dunkirk.sh
a geicko-2 based round robin ranking system designed to test c++ battleship submissions
battleship.dunkirk.sh
fix: implement proper Glicko-2 rating periods to eliminate last-submitter bias
The previous implementation updated Glicko-2 ratings after each individual
match, violating the system's core assumption of rating periods with 10-15+
games. This caused path-dependency where identical algorithms submitted at
different times received different final ratings based on opponent rating
states during their matches.
Changes:
- Refactored RecalculateAllGlicko2Ratings() to use proper rating periods
- Batches all match results per player together
- Uses opponent ratings at the start of the rating period
- Updates each player's rating once with all results
- Modified RunRoundRobinMatches() to store matches first, then recalculate
ratings once at the end (proper rating period)
- Removed per-match UpdateGlicko2Ratings() calls
- Added manual recalculation command: ./battleship-arena recalculate-ratings
- Updated leaderboard sorting: rating DESC → wins DESC → avg_moves ASC
- Pending/broken entries use avg_moves=999 to sort to bottom
Live updates still work - SSE broadcasts match progress in real-time.
Glicko ratings update at round-robin completion using proper rating periods.
References:
- Glicko-2 paper: http://www.glicko.net/glicko/glicko2.pdf
- Instant Glicko-2 implementation notes:
https://gist.github.com/gpluscb/302d6b71a8d0fe9f4350d45bc828f802
💘 Generated with Crush
Assisted-by: Claude Sonnet 4.5 via Crush <crush@charm.land>
+12
AGENTS.md
+12
AGENTS.md
···
12
12
- **scp.go** - SCP upload middleware for file submissions
13
13
- **worker.go** - Background processor (runs every 30s)
14
14
15
+
## Glicko-2 Rating System
16
+
17
+
**Important**: The system uses Glicko-2 ratings with **proper rating periods** to avoid last-submitter bias:
18
+
19
+
- All matches in a round-robin are stored first
20
+
- Ratings update **once at the end** using all match results together (proper rating period)
21
+
- This eliminates path-dependency where identical algorithms get different ratings based on submission order
22
+
- Each player's rating considers ALL their opponents' ratings at the start of the rating period
23
+
- Glicko-2 expects 10-15+ games per rating period - our round-robin satisfies this
24
+
25
+
**Manual recalculation**: Run `./battleship-arena recalculate-ratings` or `make recalculate-ratings` to recompute all ratings from scratch.
26
+
15
27
## File Upload
16
28
17
29
Students upload via SCP:
+16
-10
Makefile
+16
-10
Makefile
···
48
48
@echo "Building for production..."
49
49
@CGO_ENABLED=1 go build -ldflags="-s -w" -o bin/battleship-arena ./cmd/battleship-arena
50
50
51
+
# Recalculate all Glicko-2 ratings from scratch
52
+
recalculate-ratings: build
53
+
@echo "Recalculating all Glicko-2 ratings..."
54
+
@./bin/battleship-arena recalculate-ratings
55
+
51
56
# Show help
52
57
help:
53
58
@echo "Available targets:"
54
-
@echo " build - Build the server"
55
-
@echo " run - Build and run the server"
56
-
@echo " clean - Clean build artifacts"
57
-
@echo " test - Run tests"
58
-
@echo " gen-key - Generate SSH host key"
59
-
@echo " fmt - Format code"
60
-
@echo " lint - Lint code"
61
-
@echo " deps - Update dependencies"
62
-
@echo " build-prod - Build optimized production binary"
63
-
@echo " help - Show this help"
59
+
@echo " build - Build the server"
60
+
@echo " run - Build and run the server"
61
+
@echo " clean - Clean build artifacts"
62
+
@echo " test - Run tests"
63
+
@echo " gen-key - Generate SSH host key"
64
+
@echo " fmt - Format code"
65
+
@echo " lint - Lint code"
66
+
@echo " deps - Update dependencies"
67
+
@echo " build-prod - Build optimized production binary"
68
+
@echo " recalculate-ratings - Recalculate all Glicko-2 ratings from scratch"
69
+
@echo " help - Show this help"
battleship-arena
battleship-arena
This is a binary file and will not be displayed.
