Hi all,
I've been working on a couple of stat projects this past week that I think could bring some added depth to how we understand performance on VLR at the player and team level, and I wanted to share them here and see what people think!
1. RAR - Rounds Above Replacement
RAR (Inspired by Baseball's WAR stat) is built directly on top of VLR's existing R² metric. It estimates how many total rounds a player contributes above a replacement-level pro in a given event/match, where "replacement" is defined as the 20th percentile of R² performers in that event who meet the minimum rounds played threshold (min 20% rounds played). You can think of RAR as "how many more rounds did this player help win compared to a replacement-level pro?"
The formulas:
Per RAR = R² – Replacement_R² -> Per-round efficiency above replacement (In a given round how much did this player contribute above a replacement level pro to winning)
Total RAR = (R² – Replacement_R² x Rounds Played) -> Total impact above replacement (In a given match and/or event how many rounds did this player contribute above a replacement level pro to winning)
Most events have a replacement R² around ~0.85
So:
- Per RAR tells you how much better a player performs per round than a replacement
- Total RAR shows how much value they added overall across the rounds they played
Together, they evaluate and reward both efficiency and availability. RAR complements R² by capturing cumulative impact, not just quality per round, and I think it’d make a great addition as a sortable column next to R² on player stat tables.
Quick snapshot from Masters Toronto (before today's Grand Final):
- *Top Total RAR: Spring (126), Alfajer (112), Chronicle (101)
- *Top Per RAR: t3xture (0.38), Spring (0.36), Verno (0.33)
2. Cross-Regional ELO Team Rankings
Right now, VLR team rankings are evaluated intra-regionally with it being difficult to compare teams across different regions in real time, even outside of LANs.
So, taking inspiration from Chess.com and Nate Silver's methodologies I built a global ELO-based system that tracks team performances across all of the 2025 season, enabling direct comparisons between regions throughout the season - not just at international events.
Here's how it works:
1) Teams begin the season with a starting ELO (with 1500 being considered "Average")
- Teams who qualified for Valorant Champions 2024 are seeded based on their LAN performance
- Other Tier 1 teams who did not qualify are seeded using their implied betting odds from their first couple matches
2. Match-by-Match ELO Updates
Each match updates team ratings based on:
- Win/loss outcome factoring in the Margin of Win
- Expected result (based on pre-match ELOs)
- Match importance (with event specific K-factors being used to differentiate the importance of a match in Kickoffs v. Masters)
3. Time Decay, Regional Adjustments, Bayesian Smoothing
To adjust for a variety of factors outside of matches, the model applies:
- Time decays to teams who haven't played in a while to account for teams who don't qualify for international events or meta changes that can occur between events
- Regional Bias multipliers calculated from how regions performed relative to expectation after every international event and applied to the next regional event to enable real-time cross-region comparisons and account for difference in regional strengths
- Short-term Bayesian pull to smooth ratings from their implied 2025 initial ELOs to their current ELOs to stabilize extreme fluctuations
The result is a dynamic, constant up-to-date global leaderboard that reflects team form and allows for teams across regions to properly be compared against each other.
If people are interested, I can post a top 10 snapshot (I was personally surprised to find that 2 of the top 4 did not qualify for Toronto) or share auto-generated graphs on how your favorite team's ELO has changed throughout the season.
Would love thoughts or feedback on either of these ideas!
Edit 1: Updated the Masters Toronto Top Total RAR players using correct numbers
Discord