How the Desk Predicts
A STADIO Desk prediction is an editorial judgment designed to make a match easier to read. The Desk reviews recent matches, head-to-head history, tournament stage, and the game’s own selection structure before choosing a direction and writing down the reason.
The goal is to fix the view before play, score it by the same rule afterward, and keep measuring the quality of that judgment.
Locked before play
A published Desk prediction is finalized before the match begins. The version that readers could see beforehand is always the one that gets scored.
If a pre-match fact such as the lineup or schedule changes, the prediction can be reviewed only before the lock. A revised version is labeled on the match page so it does not look as if there was only ever one judgment.
Evidence first
A prediction combines a team name and percentage with connected form, head-to-head history, tournament stage, and series length. It then uses game-specific viewing angles: drafting and adaptation in LoL, map vetoes and sides in CS2, and maps and agent compositions in VALORANT.
A thin sample keeps the conclusion proportionally cautious, and the write-up separates what can be supported from what remains unknown.
Scored on results
Predictions that name a team are checked against the actual winner and recorded as a hit or miss, with hits and misses from the same period shown together. A matchup where neither side has a meaningful edge and no team is selected is excluded from hit-rate scoring.
The accumulated record is available by game on the prediction record page. When the sample is small, the number of scored matches is always shown alongside the rate.
The record so far
This is the current record of predictions locked before play and checked against the result. Hits and misses have equal weight here, and every figure updates with the underlying data.
Graded period 2026-07-04–2026-09-01
Only the 622 scored predictions form the hit-rate denominator. The 23 predictions awaiting a result and 60 matchups where no team was selected are excluded.
Brier measures the gap between a predicted probability and the result; lower is more accurate. A score of 0.25 is the coin-flip benchmark from calling every match 50–50.
Record by game
| Game | Scored | Hits | Misses | Hit rate | Brier |
|---|---|---|---|---|---|
| LoL | 295 | 194 | 101 | 65.8% | 0.2198n=295 |
| VALORANT | 222 | 129 | 93 | 58.1% | 0.2424n=222 |
| CS2 | 105 | 71 | 34 | 67.6% | 0.2227n=105 |
Recent form
The 30 most recently settled predictions are compared with everything before them under the same rule.| Period | Scored | Hits | Misses | Hit rate | Brier |
|---|---|---|---|---|---|
| Latest 30 | 30 | 22 | 8 | 73.3% | 0.1885n=30 |
| Earlier record | 592 | 372 | 220 | 62.8% | 0.2304n=592 |
Where the probability landed — and where it did not
The hit with the lowest Brier and the miss with the highest Brier, selected automatically from the same ledger.| Case | Match | Pre-match call | Result | Brier | Rationale |
|---|---|---|---|---|---|
| Lowest Brier | LoLBilibili Gaming vs LYON | Bilibili Gaming 92% | Hits | 0.0064 | Read rationale → |
| Largest miss | LoLTeam Secret Whales vs Top Esports | Top Esports 90% | Misses | 0.8100 | Read rationale → |
Sources and limits
Schedules, participants, and final scores are assembled from information published by tournament operators and the match feeds connected to STADIO. Names, time zones, and series records are normalized into one format before they appear on a match page. If an organizer changes an earlier announcement or corrects a result, the record here may also change after that revision has been verified.
Recent form includes only matches that can be connected with confidence, so coverage is not equally deep for every team or competition. The write-up identifies those limits, and a missing map, an unconfirmed roster, or an incomplete historical series is covered only as far as it is confirmed.
What stays private
Detailed weights and probability calculations are not published. The way team evidence and game-specific variables are combined into one judgment is a skill this desk has built up through repeated work.
The parts a reader needs to verify remain public: the selected side, probability, written rationale, actual result, and score. The Desk should be judged by the consistency of those decisions and results.
Automation, AI assistance, and human review
Automation or AI assistance may be used for schedules, results, statistical aggregation, comparison, research organization, and parts of an analysis draft. Player and team aggregates, and pages without a named author, do not carry a claim of page-level human review.
Search publication is limited to the Korean canonical match page with a pre-match Desk analysis and a result-scoring record. Human review is claimed only when an author or reviewer is explicitly identified on that page.
How to read it
A Desk prediction is a viewing reference, not a guaranteed outcome. A match contains more variables than any public dataset, including roster changes, patches, and day-of performance that cannot all be known in advance.
Read the rationale alongside the direction and compare it with your own view. After the match, return to the prediction record to see how the same judgment was scored.
The prediction scorecard shows the fixed view, tied to its reasoning and the final result.