Deta Base fetch pagination cursor binding to query predicates prevents mid-iteration filter refinement

Context

Deta Base fetch() returns a maximum of 1,000 items per call and uses an opaque last cursor for subsequent pages. The cursor is bound to the exact query predicate supplied in the initial call; altering any condition operator or attribute between paginated requests invalidates the cursor and restarts iteration from the beginning.

Goal

Enable a consumer to paginate through a large, query-bounded dataset while retaining the ability to narrow or adjust filter criteria after inspecting early pages — for example, adding a secondary attribute filter once a broad category scan reveals the data distribution.

Constraints and uncertainty

• The last cursor cannot be reused with a modified query parameter; the documentation states that changing the predicate restarts the scan.
• No offset/skip mechanism exists to jump to a logical position without a valid cursor.
• Eventual consistency means writes during pagination may appear or disappear across pages, complicating any client-side merge of separately fetched result sets.
• The 1 MB response cap can further reduce the effective page size, increasing round trips for a full traversal.

Open questions

1. Is there a supported pattern to preserve pagination position when the query predicate must change, or must the consumer always restart from the first page?
2. Can a snapshot or consistent-read mode be requested to guarantee that a multi-page traversal reflects a single point-in-time view despite eventual consistency?
3. Does the last cursor encode enough information to allow a consumer to reconstruct an equivalent query that resumes at the same logical offset after a predicate change?