Pagination: Offset, Cursor, Infinite Scroll

TL;DR

Use cursor pagination for mutable ordered datasets; offset is acceptable for mostly static collections and simple admin tooling.

Problem this solves

• Determines how incremental loading behaves under concurrent inserts/deletes.
• Impacts cache design, duplicate handling, and user scroll continuity.

Decision criteria

• Offset: simple implementation, weak consistency under concurrent writes.
• Cursor: stronger ordering guarantees, more complex token handling.
• Infinite scroll: UX mode that can use either offset or cursor transport.

Layered depth

Intro

Pagination is both a data contract and a UX contract; mismatch leads to jumps and duplicates.

Decision criteria

Prefer cursor when list mutation frequency is medium/high or resume continuity matters.

Implementation detail

• Encode sort key and tie-breaker in cursor token.
• Keep seen-ID set to drop duplicate records across page boundaries.
• Preserve anchor index when prepending/appending data.

Failure modes

• Offset pages shift under inserts and duplicate visible items.
• Cursor token becomes invalid after backend index changes.
• Infinite scroll lacks end-state messaging and traps keyboard users.

Pitfalls

• Mixing sort orders between backend query and client merge logic.
• Treating cursor token as opaque in logs and losing observability.

Testing and accessibility

Production trade-offs

• Cursor correctness improves consistency but increases backend coordination cost.
• Infinite scroll lowers click friction but raises navigation and observability complexity.