brackt/app/services/simulations/wnba-simulator.ts
chrisp 3e50619629
All checks were successful
🚀 Deploy / 🧪 Test (push) Successful in 3m1s
🚀 Deploy / ʦ🔍 Typecheck & Lint (push) Successful in 1m17s
🚀 Deploy / 🐳 Build (push) Successful in 1m9s
🚀 Deploy / 🚀 Deploy (push) Successful in 11s
claude/probability-config-review-1tdolw (#119)
Co-authored-by: Claude <noreply@anthropic.com>
Reviewed-on: #119
2026-06-30 23:24:48 +00:00

334 lines
14 KiB
TypeScript
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

/**
* WNBA Playoff Simulator
*
* Monte Carlo simulation of the WNBA playoffs including regular season
* remainder projection and seeding.
*
* Algorithm:
* 1. Load participants, regular season standings, and futures odds in parallel.
* 2. Choose Elo source automatically based on season progress:
* Pre-season (avg gamesPlayed < SRS_GAMES_THRESHOLD):
* → futures odds via convertFuturesToElo() from participantExpectedValues
* → falls back to 1500 for teams without odds
* In-season (avg gamesPlayed >= SRS_GAMES_THRESHOLD):
* → SRS from regularSeasonStandings: elo = 1500 + srs * SRS_ELO_SCALE
* → falls back to futures Elo for teams without SRS, then 1500
* 3. For each simulation:
* a. For each team, simulate remaining regular season games (40 - gamesPlayed)
* using Elo win probability vs. an average opponent (Elo 1500)
* → projectedWins = currentWins + simulatedRemainingWins
* b. Sort all teams by projected wins (desc) + random tiebreaker
* → Top 8 qualify for playoffs
* c. Simulate WNBA playoff bracket (no byes, no play-in):
* Round 1 (best-of-3): 1v8, 4v5, 2v7, 3v6
* Semifinals (best-of-5): winner(1/8) vs winner(4/5), winner(2/7) vs winner(3/6)
* Finals (best-of-7): Semifinal winners
* 4. Track placement counts per scoring tier
* 5. Convert counts to probability distributions
*
* Win probability (Elo, PARITY_FACTOR = 400):
* P(A beats B) = 1 / (1 + 10^((eloB - eloA) / 400))
*
* SRS → Elo conversion (in-season):
* elo = 1500 + srs * 20
* Calibration: WNBA SRS typically ranges ±810. A ±10 SRS gap → ±200 Elo → ~76% win
* probability for the stronger team, which is appropriate for single-game WNBA matchups.
* Source: basketball-reference.com/wnba/years/YYYY.html (SRS column in team standings)
*
* Futures → Elo conversion (pre-season):
* Uses convertFuturesToElo() from probability-engine (same as BracketSimulator / UCLSimulator).
* Reads sourceOdds (American format) from participantExpectedValues.
*
* Placement tiers → SimulationProbabilities mapping:
* probFirst = WNBA champion (1 per sim)
* probSecond = Finals loser (1 per sim)
* probThird/Fourth = Semifinal losers (2 per sim)
* probFifthEighth = Round 1 losers (4 per sim)
* Missed playoffs → all 0
*/
import { database } from "~/database/context";
import { eq } from "drizzle-orm";
import * as schema from "~/database/schema";
import type { Simulator, SimulationResult } from "./types";
import { normalizeTeamName } from "~/lib/normalize-team-name";
import { getRegularSeasonStandings } from "~/models/regular-season-standings";
import { convertFuturesToElo, eloWinProbabilityWithParity } from "~/services/probability-engine";
import { positiveConfigNumber } from "./config-access";
// ─── Simulation parameters (defaults; overridable via season config) ───────────
const DEFAULT_NUM_SIMULATIONS = 50_000;
/** Elo parity factor. WNBA defaults to 400 (standard formula). */
const DEFAULT_PARITY_FACTOR = 400;
/** WNBA regular season games per team. */
const DEFAULT_REGULAR_SEASON_GAMES = 40;
/**
* Multiplier converting SRS to Elo offset from 1500.
* elo = 1500 + srs * SRS_ELO_SCALE
*
* At scale 20:
* SRS +10 → Elo 1700, SRS -10 → Elo 1300 → P(+10 vs -10) ≈ 76%
* SRS +5 → Elo 1600, SRS 0 → Elo 1500 → P(+5 vs 0) ≈ 64%
*/
const SRS_ELO_SCALE = 20;
/**
* Average games played threshold to switch from futures-derived Elo to SRS-derived Elo.
* Once most teams have played this many games, SRS is meaningful enough to use.
*/
const SRS_GAMES_THRESHOLD = 5;
/** Number of playoff teams. */
const PLAYOFF_TEAMS = 8;
// ─── Public helpers (exported for unit testing) ───────────────────────────────
export { normalizeTeamName };
/**
* Convert an SRS rating to an Elo rating.
* An SRS of 0 maps to 1500 (league average).
*/
export function srsToElo(srs: number): number {
return 1500 + srs * SRS_ELO_SCALE;
}
/**
* Resolve the Elo to use for a team given available signals and the current mode.
*
* Priority: sourceElo (admin-entered projected wins) > SRS (in-season) > futures odds > 1500.
*/
export function resolveElo(
srs: number | null,
futuresElo: number | null,
useSRS: boolean,
sourceElo?: number | null
): number {
if (sourceElo !== null && sourceElo !== undefined) return sourceElo;
if (useSRS) {
if (srs !== null) return srsToElo(srs);
return futuresElo ?? 1500;
}
return futuresElo ?? 1500;
}
// ─── Internal types ───────────────────────────────────────────────────────────
interface TeamEntry {
id: string;
name: string;
elo: number;
currentWins: number;
remainingGames: number;
/** Elo win probability vs. average opponent (1500) — constant per team. */
winProb: number;
}
/** Simulate a best-of-N series.
