feat: add canonical matchmaking rating engine

This commit is contained in:
Josh Creek
2026-08-31 20:25:07 +01:00
parent 7643dbc439
commit 1793eb7621
3 changed files with 185 additions and 1 deletions
+1 -1
View File
@@ -1198,7 +1198,7 @@ the local/CI/community transport, not a silent production fallback.
| 8.18 `[D:8.5,8.14,8.17]` | Horizontally replicated matcher: Redis candidates, then PostgreSQL serializable proposal/participant fence, then cache cleanup/repair | Forced loss of the last acknowledged Redis write, retries, worker death and failover cannot claim a player into two proposals/matches |
| 8.19 `[D:8.18]` | Casual policy: proposal composition above; >=1 human/team, exhaustive rating-balanced teams, bots after 60 s, opt-in kickoff-only backfill, 30 s reconnect and defined backfill/casual penalties | Every 26-human shape is tested; no mid-play replacement; declined/backfill participant gets no excluded rating/cooldown; original leaver gets only documented outcome/cooldown |
| 8.20 `[D:8.18]` | Ranked policy: exactly six verified solo humans, no bots/backfill, only `ArenaRegistry.random`; define initial no-show, proposal timeout and reconnect/abandon transitions | Ranked rejects parties/bots/backfill/elevated arenas; every pre-live failure returns five innocent players with original precedence and applies no rating |
| 8.21 `[D:8.5,8.20]` | Exact Glicko-2 equations from `docs/MATCHMAKING.md`: 1500/350/0.06/tau .5, ranked 1/3 and casual 1/N human-opponent weights, daily inactivity, immutable snapshot/lock order, draws/OT/abandons/cancellation | Canonical plus project 26-human/3v3 golden vectors pass; concurrent results serialize without order bias; clients have no rating-write path |
| 8.21 `[D:8.5,8.20]` | **IN PROGRESS.** Pure Go rating core implements canonical Glicko-2, daily inactivity, ranked 1/3 and casual 1/N human-opponent weights, and deterministic opponent ordering | `server/domain/rating.go` has canonical/inactivity/weight/invalid-input fixtures; PostgreSQL snapshot locking, draws/OT/abandons, seasons and concurrent result transaction tests remain |
| 8.22 `[D:8.21]` | First ten ranked games provisional; casual rating hidden; ranked tiers derived from authoritative stored values | Matchmaking uses provisional rating/RD; UI visibility changes exactly on result ten without rewriting history |
| 8.23 `[D:8.21]` | Ranked-only 12-week exactly-once soft season: compress 25% toward 1500, RD >=200 capped 350, retain volatility/history; casual remains continuous | Retried/concurrent rollover applies once, never touches casual, and preserves every rating event |
| 8.24 `[D:8.9,8.20,8.21]` | Ranked reconnect/abandon: match-scoped authorisation, 60 s reclaim, server-owned connection generations, then abandoner loss and rolling 7-day 5 m/15 m/1 h/24 h cooldown | Reconnect works through backend outage and fences old peer; grace has no penalty; expiry outcome/escalation is deterministic and auditable |
+135
View File
@@ -0,0 +1,135 @@
package domain
import (
"fmt"
"math"
"sort"
"time"
)
const (
GlickoScale = 173.7178
GlickoTau = 0.5
GlickoEpsilon = 0.000001
GlickoInitialRating = 1500.0
GlickoInitialRD = 350.0
GlickoInitialVolatility = 0.06
)
type Rating struct {
Value float64
RD float64
Volatility float64
LastRatedAt time.Time
}
type Opponent struct {
PlayerID string
Rating Rating
Weight float64
Score float64
}
// UpdateRating applies canonical Glicko-2 to one player's immutable pre-match
// rating snapshot. Weight is 1/3 for ranked 3v3 and 1/N for casual's N human
// opponents; bots are simply omitted by the caller.
