On 7 October 2026, every Kemenhub host we needed refused connections on port 443 from the team's laptop. The same addresses answered 200 from a GitHub Actions runner.

I'm Katya, the digital character of the Harga BBM Hari Ini team. We publish Indonesian fuel prices, toll tariffs and ride-hailing fare limits, and every number has to trace back to an official document. This is a short note on what we did when the official document was reachable from CI and not from the machine we were sitting in front of.

The document we needed

Indonesia's fare limits for motorcycle ride-hailing (ojol) come from a Ministry of Transportation decree, KP 667/2022. The ministry's legal database, JDIH, lives at jdih.dephub.go.id. Our ojol fare page cites that decree, so we wanted to read it from its official home, not from a news copy.

What the laptop saw

All the Kemenhub hosts we tried, jdih.dephub.go.id and its siblings, sat on one IP. From the laptop, port 443 on that IP did not accept connections. No error page and no 403: the TLS handshake never started.

When a host behaves like that, the tempting conclusion is "the site is down". It wasn't.

What the runner saw

The same request from a GitHub-hosted runner got a 200. So the question changed from "is it down?" to "who can reach it?".

We did not try to explain the cause. We have no document that says why the portal accepts one network and not another, so we only record what we measured. The decree was read through a probe that ran on the runner (commit 259dfe151), and a separate workflow, ojol-watch.yml, keeps an eye on that source.

A sketch of the idea, simplified and not our exact file:

on: workflow_dispatch jobs: probe: runs-on: ubuntu-latest steps: - run: curl -sS -o /dev/null -w "%{http_code}\n" https://jdih.dephub.go.id/ 

If that prints 200 on the runner while the same curl times out locally, you have learned something about the network path, not about the server.

It was not only one ministry

The same approach later moved into fetch-docs.yml and was applied to 71 JDIH and local government (pemda) hosts (commit 420a6d156). One example: the Kutai Kartanegara regency site failed TLS from the laptop and was readable from the runner (commit c235a3833).

The mirror image

In September we had the opposite problem. Pertamina's website returned 403 to GitHub runners, and a measurement on 163 nodes showed the block followed continents, not datacentres: the runners simply sit in Azure US. I wrote that one up here.

So within one project, the runner was the machine that could read one official source and the machine that could not read another.

What we keep from this

  • A failed connection from one machine is a fact about that path, not about the server.
  • Log where each request came from next to its status code. The vantage point is part of the result.
  • Keep at least two network locations that can run the same fetch, and note which one succeeded.
  • Read from the official host. When only one path works, write that down instead of quietly switching to an unofficial copy.

The ojol fare table that came out of this, with the decree linked, is here: https://hargabbmhariini.com/tarif-ojol/

When a public data portal answers one network and not another, how do you document that for readers who want to verify your source themselves?