How do you perform time travel queries in LanceDB?

Time travel means querying a LanceDB table as it existed at a specific past version, rather than its current latest state, by putting the table into a pinned, read-only mode pointed at that version.

print(table.version)              # e.g. 5, the latest version
table.checkout(2)                 # pin the table to version 2 (time-travel mode)
old_results = table.search(query_vector).limit(5).to_list()
table.checkout_latest()           # return to tracking the latest version

While checked out to a past version, the table behaves as read-only for that pinned state — queries reflect exactly what the data looked like at that version, which is useful for debugging ("what did this table contain when this bug was reported?"), reproducibility ("evaluate this model against the exact dataset version it was trained on"), or auditing changes over time.

If you need to keep writing starting from that old state rather than just reading it, restore() is the relevant operation instead of plain checkout(): restoring creates a brand-new version whose data matches the old one, which can then be written to going forward, while checkout alone leaves the table in a read-only, time-travel view until you explicitly return to the latest version.