Some Episodes Just Stopped, And Nothing Told Me

Every episode in PodGlue goes through the same steps. The audio comes in, it gets turned into a transcript, and the transcript is what everything else is built from: show notes, the social posts, the newsletter.

That first step is the one everything depends on. If the transcript never happens, nothing after it happens either.

And for a while, when that first step failed, PodGlue said nothing.

What it looked like

An episode would show up in the list. It would sit there. No show notes, no newsletter draft, nothing.

Here's the part that got me: it looked completely normal. An episode where the transcript failed looked exactly the same as an episode I simply hadn't gotten to yet. Both of them just said "needs transcript." One of them was waiting for me. The other one was never going to move again, no matter how long I waited.

There was no way to tell them apart by looking. I found it by noticing an old episode I was sure I'd already handled.

Why it went quiet

The other steps in the pipeline all knew how to report a problem. If the show-notes step failed, it recorded that it failed, wrote down why, and tried again a few times before giving up.

The transcript step had none of that. It got started, and then nobody checked on it. If it worked, great. If it didn't, the failure went into a log file nobody reads and the episode was quietly dropped from the queue forever.

Worse, the daily automation run would finish and report that everything went fine. From its point of view it had. It had kicked off the work. It just never looked to see whether the work landed.

What changed

The transcript step now works like every other step.

It records that it's running. If it fails, it says so, saves the reason, and tries again. If it keeps failing, it stops and marks the episode as needing attention instead of retrying forever.

And you can see it. An episode with a failed transcript now says so, and shows you what went wrong, instead of pretending it's just waiting its turn.

There's one distinction I care about more than the rest. Some failures are worth retrying: the service was busy, the network hiccuped. Some aren't. If a file is a video instead of audio, or the episode is too long to process in one go, trying again three more times just burns time proving the file hasn't changed. Those stop on the first attempt and tell you what's actually wrong.

If something crashes halfway through, a background check picks the episode back up a few minutes later and tries again. It doesn't need me to notice.

What's next

This makes the failures visible. The next piece makes them rare, moving the transcript work somewhere it has room to finish, so long episodes stop hitting a time limit in the first place. That's already built and going out shortly.

After that: going back through the episodes that got stranded while this was silent, and getting them moving again.

The thing I keep relearning is that "no error" and "working" aren't the same thing. An episode that quietly stops is worse than one that fails loudly, because you don't go looking for a problem you don't know you have. PodGlue is supposed to be the thing that remembers your episodes for you. It can't do that if it forgets one and never mentions it.