Everyone says software support windows are getting shorter. I had a dataset sitting around that could actually check, so I did—and the first answer I got was wrong in an interesting way.
The naive answer
I took every tracked release with both a release date and a published end-of-life date—5,411 pairs across 446 products—and plotted the median support window by the year the release shipped.
| Release year | Median support window |
|---|---|
| 2016 | 2.94 years |
| 2019 | 1.95 years |
| 2022 | 1.08 years |
| 2025 | 0.69 years |
A 4x collapse in under a decade. Great chart. Almost certainly misleading.
Why it’s misleading
The set of products in the dataset is not constant over time. Far more products are tracked now than in 2016, and the ones added recently skew toward fast-moving cloud services and JavaScript ecosystem tooling that ship every few weeks.
Products with short cycles are simply overrepresented in recent years. The chart may be showing a change in what was measured, not a change in how long things are supported.
The like-for-like answer
So I restricted it to a fixed cohort: only the 144 products that have releases in both the pre-2019 period and the 2024-or-later period. Same products, both ends of the timeline.
| Release year | Median support window for the same 144 products |
|---|---|
| 2016 | 2.57 years |
| 2018 | 1.99 years |
| 2020 | 1.55 years |
| 2022 | 1.36 years |
| 2024 | 1.50 years |
| 2026 | 1.49 years |
The compression is real, but it is roughly a halving, not a 4x decline—from about 2.55 years for releases shipped through 2018 to about 1.49 years for releases shipped from 2024 onward.
The more useful finding is that it stopped. The decline runs from 2016 to around 2022 and then flattens. Support windows are not still collapsing. They compressed once and settled at roughly 18 months.
The number that actually matters operationally
While computing this, I found another figure that surprised me more:
Of 8,270 tracked releases across 446 products, 6,509—78%—are already past their published end-of-life date.
About 98% of products have at least one unsupported release, and in 89% of products, most releases are unsupported.
That is expected in the sense that old versions accumulate forever. It matters because the supported window is now narrow enough that “we upgraded recently” and “we are on a supported version” have quietly stopped meaning the same thing.
At an 18-month median, a stack you last touched two years ago is statistically likely to be out of support somewhere.
Where the data comes from
Support lifecycle dates come from endoflife.date, an open, community-maintained dataset that cites vendor announcements.
I am a user of the dataset, not its author. It deserves the credit, and anyone relying on a date for a compliance decision should confirm it through endoflife.date or the vendor’s official notice.
Anything with both a releaseDate and an eol date was included. Releases where EOL is unpublished or open-ended were excluded.
That creates another bias worth naming: products that never announce end-of-life dates are invisible in this analysis.
Checking your own stack
I put the same dataset behind a small free tool because answering “Is anything I run already unsupported?” by hand is tedious:
Paste up to 25 technologies and it shows which are past EOL, which lose support soon, and which have a recorded shutdown or acquisition event. Every result includes its source.
No login or email is required. The check runs in your browser against a static file, so nothing you type is transmitted.
Full disclosure: I built it.
Its limitations are important. Coverage is strongest for languages, databases, infrastructure, and developer tooling, but weaker for business software.
Anything the tool does not recognize is labeled unknown, not clear. Silence is not evidence of safety.
Name matching is pattern-based. I already fixed one case where jquery-file-upload incorrectly matched jQuery, so there are probably other edge cases.
What I’d like checked
I would genuinely like to know whether the plateau holds within individual ecosystems or whether it is partly an artifact of which products publish EOL dates at all.
The raw event archive is available as public JSON:
shutdown-radar.pages.dev/data.json
If the tool gives you a wrong or misleading answer for something you use, that is the single most useful thing you could tell me.
Top comments (0)