How Your Rating Is Calculated

HRI uses Glicko-2 with event-weighted K-factor multipliers, applied post-hoc. The Glicko-2 core is the same algorithm that powers lichess.org. The event weighting scales rating changes by tournament tier and elimination-round depth, without feeding inflated opponent ratings back into the engine.

This page is the authoritative methodology document. Every parameter, every guardrail, and every deliberate deviation from pure Glicko-2 is listed below. Changes to the system over time are recorded in the Changelog.

The Basics

Starting Rating
1,500
Every new player starts here. Think of this as "average."
Starting Deviation
250
Rating Deviation (RD) measures how uncertain the system is about your true skill. It starts at 250 and decreases as you play more events. A new player's credible interval is ±500 points — the rating swings hard until we have more data.
The rating you see
Raw rating
The number shown everywhere — leaderboard, profile, audit page — is your raw Glicko-2 rating. One number per format. The ± value next to it is the width of the 95% credible interval, not part of the rating.
Leaderboard Sort
Rating − 2 × RD
Ranks are computed from the lower bound of your credible interval, not the raw rating. That penalizes high uncertainty — two players both rated 1,800 won't sort the same if one has RD 40 and the other has RD 150. Tighter interval wins. New players with high RD sit naturally toward the bottom until they build a track record. The sort uses this value; the display does not.
Confidence %
(350 − RD) / 320
The Confidence column maps RD onto a 0–100% scale within its valid Glicko-2 range. 100% means RD is at the floor (maximum certainty); 0% means RD is at the cap (essentially unrated). A rating at RD 30 reads 100%, at RD 80 reads 84%, at RD 150 reads 63%. Hover the cell to see the raw RD.

What Affects Your Rating

After each tournament, the Glicko-2 algorithm looks at every match you played and adjusts your rating based on three factors:

1

Did you win or lose?

Wins increase your rating. Losses decrease it. Draws have a small effect that depends on the opponent's strength.

2

How strong was your opponent?

Beating a 1,800-rated player matters more than beating a 1,200-rated player. Losing to a strong player costs you less than losing to a weak one.

3

How confident is the system about both ratings?

If either player has a high RD (few events), the result causes a bigger rating swing. As both players play more, the RD shrinks and results cause smaller, more precise adjustments.

Event-Weighted K-Factor

HRI runs pure Glicko-2 with real opponent ratings, then scales the resulting rating change by a multiplier based on tournament tier, elimination round, and whether you won the tournament. This preserves the statistical integrity of the engine — we do not feed inflated opponent ratings back into Glicko-2 — while still rewarding harder events more than easier ones.

The total multiplier for any match is: Tier Multiplier + Top Cut Bonus + Winner Bonus

1. Tournament Tier Multiplier

Every match at a tournament gets a baseline multiplier based on the tier. Higher-tier events have stronger fields, so your rating change from each match is amplified.

Tier Multiplier Description
Planetary 1.0x Baseline. No modification to your rating change.
Sector 1.15x Regional talent pool. Rating changes amplified 15%.
Regional 1.3x Top players travel for these. Rating changes amplified 30%.
Galactic 1.5x Best players in the world. Rating changes amplified 50%.

2. Top Cut Multiplier

Making the elimination bracket matters. Top-cut matches get a progressive multiplier bonus added on top of the tier multiplier. The deeper you go in the bracket, the more each match is amplified.

Round Bonus Added to tier multiplier
Swiss Rounds +0 No top-cut bonus. Only tier multiplier applies.
Quarterfinals +0.15 You made top 8. Playing the best in the room.
Semifinals +0.25 Top 4. Win-or-go-home against proven players.
Finals +0.35 The championship match. Maximum stakes.

3. Tournament Winner Bonus

The player who wins the Finals gets an additional +0.15 added to their multiplier for that match. Winning the whole tournament is the strongest signal of skill.

Finals Winner
+0.50
+0.35 (Finals) + 0.15 (Winner Bonus) added to tier multiplier. The champion's Finals match is amplified more than the runner-up's.
Finals Runner-Up
+0.35
+0.35 (Finals) only, added to tier multiplier. Still a significant match, but no winner bonus.

Guardrails

Rating systems occasionally produce extreme outputs — a badly-seeded player's first event can generate a 700-point correction that is mathematically correct but visually shocking. HRI applies two guardrails to catch these without compromising the integrity of Glicko-2's self-correction.

