Paging

Three endpoints return more rows than fit in one response, and all three page identically:

• GET /v1/accounts — your users
• GET /v1/connections — their authorizations
• GET /v1/activities — the workouts you have been granted

GET /v1/activities?limit=100 HTTP/1.1

{
  "activities": [ "…" ],
  "has_more": true,
  "next_starting_after": "7e14a9c3-2b60-4f85-9d3a-6c081ef47b52",
  "total": 1284
}

Send next_starting_after back as starting_after and you get the next page. Keep going while has_more is true:

let cursor: string | undefined;

do {
  const page = await stridee.get('/v1/activities', { since, until, limit: 200, starting_after: cursor });
  for (const activity of page.activities) await handle(activity);
  cursor = page.next_starting_after;
} while (cursor);

starting_after takes the id of the last row you received — a value you already have and can read in a log line, not an opaque token. It must name a row of yours, or you get a 400: an unknown cursor is a bug in a loop, and answering it with an empty page would look exactly like reaching the end.

Two fields worth reading carefully#

has_more is the paging control, not total. It comes from a row the endpoint fetched and discarded, so it is exact. Stop when it is false, not when your row count reaches some number you computed.

total is on the first page only — the page you asked for without a starting_after — and null on every page after it. It does not change as you page, and counting a quarter of a million rows again per page is work with no reader. It is the number you put above a table; it is not part of the loop.

Why there is no offset#

Every list here is newest-first over data that keeps arriving. With offset, a workout landing while you page pushes every row down one, so page two re-serves rows page one already gave you — and no client should have to be told to dedupe around that. A cursor names a row rather than a distance from one, so nothing arriving above it moves your position.

It is also the difference between a backfill that finishes and one that crawls: offset makes the database walk every row it is skipping, so page 5,000 costs thousands of times page one. A cursor is one seek whatever page you are on.

If you are re-reading a fixed window — a nightly reconcile — pass both ends: since=<last run>&until=<this run>, half-open, and next time since is this run's until. That window cannot change while you walk it, and nothing is counted twice.