IVK Skill

Skill ratings
players trust.

A player-facing rating that combines skill and progression and stays stable under real matchmaking. One API call to start, hosted or licensed.

Let's talk Read the docs
Player-facing·Stateless·30+configurable inputs
Ranked 3v3 TDMSeason 13
GigitiGoo#3356
Platinum III
33,070MMR
↑ +11,070 this season
last 10 matches
the problem

Every rating system breaks
under skill-based matchmaking.

Elo, TrueSkill, OpenSkill, and Glicko were never built for it. Each one fails differently, and all of them make you pay.

01 · A FORCED TRADEOFF

Pick your poison.

Elo is slow to adapt, sometimes taking hundreds of games to get to a player's real skill level. TrueSkill and OpenSkill solve that, but once you reach your real skill level, it is very difficult to adapt.

02 · CAN'T BE SHOWN

Hidden MMR by default.

Each system measures skill or progression, never both, and the number swings too much to show a player anyway. So studios hide the real rating and hand players a cosmetic rank instead.

03 · NEVER STABLE

Forever maintenance.

Under live matchmaking the distribution drifts. Elo inflates, TrueSkill slides, and the distribution you planned for requires constant patchwork.

ivk skill

IVK Skill
The PvP skill ratings engine.

Most rating systems were never built for modern games, where matchmaking, progression, and how players read the number all matter at once. IVK Skill keeps ratings stable, progression intuitive, and every behaviour under your control.

Accurate ratings

IVK Skill folds performance and progression into a single, player-facing value. That same value drives matchmaking directly.

Intuitive changes

Rating movement that reads the way players expect. Win and it goes up, lose and it comes down, and improvement shows as steady progress.

30+ configurable inputs

Shape how the system behaves with clear parameters: convergence speed, placement matches, team weighting, distribution shape, and more.

RANKED MMR DISTRIBUTION010k20k30k40k50k60k70k80k90k
accurate

Predictable
before day one.

  • Bounded from 0 to 1
    Every rating sits between 0 and 1. One is a god-level player, zero is an AFK, and we can even let it drop below zero for someone actively working against their team.
  • Know the curve before launch
    Because the scale is fixed, we can show you what your player distribution will look like after 50, 100, or 1,000 matches, before a single game is played.
  • No drift, no remaps
    The distribution holds its shape on its own. No remap every few weeks, no rebalancing sprint each quarter.
LEADERBOARD · SEASON 18
#24H CHANGEPLAYERSKILL RATING
1+312
VXLRstatic#4827
70,956
2-89
VXLRquietriot#1903
70,648
3+204
XXXXLaske#6510
69,783
4—
KZTNt-grave#0001
68,642
5-43
KZTNGigitiGoo#0002
68,572
6+178
XXXXnullVect#3204
66,870
7+521
KZTNpois00nP#8061
66,733
8-167
QRTvroomtv#5429
66,657
9—
QRTwoodyB#2716
66,190
10-92
XXXXDeadplate#9038
65,651
player-facing

Skill and progression,
combined.

  • Player-facing MMR
    Skill drives matchmaking accuracy. Progression gives players something to chase. Both live in one number.
  • Converges in ~5 games
    Set placement matches however long you like. In practice we are confident about a player in around five games, depending on the game type and mode.
  • Built to be shown
    Honest enough to put on screen. The Finals shows players their actual skill rating.
ranked config
{
config_id: "k5982hhd8",
base_step_size: 0.03,
player_model: {
default_mmr: 0.14,
placement_length: 10,
same_team_blend: 1.0,
bot_blend: 0.75,
player_perf_weight: 0.7,
perf_beta: 0.11,
pre_placement_alpha: 3,
},
team_model: { interdependence: 1 },
population_model: { type: "beta", a: 3, b: 3 },
}
configurable

Tune it to
your game.

Over thirty variables let you shape the algorithm into something that is unmistakably yours, and our team is with you every step of the way to guide the tuning.

features

Built for real games,
not textbooks.

Pre or post match updates
Run traditional post-match skill calculations or calculate all possible MMR changes pre-match.
Controlled streak breaking
We catch unusual performance runs and smooth the adjustment, so no rating whiplash from a hot or cold streak.
Simple inputs
Built for developers, not mathematicians. No need to learn sigma, mu, or other statistical variables.
Stateless
A single match result is enough to update ratings. No need for large data syncs or long histories.
Team dynamics
Tune how much individual versus team performance moves each rating, per game mode.
Placement matches
Configure how placement matches behave, including how many are played, how volatile early ratings are, and how quickly players converge to their true MMR.
Partial and match rejoins
Support full or partial MMR updates for players who quit, drop and rejoin, or even switch teams mid-match.
Retries and fallback
Retry and fallback logic is built into the SDKs, so failures recover automatically with no error handling of your own.
Game mode agnostic
Works out of the box for 1v1, 5v5, FFA, battle royale, objective-based modes, and anything in between.

Ready to solve
your matchmaking?

Talk to engineers who've shipped matchmaking at scale.

The Invokation wizard, casting