RD Bounds
30 – 350
Rating Deviation is clamped to a sane range. RD cannot drop below 30 (absolute maximum confidence) or rise above 350 (effectively unrated).
Multiplier Bonus Cap
±150
The extra points added by the event multiplier are capped at ±150. The raw Glicko-2 change itself is uncapped — we want the engine to correct bad ratings fast. Tournament rows where this clamp fires are marked "(clamped)" so you can audit them. If the clamp fires on more than 0.5% of updates, we treat that as a data-quality signal and investigate.

Putting It All Together

Here's the total multiplier for every combination. Your rating change from Glicko-2 is multiplied by this factor after the calculation runs.

Situation Total Breakdown
Planetary Swiss 1.0x 1.0 tier (no modification)
Planetary QF 1.15x 1.0 tier + 0.15 QF
Planetary Finals (winner) 1.5x 1.0 tier + 0.35 finals + 0.15 winner
Sector Swiss 1.15x 1.15 tier
Regional Swiss 1.3x 1.3 tier
Regional Finals (winner) 1.8x 1.3 tier + 0.35 finals + 0.15 winner
Galactic Swiss 1.5x 1.5 tier
Galactic Finals (champion) 2.0x 1.5 tier + 0.35 finals + 0.15 winner. The highest-value match in the system.

Where HRI Deviates From Pure Glicko-2

Pure Glicko-2 treats a rating period as a single batch: all matches within the period use each player's start-of-period rating, and ratings update once at the end. HRI makes two deliberate deviations from this, documented here for transparency.

Phased top-cut updates
Swiss rounds are processed together as one Glicko-2 batch. Each elimination round (Quarterfinals, Semifinals, Finals) is then processed as its own batch, using ratings updated by the previous phase. This means top-cut matches see the Swiss result reflected in the players' ratings. Pure Glicko-2 would ignore that signal within the same rating period.
Post-hoc K-factor scaling
We run Glicko-2 with real opponent ratings, then multiply the resulting rating change by the event weight (tier + top-cut + winner bonus). This is different from adjusting opponent ratings before the Glicko-2 run (which would be the "opponent inflation" pattern that distorts the engine). HRI avoids that pattern.

Rating Deviation Bands

Every rating has a deviation (RD). All players appear on a single leaderboard, sorted by their conservative rating (see "Leaderboard Sort" above) — so provisional players naturally fall toward the bottom until they build a record. For quick visual context, RD is grouped into three color-coded bands:

Established
RD ≤ 80
Enough events that the rating is trusted. Confidence displays in green.
Developing
80 < RD ≤ 120
Still stabilizing. Typical for players with a handful of events. Confidence displays in amber.
Provisional
RD > 120
Very few events — the rating could move sharply. Confidence displays in gray.
Time decay
If you stop playing, your rating itself stays the same but RD gradually increases — the system becomes less certain the longer it's been since you played. Compete again and RD drops back.

Rating Formats

HRI tracks independent ratings for each format. Your Premier rating is not affected by your Limited results, and vice versa.

O

Overall — the canonical cross-format rating

The Galactic Championship requires every player to compete in BOTH Premier and Limited. Overall is built to seed that event — the equal-weighted average of your Premier and Limited ratings:

Overall rating = (Premier + Limited) / 2
Overall RD = (Premier_RD + Limited_RD) / 2

If you've only played one format (as of 2026-04-17), the missing format is filled with a correlation-shrinkage prior, not the population mean. Card-game skill transfers: the empirical correlation between Premier and Limited ratings among players who play both is ρ = 0.65. So we estimate your missing-format rating by pulling your known rating partway toward 1,500:

missing_prior = 1,500 + 0.65 × (known_rating − 1,500)
missing_RD = 150

A 1,900 Premier player with no Limited history gets a Limited prior of 1,760, for an Overall of 1,830. Not 1,700, the way it used to be. The old 1,500 / RD 350 prior assumed zero correlation between formats, which let mediocre dual-format players outrank elite single-format players on Overall — the v3.4 change fixes that pathology while still discounting unmeasured formats via RD (conservative sort penalizes high-RD).

Eternal is intentionally excluded. Players with only Eternal experience don't appear on the Overall leaderboard — that format has its own.

P

Premier

Constructed format. Bring your own deck. The most commonly played competitive format. Feeds Overall.

L

Limited

Sealed or draft format. Build your deck from packs at the event. Feeds Overall.

E

Eternal

All cards legal. A wider card pool for experienced players. Tracked independently; not part of Overall.