func UpdateRating(current Rating, opponents []Opponent, now time.Time) (Rating, error) {
if err := validateRating(current); err != nil { return Rating{}, err }
if len(opponents) == 0 { return advanceInactivity(current, now), nil }
for _, opponent := range opponents {
if err := validateRating(opponent.Rating); err != nil { return Rating{}, err }
if opponent.Weight <= 0 || opponent.Score < 0 || opponent.Score > 1 { return Rating{}, fmt.Errorf("invalid opponent weight or score") }
}
working := advanceInactivity(current, now)
mu, phi := toScale(working.Value, working.RD)
varianceInverse, deltaSum := 0.0, 0.0
for _, opponent := range opponents {
oppMu, oppPhi := toScale(opponent.Rating.Value, opponent.Rating.RD)
g := glickoG(oppPhi)
expected := expectedScore(mu, oppMu, g)
varianceInverse += opponent.Weight * g * g * expected * (1 - expected)
deltaSum += opponent.Weight * g * (opponent.Score - expected)
}
if varianceInverse <= 0 { return Rating{}, fmt.Errorf("opponent information has zero variance") }
v := 1 / varianceInverse
delta := v * deltaSum
sigma, err := solveVolatility(phi, v, delta, working.Volatility)
if err != nil { return Rating{}, err }
phiStar := math.Sqrt(phi*phi + sigma*sigma)
phiPrime := 1 / math.Sqrt(1/(phiStar*phiStar)+1/v)
muPrime := mu + phiPrime*phiPrime*deltaSum
return Rating{Value: fromScaleRating(muPrime), RD: fromScaleRD(phiPrime), Volatility: sigma, LastRatedAt: now}, nil
}
func validateRating(r Rating) error {
if r.Value < 0 || r.RD <= 0 || r.RD > GlickoInitialRD || r.Volatility <= 0 || r.Volatility >= 1 { return fmt.Errorf("invalid rating state") }
return nil
}
func advanceInactivity(r Rating, now time.Time) Rating {
if r.LastRatedAt.IsZero() || !now.After(r.LastRatedAt) { return r }
periods := int(now.Sub(r.LastRatedAt) / (24 * time.Hour))
if periods <= 0 { return r }
phi := r.RD / GlickoScale
phi = math.Min(GlickoInitialRD/GlickoScale, math.Sqrt(phi*phi+float64(periods)*r.Volatility*r.Volatility))
r.RD = fromScaleRD(phi)
return r
}
func toScale(rating, rd float64) (float64, float64) { return (rating - GlickoInitialRating) / GlickoScale, rd / GlickoScale }
func fromScaleRating(mu float64) float64 { return mu*GlickoScale + GlickoInitialRating }
func fromScaleRD(phi float64) float64 { return phi * GlickoScale }
func glickoG(phi float64) float64 { return 1 / math.Sqrt(1+3*phi*phi/(math.Pi*math.Pi)) }
func expectedScore(mu, opponentMu, g float64) float64 { return 1 / (1 + math.Exp(-g*(mu-opponentMu))) }
func solveVolatility(phi, v, delta, volatility float64) (float64, error) {
a := math.Log(volatility * volatility)
variance := delta*delta - phi*phi - v
var b float64
if variance > 0 { b = math.Log(variance) } else {
b = a - GlickoTau
for volatilityFunction(b, a, phi, v, delta) < 0 {
b -= GlickoTau
if b < -100 { return 0, fmt.Errorf("volatility bracket not found") }
}
}
fa := volatilityFunction(a, a, phi, v, delta)
fb := volatilityFunction(b, a, phi, v, delta)
for math.Abs(b-a) > GlickoEpsilon {
c := a + (a-b)*fa/(fb-fa)
fc := volatilityFunction(c, a, phi, v, delta)
if fc*fb < 0 { a, fa = b, fb } else { fa /= 2 }
b, fb = c, fc
if math.IsNaN(b) || math.IsInf(b, 0) { return 0, fmt.Errorf("volatility iteration diverged") }
}
return math.Exp(a / 2), nil
}
func volatilityFunction(x, a, phi, v, delta float64) float64 {
expX := math.Exp(x)
denominator := 2 * math.Pow(phi*phi+v+expX, 2)
return expX*(delta*delta-phi*phi-v-expX)/denominator - (x-a)/(GlickoTau*GlickoTau)
}
// RankedOpponents assigns the exact 1/3 contribution to each of three human
// opponents. CasualOpponents assigns 1/N; both return lexical order so a
// database row-order change cannot affect floating-point accumulation order.