+13
cmd/battleship-arena/main.go
+13
cmd/battleship-arena/main.go
···
61
61
if err := initStorage(cfg); err != nil {
62
62
log.Fatal(err)
63
63
}
64
+
65
+
// Check for special commands
66
+
if len(os.Args) > 1 {
67
+
switch os.Args[1] {
68
+
case "recalculate-ratings":
69
+
log.Println("Recalculating all Glicko-2 ratings from scratch...")
70
+
if err := storage.RecalculateAllGlicko2Ratings(); err != nil {
71
+
log.Fatalf("Failed to recalculate ratings: %v", err)
72
+
}
73
+
log.Println("✓ Ratings recalculated successfully")
74
+
return
75
+
}
76
+
}
64
77
65
78
server.InitSSE()
66
79
server.SetConfig(cfg.AdminPasscode, cfg.ExternalURL)
+9
-8
internal/runner/runner.go
+9
-8
internal/runner/runner.go
···
329
329
map[int]string{newSub.ID: newSub.Username, opponent.ID: opponent.Username}[winnerID])
330
330
}
331
331
332
-
matchID, err := storage.AddMatch(newSub.ID, opponent.ID, winnerID, player1Wins, player2Wins, avgMoves, avgMoves)
332
+
_, err := storage.AddMatch(newSub.ID, opponent.ID, winnerID, player1Wins, player2Wins, avgMoves, avgMoves)
333
333
if err != nil {
334
334
log.Printf("Failed to store match result: %v", err)
335
-
} else {
336
-
if err := storage.UpdateGlicko2Ratings(newSub.ID, opponent.ID, player1Wins, player2Wins); err != nil {
337
-
log.Printf("Glicko-2 update failed: %v", err)
338
-
} else {
339
-
recordRatingSnapshot(newSub.ID, int(matchID))
340
-
recordRatingSnapshot(opponent.ID, int(matchID))
341
-
}
342
335
}
343
336
}
344
337
345
338
log.Printf("✓ Round-robin complete for %s (%d matches)", newSub.Username, totalMatches)
339
+
340
+
// Update Glicko-2 ratings using proper rating periods (batch all matches together)
341
+
log.Printf("Updating Glicko-2 ratings (proper rating period)...")
342
+
if err := storage.RecalculateAllGlicko2Ratings(); err != nil {
343
+
log.Printf("Failed to update Glicko-2 ratings: %v", err)
344
+
} else {
345
+
log.Printf("✓ Glicko-2 ratings updated")
346
+
}
346
347
}
347
348
348
349
func recordRatingSnapshot(submissionID, matchID int) {
+105
-3
internal/storage/database.go
+105
-3
internal/storage/database.go
···
191
191
192
192
func GetLeaderboard(limit int) ([]LeaderboardEntry, error) {
193
193
// Get submissions with matches
194
+
// Rankings use Glicko-2 with proper rating periods:
195
+
// - All round-robin matches are batched together before rating updates
196
+
// - This prevents last-submitter bias from path-dependent rating changes
194
197
query := `
195
198
SELECT
196
199
s.username,
···
216
219
350.0 as rd,
217
220
0 as total_wins,
218
221
0 as total_losses,
219
-
0.0 as avg_moves,
222
+
999.0 as avg_moves,
220
223
s.upload_time as last_played,
221
224
1 as is_pending,
222
225
0 as is_broken
···
234
237
0 as rd,
235
238
0 as total_wins,
236
239
0 as total_losses,
237
-
0.0 as avg_moves,
240
+
999.0 as avg_moves,
238
241
s.upload_time as last_played,
239
242
0 as is_pending,
240
243
1 as is_broken
241
244
FROM submissions s
242
245
WHERE s.is_active = 1 AND s.status = 'compilation_failed'
243
246
244
-
ORDER BY is_broken ASC, is_pending ASC, rating DESC, total_wins DESC
247
+
ORDER BY is_broken ASC, is_pending ASC, rating DESC, total_wins DESC, avg_moves ASC
245
248
LIMIT ?
