Dev.to WebDev 🛠 Dev 👁 0 📖 6 min read

Nothing in our dependency tree is newer than 168 days: 42 of 50 behind, nine of them by a major

PubTrivia has 50 direct dependencies across its web app. I ran pnpm outdated, then dated every installed version and every current version against the npm registry, so "behind" could be a number of days rather than a fee

PubTrivia has 50 direct dependencies across its web app. I ran pnpm outdated, then dated every installed version and every current version against the npm registry, so "behind" could be a number of days rather than a feeling.

The result in one line: the most recently published version anywhere in the tree came out 168 days ago. Not the oldest, the newest.

That lines up with the lockfile's history. Its last dependency-changing commit was 28 June; the only commit to touch it since was on 13 October, pinning pnpm and dropping tsx so a Railway build would succeed. So this is a June install with five months of drift on top.

The lag

package                        have       latest     installed  days    stable
                                                      version   behind  releases
                                                      age (d)           skipped
-----------------------------  ---------  ---------  ---------  ------  --------
@radix-ui/react-dialog         1.1.15     1.3.1            461     425        11
@radix-ui/react-dropdown-menu  2.1.16     2.2.1            461     425        11
@radix-ui/react-select         2.2.6      2.4.1            461     425        11
@radix-ui/react-tabs           1.1.13     1.2.1            461     425        10
@radix-ui/react-toast          1.2.15     1.3.1            461     425        11
@upstash/ratelimit             2.0.8      2.2.1            309     304         6
postgres                       3.4.9      4.1.1            226     219         8
prettier                       3.8.3      3.10.1           217     211        15
dotenv                         17.4.2     18.0.10          219     210         9
framer-motion                  12.40.0    14.14.0          180     176        40
lucide-react                   1.17.0     1.75.0           174     173        58
@sentry/nextjs                 10.56.0    11.14.0          168     168        43
@types/node                    25.9.1     26.6.6           182     162        46
next                           16.2.7     16.4.6           169     161        32
stripe                         22.2.0     23.1.0           174     154        14
react / react-dom              19.2.7     19.3.1           169     134         8
typescript                     6.0.3      7.0.2            215      82       129

Summed across the 42 outdated packages: 8,960 days of release time, which is 24.5 years, and 595 stable releases we have skipped. If you count prereleases the number is 3,499, which says more about how often Next.js and React publish canaries than about this project, so the honest figure is 595.

The median installed version is 217 days old. The median gap between what we have and what is current is 176 days.

Nine majors, sorted by what they touch

postgres        3.4.9   -> 4.1.1     the driver holding every connection
stripe          22.2.0  -> 23.1.0    checkout, webhooks, the money path
@sentry/nextjs  10.56.0 -> 11.14.0   the thing that tells us when it breaks
vitest          4.1.8   -> 5.1.1     the test runner
@vitest/...     4.1.8   -> 5.1.1     its coverage plugin, which nothing uses
typescript      6.0.3   -> 7.0.2     the compiler
@types/node     25.9.1  -> 26.6.6    types for a runtime we pin at 20 in .nvmrc
framer-motion   12.40.0 -> 14.14.0   two majors, 40 stable releases
dotenv          17.4.2  -> 18.0.10   local env loading only

Two of those matter more than the rest. postgres is the driver in the deployed WebSocket container, which a licence audit on this project found is also the one Unlicense package in a 23-package production image. stripe is the SDK on the paid path, where a major version is where API version pinning and typed resource changes live, and where a 42-test billing suite with no live Stripe calls is the only thing standing between an upgrade and a surprise.

The two that cost nothing: dotenv is dev-only, and @vitest/coverage-v8 is installed and never invoked, which an audit of this project's dead configuration found a month ago and is still true.

Old is not the same as behind

Eight of the 50 are exactly current, and five of those are the oldest things in the tree:

package                   installed version age   status
-----------------------   ---------------------   ------------------
tailwindcss-animate                  1,177 days   current, 8 releases ever
clsx                                   939 days   current, 15 releases ever
qrcode                                 834 days   current, 50 releases ever
class-variance-authority               722 days   current, 27 releases ever
next-themes                            616 days   current, 55 releases ever

clsx has shipped fifteen versions in its life and the newest is three years old. That is not neglect, it is a function that joins class names and is finished. The same tool that flags 42 packages as outdated is silent about the five oldest files in node_modules, which is the right behaviour and a useful reminder that version age measures the maintainer's habits, not your risk.

