Pinned snapshot or floating alias in production - what actually broke for you when the default moved?

Pin. A floating alias is an unannounced deploy of somebody else's code into the middle of your critical path, scheduled at their convenience. You would not accept that from a library dependency and this one is less deterministic than a library.

We do both, split by blast radius. The internal summariser floats, because if it gets weird someone notices in a channel and nothing downstream cares. Anything whose output is written to a database or shown to a customer is pinned, and moving that pin is a normal change with a normal review. The thing that made this workable was a fixture set - a couple of hundred real documents with known-good extractions - that runs nightly against both the pinned and the current default, so we see divergence before we choose to move.

Adding a number from our own harness: on one version change, our fixture set went from almost everything parsing cleanly to roughly three in a hundred failing schema validation, and the failures were not random - they clustered on the longest inputs, which is exactly where a spot check of five examples would have missed them. That is the argument for a fixture set over eyeballing. If you build nothing else this week, build the thing that runs the same hundred inputs and diffs.

Worth saying that pinning has its own failure mode, which people discover later: snapshots get retired, and you find out on a schedule that is not yours. If you pin and then never touch it for a year, you have not avoided the migration, you have deferred it and made it bigger, because you will be moving across a much wider gap. Pin, and put a recurring reminder to re-evaluate every quarter so the jump stays small.

We floated for eight months and it was fine, right up until it was not, and the way we found out was a customer email rather than an alert. The lesson was less about pinning and more that we had no output monitoring at all - no schema validation rate, no length distribution, nothing. Pin first because it is one line, then build the monitoring, because you need it either way.