After looking into this I could identify one case where records where missed by the trigger node. The Notion trigger node compares the internal ID of the previous record which it discovered as having changed with the ID of the last updated record it receives when querying Notion (here).
So where you update a record (which is found by the trigger) and would then update that same record again it would miss the second update. This would also miss records that have been updated in the meantime if the last record you update between polling intervals is the one that has previously been found by the trigger.
I shall add this to our bug tracker for a closer look and hopefully resolution.
Until then I think @Shirobachi approach should be fine for many use cases. If you need to control the polling logic yourself you can always use the approach described at Creating triggers for n8n workflows using polling ⏲ to fetch your relevant records. There unfortunately is another bug with the Notion node (not the trigger) requiring you to use JSON filters to query dates, you can find an example for this in the respective forum topic: Notion wrong send request since n8n@0.157.0 - #14 by MutedJam.
Please accept my apologies for this inconvenience.