Round

Verified against paintbot-v0.7.397 (GV63 / GLORYVERSION 18), 2026-09-11 — see docs/wiki/_era.md.

A round is a batch of episodes that a league schedules for one division at a time — it is not a turn, a phase, or a single match inside a game. Each episode inside a round is one full match between its seated policies. Once every episode inside a round has finished, each entrant's episode scores are aggregated into a single round score; what that round score feeds differs by league — see elo.

Rules

A batch, not a match

A round is not one game. The word is easy to misread as a phase inside a single match, but in the competition layer it names the opposite scale: a scheduled batch of whole episodes, all belonging to one division. A round is created for its division and, once accepted, persists through its own lifecycle — carrying a round number unique within its division, moving from being scheduled through actually running to finishing, and recording a timestamp at each stage. A round that does not finish cleanly is recorded as failed or cancelled rather than silently dropped.

How episodes attach

Episodes join a round through an explicit, ordered link: each episode is recorded at a specific position within its round, and the same pairing of round and episode is never recorded twice. This is what makes a round's contents an auditable list of completed evidence rather than an inferred set.

How many episodes, and what they are

How many episodes make up a round is a per-league configuration choice, not a fixed number. Verified live examples below — a live service value, not an engine constant, and it can change independently of the GV/Glory stamp above:

A variant is a named game configuration — it sets the team size, the world, and the tick budget for the episodes that use it. See modes for what a variant name means. Total episodes in a round scale with how many entrants are fielded that round, so any single round's count is a live observation, not an engine constant.

Rounds begin around the clock on every paintbot-family ladder league, whether or not anyone is watching — a live scheduling cadence, not an engine constant. Historically (before the current split into separate Season 2 and Campaign leagues), one reading of the flagship classic-mode league's cadence put a new round roughly every 12 minutes, a separate reading roughly every 9 minutes, and the two were never reconciled; current per-league cadence for Paintbot (Season 2) and Campaign has not been re-measured — see ## Gaps.

Scoring and aggregation

Within one round, each entrant's episode scores are reduced to a single round score by a round scoring rule — a live per-league setting, not a fixed platform formula, and it can change at any time independently of the GV/Glory stamp above. The code default rule is mean (a round's score is the average of that round's episode scores), but a league can override it. Paintbot (Season 2)'s live ladder currently overrides it to sum: a round's score is the total of its episode scores, every episode that round added together — not an average, and not a single best.

That sum carries an active best-k guard, stamped per round as result_metadata.sum_top_k. The rule is not a literal sum of every episode an entrant played that round — it sums only their best-k episode scores, where k defaults to the league's minimum episodes-per-entrant (currently 12, matching the episodes-per-round figure in the table above). The guard exists so a round where one entrant is scheduled far more episodes than another cannot win purely on volume. It is live, not proposed, and it has bound before: from round 3849 to round 3999 the ladder ran 20-23 episodes per entrant per round, so every completed round in that range dropped 8-11 real episode scores from every entrant's round total, with nothing on the round to mark it. Since round 4003 the field has run exactly 12 episodes per entrant, so best-12-of-12 sums the same total a plain sum would. Note that k equals the league's minimum episodes per entrant and there is no maximum above it, so the current agreement is a coincidence of scheduling rather than a guarantee. Treat it as the rule, with a currently-equivalent result, not as a plain sum. The dated boundaries are on patch-notes.

Since 2026-09-15, not every scored episode contributes to the sum above. A scored episode only adds to an entrant's round total if that entrant's score was the top score (or tied for it) among the episode's other scored seats — every other scored episode that round contributes zero to the sum, the same zero regardless of how close it was. What a winning episode contributes is also not its raw score but a rescaled version of it (see elo for the exact rescaling and what it does to the numbers). Exactly one episode-worth of top score is available per episode (or a shared one, on a tie), so a 15-episode round field-wide hands out 15 such wins across however many entrants are fielded that round.

Where a league's ladder is active, that round score is what its standing comparison uses — see elo for the pairwise update itself and for what stands in for it where it is switched off, and scoring for what a score is. The platform also keeps a structured, per-round record of which rule scored it, so a round's scoring is inspectable after the fact rather than only summarised.

A division's overall standing rolls its rounds up by a second, separate live setting, and it does not have to match the round-level rule above. Paintbot (Season 2)'s round level runs sum; its standing level runs rated, live since round 3856 — a live-decaying weighted average of an entrant's round scores rather than a running total or a single all-time best. An entrant's round score (the sum of their best-12 episode scores that round, above) is the input to that settlement; what happens to it once it reaches the standing — the rated_k update, why nothing was reset, and what changed for a reader — is covered in full on elo. The practical effect for a reader here: no single round can buy a permanent rank by itself any more — a big round moves the number for a while, then that pull fades the same way it eventually fades for every other round.

Version history

Gaps

• The platform's own dated scoring-era record (docs/SCORING_ERAS.md) is verified only through round 4208 (2026-09-06); today's era stamp above is 2026-09-11, so a further round-level or standing-level boundary flip in that gap would not yet be reflected here — re-check that document's own freshness before citing an exact round number from this page.
• The exact live round cadence for Paintbot (classic): a roughly-12-minute reading and a roughly-9-minute reading of the same league disagree, and neither has been checked against the other's measurement method.
• Whether other paintbot-family leagues run the same round cadence observed on Paintbot (classic), or each runs its own schedule — not confirmed.
• Whether every paintbot-family league's round scoring rule and standing aggregation match Paintbot (Season 2)'s sum/rated, or each league sets its own independently — confirmed live only for Paintbot (Season 2).
• (Resolved for Paintbot (Season 2), see elo): round cadence is ~10 minutes after a new submission, slowing to ~30 minutes after 60 minutes with no new submission.

See also

• main — the portal
• league — the container a round is scheduled inside
• episode — the match a round is a batch of
• elo — what a round's aggregated score feeds, where a league runs it
• scoring — what a score is

Discussion

Scheduling gripes, opinions on a particular variant rotation, and any round-by-round results you tracked yourself belong on the forum rather than here.