years for an obsolescence.
First revision within 10 years, by publication cohort
| Cohort | n | Obsoleted | Updated | Either | First revision is an Update |
|---|---|---|---|---|---|
| 1970s | 691 | 0.126 | 0.119 | 0.221 | 0.516 (79 of 153) |
| 1980s | 376 | 0.301 | 0.074 | 0.354 | 0.203 (27 of 133) |
| 1990s | 1,593 | 0.299 | 0.068 | 0.341 | 0.193 (105 of 544) |
| 2000s | 2,888 | 0.104 | 0.131 | 0.220 | 0.586 (373 of 636) |
| 2010-16 | 2,262 | 0.062 | 0.100 | 0.154 | 0.645 (225 of 349) |
The switch does not close the gap. The share of RFCs revised in any way within 10 years fell from 0.341 for the 1990s to 0.220 for the 2000s and 0.154 for 2010 to 2016. So the decline is partly a change in how the IETF revises and partly a real fall in how often it revises within a decade.
Held errata mostly stay held
In the companion study, 1,113 RFCs carry at least one erratum marked Held for Document Update. Of these, 455 (40.9%) have ever been updated or obsoleted, about the 40% that McQuistin and colleagues reported for 2022. Ten years after the first held erratum, 0.660 of these RFCs still have no later revision of any kind. Spot-checks show this is generous: RFC 5321 (SMTP) counts as revised because of a narrow later Update that did not fold in its six held errata.
That does not mean held errata are ignored. Against matched RFCs with no errata at all, from the same year and status group, RFCs with a held erratum were revised within five years about 2.4 times as often. Errata mark documents that people read and use, and those get revised more. Separately, the share of processed errata marked as held fell from 0.357 (January 2018 to April 2021) to 0.182 (June 2021 to December 2024), after the IESG's 2021 statement on errata.
Revised within 5 years: RFCs with errata against matched errata-free RFCs
| Arm | Matched | Subjects | Matched controls | Lift |
|---|---|---|---|---|
| Held erratum, any revision | 963 | 0.199 | 0.084 | 2.37 |
| Verified-only errata, any revision | 1,071 | 0.177 | 0.086 | 2.05 |
| Held erratum, Obsoletes only | 1,079 | 0.068 | 0.035 | 1.95 |
What it means
An RFC number is a durable reference. Once a document survives its first few years, it is very likely to stay in force. But "not obsoleted" no longer means "unchanged": for anything published since 2000, the current meaning of a standard often lives in a chain of Updates. Readers who need the latest rules should follow the updated-by links, not only the obsoleted-by ones.


