guides / index 23 publishedmechanism first man 5 elf

Binary compatibility check

Measured, not guessed 2 business days Fixed price Request a check →

You have a binary or a container image that runs on one machine and refuses to run on another. The usual answer is a guess (“maybe glibc is too old”). This is the measurement: I run it against the distributions you care about and tell you exactly what fails, with the command that shows it.

The methods are the ones published on this site — the exec format error guide, the 27-distribution matrix and the Sentry benchmark were all produced the same way.

What you get

  • A verdict table — for each target distribution: runs, does not run, or runs with a caveat.
  • The exact symbol that is missing when it does not run — GLIBC_2.34, GLIBCXX_3.4.29, a missing interpreter, or the wrong machine type.
  • Reproduce commands — the same four commands in your own CI, so you can re-check after every change.
  • Migration notes — what breaks if you move to Alpine, distroless or a newer base image, in the order it breaks.
  • A short call (30 minutes) if the table raises questions you want to talk through.

Where this is worth paying for

Moving to a smaller base image. Alpine is musl, not glibc. A binary that works on Ubuntu fails there with a message that says the file does not exist, while the file is present.

Shipping one prebuilt binary to many customers. The oldest glibc among your customers sets your build floor, and it is usually older than the runner your CI uses.

Upgrading a base image. The upgrade succeeds, the container starts, and something subtle stops working two weeks later. The version deltas are the reason.

Price

ScopeWhat is measuredPrice
Single checkOne binary or image, three target distributions of your choice$199
Release setUp to five binaries or images, eight distributions, plus a written migration plan$449

Fixed price, paid before the work starts. If the check shows your binary is fine everywhere, that is the answer you get — the price is for the measurement, not for a finding.

How it works

  1. You send the binary or image reference and the distributions you care about.
  2. I run it on clean hosts in those distributions — pull, run, read the loader output, check the symbols.
  3. You get the verdict table and the reproduce commands within two business days.
  4. If something does not match your environment, it gets re-measured before the report is delivered.

What this is not

  • Not a penetration test or a CVE scan — this is about whether the thing starts and links.
  • Not a code review, and not a promise to fix your build.
  • Not a benchmark of performance — that is a separate, longer job.
  • No NDA needed for the work itself; if you need one, say so in the request.

Questions

Can you work with a private image? Yes — a registry pull token with read-only access to that one repository is enough, and it is revoked after the run.

What if my target is Windows or macOS? The check covers Linux distributions. For those platforms I can tell you what the correct build is, but the measurement is Linux.

Do you keep my binary? No. Everything is deleted after the report is delivered, except the version numbers that go into the table.

How fast can you start? Two business days from payment, unless you need a specific date.

What if I need it today? Say so in the request. If a slot exists it costs the same; if not, I will tell you rather than take the money.