How the rating works

Rating is respect points. A newcomer gets 800 — starting capital, the rest is on you.

• Every game is 7 little duels: you against everyone else at the table. Finish higher — the duel is yours.
• Beat someone strong — take a big cut. Beat someone weak — good job, here's a penny. Lose to someone weak — that hurts. Lose to the champion — no big deal, everyone saw it coming.
• Tournaments are serious business. The deeper the stage, the more a game is worth: the final counts one and a half times the qualification.
• Lobbies are joyrides: they weigh ×0.5 of a tournament game. And if you grind all evening long, the system starts yawning and nearly stops handing out points. You can't no-life your way to the top.
• Points don't appear out of thin air — they flow from the losers to the winners. Want to climb — beat the people above you. That's it.
• Your first 5 games are calibration: the table is sizing you up and won't give you a place yet. Play all five — welcome to the overall standings. The first 10 games move you twice as fast, but with no tournament bonuses: calibration measures your level, not where you played it.
• Life at the top is hard: top players get less for their wins, and losses hit harder. That's by design.
• A single game can never move you by more than ±50 — even if everything went completely sideways (internet died, the lobby fell apart). One bad game is a scratch, not a catastrophe.

Craving formulas and coefficients? Hit “Full math version”

In short

• Every 8-player game = 7 head-to-head duels, one against each opponent: finish higher — you win the duel.
• Beating a strong player pays more than beating a weak one (Elo expectation) — head-to-head is baked into the core.
• Points flow between players: to climb, you have to win. Game volume alone doesn't raise your rating.
• Tournaments carry full weight, lobbies ×0.5, and deep stages weigh more than the qualification (up to ×1.5).
• Newcomers start at 800 and are “calibrating” for their first 5 games — they don't appear in the overall standings.

1. The core: pairwise Elo

For player i against each opponent j in a game:

Dividing by 7 normalizes a game to “one unit”: an 8-player game doesn't weigh more than a 1-on-1 duel. A 400-point gap = a 10:1 win expectation. All changes within a game are computed from the ratings before the game and applied simultaneously.

2. K-factor — how fast you move

• First 10 games: K = 64 — a newcomer quickly rides up to their real level. In those games a tournament game counts at weight 1.0 — no stage or category multiplier (lobbies as usual, ×0.5). Otherwise a Major final during calibration would be a springboard nobody could match later.
• After that: K = 32.
• At a rating ≥ 1000: K = 24 — the top of the table is more stable, one random game doesn't wreck it.

3. Game weight W

W = W_type · W_stage · W_tournament · W_day

• Type: a tournament game is ×1.0, a rated lobby ×0.5 — lobbies count, but far less.
• Stage depth: linear from ×1.0 (qualification) to ×1.5 (final). For the 4 stages of the Free Cup: qual 1.0 → quarterfinal ≈1.17 → semifinal ≈1.33 → final 1.5. Elimination games are worth more.
• Tournament importance: by category — Major x1.5, Free x1.0, Fast x0.8 (a specific tournament may have its own coefficient); hard limits 0.5-2.0 that the system will not let you exceed.
• Daily lobby decay (anti-farm): within one day, lobby games weigh 1.0 → 1.0 → 1.0 → 0.75 → 0.75, then 0.5. A normal evening (up to 5 games) is barely trimmed; a marathon counts at half strength.

4. Anti-farm, or “no rocketing into the heavens”

• Elo is nearly zero-sum: your plus is someone else's minus. Playing a lot ≠ climbing; climbing = winning.
• The higher your rating, the less you get for beating weaker players: against an 800-rated field a top player's expectation is already 0.75+, so a win brings crumbs while a loss hits hard.
• Lobbies at ×0.5 + daily decay: farming with the same crew yields very little.
• K = 24 at the summit: the top moves slowly, and deservedly so.
• Per-game cap: ±50. Even in the worst combination (calibration K, a final, an expensive tournament) a single game can't change a rating by more. A “top-1 vs top-50” match will never produce −1000: without the cap, the maximum for a regular game is ±32 and for a final ±48 — the cap insures against outliers (technical issues, future coefficients).

5. Newcomers

Everyone starts at 800. Until a player has 5 games played, they are “calibrating”: shown in a separate block below the table with a “~N” rating and excluded from the overall places. That way a fresh 800 never ranks “above” a veteran who has actually played: the table only lists those the rating has already measured.

6. Three tables — one engine

• Overall — everything above: tournaments at full weight, lobbies ×0.5. Medals, milestones, player of the day and default tournament seeding use it.
• Tournaments — same formula, tournament games only. Who really carries in tournaments; lobbies don't dilute it.
• Lobbies — rated lobbies only, weight ×1.5, no calibration (K = 32 from game one) and no daily cut: every game at full weight, up and down. Grind and win — you're on top; one tournament a month won't get you here. A place from 3 games.
  The lobby table is seasonal: it counts only the lobbies of the current time of year and resets when the next one starts (winter, spring, summer, autumn). Seasons never touch the all-time and tournament ratings.
• The organizer picks in the tournament settings which table seeds and grants direct entry: tournaments or overall for majors and leagues, lobbies to select the active ones.

7. What does NOT affect it

• No-shows — a game without you doesn't count as played (the rating is about the game; discipline is handled by tournament penalties).
• Penalties and DQs — deducted tournament points don't touch the rating: placements in played games remain a fact.
• Placeholder bots — thrown out of every duel.
• Time — there's no inactivity decay yet (the community is small; it can be added later).

Back-of-the-napkin examples

Tournament coefficient: limits and recommendations

The coefficient multiplies the price of every game in the tournament. The system won't let it leave 0.5–2.0: already at 2.0 a final carries an effective K = 96 — one game moves a top player by ±48 points, and a full tournament (~11 games) can yield +300. Any higher and the rating turns into a one-evening lottery. Recommendations:

Rule of thumb: keep adjacent tournament types 0.2–0.3 apart, no more — otherwise the “cheap” tournaments stop mattering and the motivation to play them drops.

Ideas for later: seasons (a soft reset toward the middle), inactivity decay, tournament coefficients based on the prize pool, a confidence interval instead of a calibration threshold (Glicko-2).