P215 · Tool use

Prove paginated retrieval is complete

Follow each endpoint’s continuation contract and distinguish complete from partial retrieval.

Editorially reviewed

These examples and illustrative results are independently authored teaching materials, not measured model results.

Use case

Aggregate all subscribers in a declared filter/access scope. Teaching page 1 has 50 and next_cursor=c2; page 2 has 20 and null. A successful first response is not complete retrieval; use returned cursors unchanged.

Mechanism

Inspect paging/termination, record scope/page counts/continuation and sequentially accumulate stable IDs. Claim scoped completeness only after every page succeeds and the documented terminator. Failures, denied access, repeated cursors or caps retain partial state/next position.

Bad example

Use first-page 50 for all-subscriber metrics and call 200 complete without c2/failure inspection.

Good example

Follow teaching page 1→c2→page 2→null under one filter/account, reconcile 70 IDs and claim complete only with both successes/terminator. Failure/cap reports actual pages/count/cursor as partial without invented rows or quota bypass.

Why the change matters

A page response covers that page only; explicit termination establishes traversal while scoped continuation prevents mixed filters.

Observable expectation

Two successful pages yield 70/null; failed second page yields 50/partial. Repeated c2 fails rather than loops.

Inspect identity/termination, not counts alone; no subscribers are queried.

Limits

Completeness is scoped and changing data may need snapshots. Endpoint page/offset/cursor contracts differ; frozen null fields are examples, not universal termination.

Sources and evidence

Read the editorial criteria