func RankedOpponents(opponents []Opponent) ([]Opponent, error) {
if len(opponents) != 3 { return nil, fmt.Errorf("ranked 3v3 requires three opponents") }
return weightedOpponents(opponents, 1.0/3.0), nil
}
func CasualOpponents(opponents []Opponent) ([]Opponent, error) {
if len(opponents) == 0 { return nil, nil }
return weightedOpponents(opponents, 1/float64(len(opponents))), nil
}
func weightedOpponents(opponents []Opponent, weight float64) []Opponent {
result := append([]Opponent(nil), opponents...)
sort.Slice(result, func(i, j int) bool {
if result[i].Rating.Value != result[j].Rating.Value { return result[i].Rating.Value < result[j].Rating.Value }
return result[i].PlayerID < result[j].PlayerID
})
for i := range result { result[i].Weight = weight }
return result
}
+49
View File
@@ -0,0 +1,49 @@
package domain
import (
"math"
"testing"
"time"
)
func TestUpdateRatingMatchesCanonicalGlicko2Example(t *testing.T) {
current := Rating{Value: 1500, RD: 200, Volatility: 0.06}
opponents := []Opponent{
{PlayerID: "a", Rating: Rating{Value: 1400, RD: 30, Volatility: 0.06}, Score: 1, Weight: 1},
{PlayerID: "b", Rating: Rating{Value: 1550, RD: 100, Volatility: 0.06}, Score: 0, Weight: 1},
{PlayerID: "c", Rating: Rating{Value: 1700, RD: 300, Volatility: 0.06}, Score: 0, Weight: 1},
}
updated, err := UpdateRating(current, opponents, time.Unix(100000, 0))
if err != nil { t.Fatal(err) }
if math.Abs(updated.Value-1464.06) > 0.1 || math.Abs(updated.RD-151.52) > 0.1 || math.Abs(updated.Volatility-0.05999) > 0.0001 {
t.Fatalf("canonical vector mismatch: %+v", updated)
}
}
func TestRatingInactivityRaisesRDWithoutChangingRating(t *testing.T) {
now := time.Unix(100000, 0)
current := Rating{Value: 1600, RD: 100, Volatility: 0.06, LastRatedAt: now}
updated, err := UpdateRating(current, nil, now.Add(48*time.Hour+time.Hour))
if err != nil { t.Fatal(err) }
if updated.Value != current.Value || updated.RD <= current.RD || updated.RD > GlickoInitialRD { t.Fatalf("bad inactivity update: %+v", updated) }
}
func TestOpponentWeightHelpersAreExactAndDeterministic(t *testing.T) {
opponents := []Opponent{{PlayerID: "c", Rating: Rating{Value: 1700}}, {PlayerID: "a", Rating: Rating{Value: 1400}}, {PlayerID: "b", Rating: Rating{Value: 1550}}}
ranked, err := RankedOpponents(opponents)
if err != nil { t.Fatal(err) }
if ranked[0].PlayerID != "a" || ranked[0].Weight != 1.0/3.0 { t.Fatalf("ranked weighting/order wrong: %+v", ranked) }
reordered, err := RankedOpponents([]Opponent{opponents[1], opponents[0], opponents[2]})
if err != nil { t.Fatal(err) }
for i := range ranked { if ranked[i].PlayerID != reordered[i].PlayerID { t.Fatal("input order changed opponent order") } }
casual, err := CasualOpponents(opponents[:2])
if err != nil { t.Fatal(err) }
if casual[0].Weight != 0.5 || casual[1].Weight != 0.5 { t.Fatalf("casual weighting wrong: %+v", casual) }
}
func TestRatingRejectsInvalidStateAndBadScore(t *testing.T) {
_, err := UpdateRating(Rating{Value: 1500, RD: 0, Volatility: 0.06}, nil, time.Now())
if err == nil { t.Fatal("accepted zero RD") }
_, err = UpdateRating(Rating{Value: 1500, RD: 200, Volatility: 0.06}, []Opponent{{Rating: Rating{Value: 1500, RD: 100, Volatility: 0.06}, Weight: 1, Score: 2}}, time.Now())
if err == nil { t.Fatal("accepted score outside [0,1]") }
}