Three more are pinned exactly rather than with a caret: jspdf at 4.2.1, next and react at exact versions, and babel-plugin-react-compiler at 1.0.0. Pinning the renderer and the compiler plugin is deliberate on a project where the React Compiler is switched on.

The pacing rule that points at the past

pnpm-workspace.yaml carries this:

minimumReleaseAgeExclude:
  - '@next/[email protected]'
  - '@next/[email protected]'
  - '@next/[email protected]'
  ... nine more at 15.5.19 ...
  - [email protected]

Twelve entries, every one at 15.5.19, in a project whose next is 16.2.7. The field exists to exempt named packages from minimumReleaseAge, which delays installing a version until it has been on the registry for some number of hours, and which is the single best defence against an npm account takeover shipping a poisoned patch release.

pnpm config get minimumReleaseAge returns nothing. So twelve exemptions are listed against a rule that is not switched on, for a major version the project left behind, and a supply chain measure that would actually help here is one line that nobody has written. An audit on this project found all of that 37 days ago and the file has not changed since, which is the part I find most instructive: an inventory finding without an owner does not fix itself.

The other half of the deploy

The WebSocket container is a separate workspace with seven direct dependencies, and it is the half that runs in production rather than on a CDN:

drizzle-orm   ^0.45.2    194 days behind
postgres      ^3.4.9     219 days behind, one major
ws            ^8.21.0    179-day-old spec, latest published 5 days ago
ioredis        5.6.1     585-day-old pin, latest is 6.0.0
@types/ws     ^8.18.1    596-day-old spec
@types/node   ^25.9.3    one major behind
typescript    ^6.0.3     one major behind

ioredis is the one to look at: pinned exactly, with no caret, to a version published 585 days ago, while the current major is 6.0.0. An exact pin on a Redis client is a defensible decision and an undocumented one, and that is the whole problem with pins: six months later nobody can tell a decision from an accident.

There is also a straight contradiction sitting in two files. This workspace's engines says node >=24. The repository's .nvmrc says v20.18.1. The machine I measured on is running 24.22.1, which satisfies one of those and ignores the other.

What this measurement does not say

"Behind" is not "vulnerable". I ran no audit, I am claiming no advisory, and nothing here is evidence of a security problem. pnpm outdated reports what the registry calls latest, including majors you may have excellent reasons never to take: framer-motion 14 is two majors of API change for a site whose animation budget is a footer watermark fading in.

Days-behind is release time, not risk. The honest reading is narrower: the tree was installed in June, nothing has been reviewed since, and the gap is now wide enough that any single upgrade pulls a bigger diff than it would have done in July. lucide-react is 58 releases behind and will almost certainly be a no-op bump; stripe is 14 behind and will not be.

What I would do, in order

  1. One sweep of the minor and patch updates on the 33 non-major packages, which is a single command and one run of the existing test suite.
  2. Switch on minimumReleaseAge, delete the twelve exemptions for a version that is gone, and keep the field for when something genuinely needs it.
  3. Take stripe and postgres as two separate, deliberate upgrades with the billing suite run in between.
  4. Delete fast-check and @vitest/coverage-v8, which are installed and unused, and delete eslint or make it run. Since Next 16 removed next lint, the lint script in this project fails with Invalid project directory provided, no such directory: .../lint, so 14 skipped eslint releases are 14 skipped releases of a tool that cannot execute.
  5. Write down why ioredis is pinned, next to the pin.

If you want to see these versions doing their jobs, every one of them is on the live site: the Resources menu in the header is a 425-day-old Radix dropdown and works perfectly (any page will do), the icons in the breadcrumb trail come from a lucide-react 58 releases behind, the footer wordmark fade is framer-motion 12, and the pricing page is where Stripe 22 talks to the API. The point of the audit is not that anything is broken. It is that nobody could have told you whether it was.

📰 Read the original article on Dev.to WebDev

Originally published by Dev.to WebDev. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.