List matches for a game

Game slug. cs2 is the only populated title today; an un-onboarded game 404s cleanly.

Filter by match status. scheduled excludes rows more than 6 hours past their slot; cancelled is the explicit way to list called-off fixtures.

Filter by tournament (UUID).

Filter by team (UUID).

Only matches scheduled on or after this day, as YYYY-MM-DD (e.g. 2026-07-01). Compared against scheduled_at in UTC and inclusive of the whole day. A malformed value is a 400, not a silent no-op.

Only matches scheduled on or before this day, as YYYY-MM-DD. Inclusive — date_to=2026-07-01 includes everything on 1 July, not just midnight.

Keep only matches that actually carry the named sub-resources. Comma-separated and AND-ed, e.g. has=rounds,depth returns only matches with BOTH round data and the per-player depth layer. Values are the same keys as data_available. An unknown name is a 400, never a silently-ignored filter.

has=map_stats is narrower than has=stats on the same endpoint: it keeps only matches whose /stats carries the per-map rows, not just the whole-match aggregate.

has=odds is the one value that depends on your access rather than only on what is stored: it keeps matches for which we hold at least one odds line you can retrieve, so it never returns a match whose /odds would come back empty for you.

Only matches whose changed_at is at or after this instant (RFC 3339, inclusive) — the change feed. Pair it with sort=changed_at and page to the end, then store the newest changed_at you saw and pass it back next time.

Three things to know. It sees the match's own fields — status, scores, winner, times, stage, format, teams and tournament — and the first arrival of each sub-resource: the instant data_available gains a true, because maps, stats, per-map stats, vetoes, rounds, depth or lineups landed for the first time. Later corrections to those rows do not move it again; only the transition from "we hold none" to "we hold some" does. It includes cancelled and long-overdue rows that the default list hides, because a cancellation is precisely a change you need to hear about. And ingestion commits in batches, so a row can become visible a few minutes after the changed_at it carries: overlap your window by 10 minutes and de-duplicate on id rather than resuming from the exact last value. changed_at starts on 2026-09-18; rows untouched since then all carry that date. A malformed value is a 400.

Ordering. -scheduled_at (default, newest fixture first), scheduled_at, changed_at or -changed_at. An unrecognised key returns 400 rather than being ignored. A cursor is bound to the sort it was minted under, so change sort and start again from the first page.

Page size. The default and maximum vary per endpoint — this shared definition documents the common case (default 100, max 500); each operation that differs states its own numbers. cursor carries the page position, not limit.

A value above the endpoint's maximum is silently clamped, not rejected. A missing, zero, negative or non-numeric value silently falls back to the default — so ?limit=abc and ?limit=0 both return the default page size rather than a 400. Read meta.count if you need to know how many rows you actually got.

Opaque forward pagination cursor — pass the previous page's meta.next_cursor verbatim. Omit for the first page. Never construct one by hand.