246
249
`
247
250
···
732
735
)
733
736
return err
734
737
}
738
+
739
+
// RecalculateAllGlicko2Ratings recalculates all Glicko-2 ratings from scratch
740
+
// using proper rating periods where all matches for a player are batched together
741
+
func RecalculateAllGlicko2Ratings() error {
742
+
// Reset all active submissions to initial ratings
743
+
_, err := DB.Exec(`
744
+
UPDATE submissions
745
+
SET glicko_rating = 1500.0, glicko_rd = 350.0, glicko_volatility = 0.06
746
+
WHERE is_active = 1 AND status = 'completed'
747
+
`)
748
+
if err != nil {
749
+
return err
750
+
}
751
+
752
+
// Get all active player IDs
753
+
var playerIDs []int
754
+
rows, err := DB.Query("SELECT id FROM submissions WHERE is_active = 1 AND status = 'completed'")
755
+
if err != nil {
756
+
return err
757
+
}
758
+
for rows.Next() {
759
+
var id int
760
+
if err := rows.Scan(&id); err != nil {
761
+
return err
762
+
}
763
+
playerIDs = append(playerIDs, id)
764
+
}
765
+
rows.Close()
766
+
767
+
// For each player, collect ALL their match results and update once (proper rating period)
768
+
for _, playerID := range playerIDs {
769
+
// Get player's current rating
770
+
var rating, rd, volatility float64
771
+
err := DB.QueryRow(
772
+
"SELECT glicko_rating, glicko_rd, glicko_volatility FROM submissions WHERE id = ?",
773
+
playerID,
774
+
).Scan(&rating, &rd, &volatility)
775
+
if err != nil {
776
+
continue
777
+
}
778
+
779
+
// Collect ALL match results for this player in this rating period
780
+
var results []Glicko2Result
781
+
782
+
rows, err := DB.Query(`
783
+
SELECT
784
+
CASE WHEN player1_id = ? THEN player2_id ELSE player1_id END as opponent_id,
785
+
CASE WHEN player1_id = ? THEN player1_wins ELSE player2_wins END as my_wins,
786
+
CASE WHEN player1_id = ? THEN player2_wins ELSE player1_wins END as opponent_wins
787
+
FROM matches
788
+
WHERE (player1_id = ? OR player2_id = ?) AND is_valid = 1
789
+
ORDER BY timestamp ASC
790
+
`, playerID, playerID, playerID, playerID, playerID)
791
+
792
+
if err != nil {
793
+
continue
794
+
}
795
+
796
+
for rows.Next() {
797
+
var opponentID, myWins, opponentWins int
798
+
if err := rows.Scan(&opponentID, &myWins, &opponentWins); err != nil {
799
+
continue
800
+
}
801
+
802
+
// Get opponent's rating at the START of this rating period (not current)
803
+
var oppRating, oppRD float64
804
+
err := DB.QueryRow(
805
+
"SELECT glicko_rating, glicko_rd FROM submissions WHERE id = ?",
806
+
opponentID,
807
+
).Scan(&oppRating, &oppRD)
808
+
if err != nil {
809
+
continue
810
+
}
811
+
812
+
totalGames := myWins + opponentWins
813
+
score := float64(myWins) / float64(totalGames)
814
+
815
+
results = append(results, Glicko2Result{
816
+
OpponentRating: oppRating,
817
+
OpponentRD: oppRD,
818
+
Score: score,
819
+
})
820
+
}
821
+
rows.Close()
822
+
823
+
// Update this player's rating based on ALL results at once (proper rating period)
824
+
if len(results) > 0 {
825
+
player := Glicko2Player{Rating: rating, RD: rd, Volatility: volatility}
826
+
newPlayer := updateGlicko2(player, results)
827
+
828
+
DB.Exec(
829
+
"UPDATE submissions SET glicko_rating = ?, glicko_rd = ?, glicko_volatility = ? WHERE id = ?",
830
+
newPlayer.Rating, newPlayer.RD, newPlayer.Volatility, playerID,
831
+
)
832
+
}
833
+
}
834
+
835
+
return nil
836
+
}