Which Events Count?

Only sanctioned competitive events are included in HRI ratings.

Included
Planetary Qualifier, Sector Qualifier, Regional Championship, Galactic Championship, Galactic Open. All three formats: Premier, Limited, Eternal.
Excluded
Store Showdowns, Minor Tournaments, Online Tournaments, Weekly Play, Special Events.

Data Source

All match data comes from Melee.gg, the official tournament platform for Star Wars: Unlimited. Tournament discovery and tier/format classification comes from the SWU Competitive Hub. HRI processes every round of every qualifying tournament, including all Swiss pairings and the full elimination bracket.

Database-level constraints prevent malformed matches from entering the system: a player cannot appear as their own opponent, and the same match cannot be counted twice even if imported with players in swapped positions.

Common Questions

Why is my rating lower than someone with a worse record?

Record alone doesn't determine rating. A 10-2 record against top-100 players at a Regional is worth more than a 10-2 record at a small Planetary. Glicko-2 accounts for opponent strength; HRI's event weighting adds tournament tier, top-cut depth, and tournament victory.

Why are some players near the bottom with low Confidence %?

Everyone appears on the same leaderboard, but it sorts by Rating − 2 × RD — the lower bound of the 95% credible interval. A player with RD 200 sits 400 points below their raw rating for sorting purposes, so provisional players fall naturally to the bottom until they build a track record. Their Confidence % reflects the RD level, so a low number means "we don't yet know where this player truly belongs."

How do I improve my rating the fastest?

Win matches at high-tier events, make top cut, and win the tournament. The single highest-value thing you can do is win the Finals at a Galactic Championship (2.0x multiplier). Consistency at Sector and Regional events also builds your rating steadily. Playing more events tightens your RD, which raises your ranked-leaderboard position even if your raw rating stays flat.

Why are Premier, Limited, and Eternal rated separately?

Different formats test different skills. A top Premier player might be average at Limited. Separate tracks give an accurate picture of skill per format. The Overall view combines them via a match-weighted average — see the Formats section above.

How is my Overall rating calculated?

If you have both a Premier and a Limited rating, Overall is the equal-weighted average: (Premier + Limited) / 2. If you're rated in only one format, the missing format is filled via a correlation-shrinkage prior: 1,500 + 0.65 × (known − 1,500). See the Overall section above for the full math. Eternal isn't included.

I'm only rated in Premier. Why does my Overall rank look so different from my Premier rank?

Overall is a Galactic Championship readiness rating, not a "best format" rating. The Galactic Championship tests both Premier and Limited, so Overall blends them. If you haven't played Limited, the system estimates your Limited rating from your Premier rating (they correlate 0.65 empirically), then averages. Your Overall number stays close to your Premier, but the conservative sort (rating − 2 × RD) still discounts unmeasured formats because we're less certain about them. To close the gap completely, play a Limited event — even one tightens the RD on your composite.

What does the "(clamped)" marker on a tournament mean?

The event multiplier added more than ±150 points of bonus on top of the raw Glicko-2 change, and the guardrail caught it. This usually happens the first time a strong new player hits a high-tier event. Your raw Glicko-2 change is preserved — the clamp only limits the extra points from the multiplier.

What is Glicko-2?

Glicko-2 is a rating system created by Professor Mark Glickman at Harvard. It improves on traditional Elo by tracking how confident the system is in each rating. The full paper is at glicko.net.

System Parameters

For the statistically curious or the implementation-auditing: HRI's published parameters.

Parameter Value Notes
Default rating 1500 Standard Glicko-2 default.
Default RD 250 Allows first tournament to tighten uncertainty faster than RD = 350.
Default volatility 0.06 Glickman's recommended default.
Tau (volatility constraint) 0.5 Within Glickman's recommended 0.3 – 1.2 range. Lower values dampen volatility runaway.
RD floor / cap 30 / 350 Bounds on rating deviation.
Established RD band 80 RD at or below 80 classifies as Established (green Confidence). Visual only — does not gate leaderboard visibility.
Multiplier bonus cap ±150 Maximum extra points the event weighting can add on top of the raw Glicko-2 change.
Observed clamp rate 0.031% How often the ±150 cap actually fires. If this rises above 0.5% we investigate — it's the system's self-check for bad inputs.
RD decay period 30 days Per-player: RD increases by one Glicko-2 decay step for every 30 days of inactivity in that format.