New Feature

Introducing Glicko-2 Driver Ratings

A New Way to Rank Autocross Drivers

AutocrossRank has always used our ARS (AutocrossRank Score) system to rank drivers —
a margin-based approach that rewards you for beating competitors by larger time margins
within your class. That system isn't going anywhere, but we've now added a second ranking
algorithm alongside it: Glicko-2.

You can switch between them on the Rankings page using the Algorithm dropdown,
and on any driver's profile page under their Scoring Breakdown.


What Is Glicko-2?

Glicko-2 is a rating system developed by statistician Mark Glickman, and it's the same
family of algorithms used to rate chess players, online gamers, and competitive sports
leagues around the world. The core idea is simple: your rating reflects how you perform
head-to-head against your class competitors, adjusted for how well-established those
competitors' ratings are
.

Every driver has three numbers:

  • Rating — your skill estimate, starting at 1500 for everyone
  • Rating Deviation (RD) — how confident the system is in your rating. New drivers
    start at 350 (high uncertainty). Active drivers converge toward 50–100 over time.
  • Volatility — how consistent your performances are event to event

Your position on the Glicko-2 leaderboard is determined by a conservative estimate:
Rating - 2 × RD. This intentionally ranks drivers with high uncertainty lower until
they've proven themselves with more results.


How It Works at Each Event

Each autocross event is treated as a rating period. Within your class, the system
creates a head-to-head record for every pair of drivers: the faster driver wins, the
slower driver loses. That's it — no margin weighting, just who beat whom.

After all pairings are resolved, your rating updates based on:

  1. How you did against each opponent (wins and losses)
  2. How established each opponent's rating is (high-RD opponents count for less)
  3. How surprising your performance was relative to what was expected

Examples

Example 1 — New Driver Upsets an Established Driver

Sarah has just entered her first SCCA event. She starts at 1500 ± 350 (high
uncertainty — the system doesn't know much about her yet).

Mike is a seasoned competitor with a rating of 1750 ± 60 (well established over
many events).

Sarah beats Mike in class. Because Mike is a strong, well-known quantity, this is a
significant upset. Sarah's rating jumps considerably — maybe up to 1620. Her RD also
drops since we now know more about her. Mike's rating drops slightly but not dramatically;
one loss to an unknown driver is weighted cautiously.


Example 2 — Expected Result, Small Movement

Tom is rated 1800 ± 55 and beats Dave who is rated 1600 ± 65.

The system expected Tom to win. Because this result was predictable, neither rating moves
much — Tom might go to 1808, Dave to 1592. The system doesn't reward you heavily
for winning battles you were already expected to win.


Example 3 — Solo Competitor

Lisa shows up as the only driver in her class at a regional event. Since there's no
one to compare against, no head-to-head games are recorded. Her rating stays unchanged,
but her RD increases slightly — the system becomes a little less certain about her
skill level because she sat out a rating period without competing.


Example 4 — The Leaderboard Conservative Score

Two drivers are compared for a leaderboard spot:

Driver Rating RD Conservative Score (Rating - 2×RD)
Alex 1900 150 1600
Jordan 1750 45 1660

Even though Alex has a higher raw rating, Jordan ranks higher. Alex has only attended a
handful of events and the system isn't confident in that 1900 yet. Jordan has a large
history and their 1750 is solid. The conservative score protects the leaderboard from
being dominated by drivers who got lucky at one or two big events.


ARS vs Glicko-2 — Which Is Right?

They measure different things:

ARS (Margin Scoring) Glicko-2
Rewards Beating competitors by larger margins Beating strong, established competitors
Event contribution Larger time gaps = bigger score swings Win/loss only — margins don't matter
New driver experience Starts earning immediately Starts at 1500 with high uncertainty
Leaderboard logic Cumulative score Conservative skill estimate

Neither is definitively "better" — they highlight different aspects of competitive
performance. We'll continue running both and are interested in how the community uses them.


Glicko-2 was developed by Mark Glickman (Boston University). For the technical details
see glicko.net.