* @param winTarget wins needed to win the series (2 for BoX3, 3 for BoX5, 4 for BoX7) */
export function simSeriesN(
a: TeamEntry,
b: TeamEntry,
winTarget: number,
parityFactor = DEFAULT_PARITY_FACTOR
): { winner: TeamEntry; loser: TeamEntry } {
const winProb = eloWinProbabilityWithParity(a.elo, b.elo, parityFactor);
let winsA = 0;
let winsB = 0;
while (winsA < winTarget && winsB < winTarget) {
if (Math.random() < winProb) winsA++;
else winsB++;
}
return winsA === winTarget ? { winner: a, loser: b } : { winner: b, loser: a };
}
/** Simulate remaining regular season games for a team. Returns projected total wins. */
function simulateProjectedWins(team: TeamEntry): number {
let extra = 0;
for (let g = 0; g < team.remainingGames; g++) {
if (Math.random() < team.winProb) extra++;
}
return team.currentWins + extra;
}
// ─── Simulator ────────────────────────────────────────────────────────────────
export class WNBASimulator implements Simulator {
async simulate(sportsSeasonId: string, config: Record<string, unknown> = {}): Promise<SimulationResult[]> {
const db = database();
const parityFactor = positiveConfigNumber(config, "parityFactor", DEFAULT_PARITY_FACTOR);
const numSimulations = Math.round(positiveConfigNumber(config, "iterations", DEFAULT_NUM_SIMULATIONS));
const seasonGames = Math.round(positiveConfigNumber(config, "seasonGames", DEFAULT_REGULAR_SEASON_GAMES));
// 1. Load participants, standings, and futures odds in parallel.
const [participantRows, standings, evRows] = await Promise.all([
db
.select({ id: schema.seasonParticipants.id, name: schema.seasonParticipants.name })
.from(schema.seasonParticipants)
.where(eq(schema.seasonParticipants.sportsSeasonId, sportsSeasonId)),
getRegularSeasonStandings(sportsSeasonId),
db
.select({
participantId: schema.seasonParticipantExpectedValues.participantId,
sourceOdds: schema.seasonParticipantExpectedValues.sourceOdds,
sourceElo: schema.seasonParticipantExpectedValues.sourceElo,
})
.from(schema.seasonParticipantExpectedValues)
.where(eq(schema.seasonParticipantExpectedValues.sportsSeasonId, sportsSeasonId)),
]);
if (participantRows.length === 0) {
throw new Error(
`No participants found for sports season ${sportsSeasonId}. ` +
`Add WNBA teams as participants before running simulation.`
);
}
if (participantRows.length < PLAYOFF_TEAMS) {
throw new Error(
`WNBA simulation requires at least ${PLAYOFF_TEAMS} participants ` +
`(got ${participantRows.length}). Add all WNBA teams before running simulation.`
);
}
// 2. Build futures Elo map from championship odds and sourceElo map from projected wins.
const oddsInput = evRows
.filter((r) => r.sourceOdds !== null)
.map((r) => ({ participantId: r.participantId, odds: r.sourceOdds ?? 0 }));
const futuresEloMap: Map<string, number> =
oddsInput.length > 0
? convertFuturesToElo(oddsInput, "american")
: new Map();
const sourceEloMap = new Map<string, number>();
for (const row of evRows) {
if (row.sourceElo !== null && row.sourceElo !== undefined) {
sourceEloMap.set(row.participantId, row.sourceElo);
}
}
// 3. Build standings lookup and determine simulation mode.
const standingsMap = new Map(standings.map((s) => [s.participantId, s]));
const participantIds = participantRows.map((r) => r.id);
const totalGamesPlayed = participantRows.reduce((sum, r) => {
return sum + (standingsMap.get(r.id)?.gamesPlayed ?? 0);
}, 0);
const avgGamesPlayed = totalGamesPlayed / participantRows.length;
const useSRS = avgGamesPlayed >= SRS_GAMES_THRESHOLD;
// 4. Construct team entries with resolved Elo.
const teams: TeamEntry[] = participantRows.map((r) => {
const standing = standingsMap.get(r.id);
const srs =
standing?.srs !== null && standing?.srs !== undefined
? parseFloat(standing.srs)
: null;
const futuresElo = futuresEloMap.get(r.id) ?? null;
const elo = resolveElo(srs, futuresElo, useSRS, sourceEloMap.get(r.id) ?? null);
const gamesPlayed = standing?.gamesPlayed ?? 0;
return {
id: r.id,
name: r.name,
elo,
currentWins: standing?.wins ?? 0,
remainingGames: Math.max(0, seasonGames - gamesPlayed),
winProb: eloWinProbabilityWithParity(elo, 1500, parityFactor),
};
});
const source = useSRS
? "wnba_bracket_monte_carlo_srs"
: "wnba_bracket_monte_carlo_futures";
/** Seed teams 18 by projected wins for this simulation. */
const buildSeededBracket = (): TeamEntry[] => {
const projected = teams.map((t) => ({
team: t,
projectedWins: simulateProjectedWins(t),
tiebreaker: Math.random(),
}));
projected.sort((a, b) => b.projectedWins - a.projectedWins || b.tiebreaker - a.tiebreaker);
return projected.slice(0, PLAYOFF_TEAMS).map((x) => x.team);
};
// 5. Integer placement count maps — initialized to 0 for all participants.
const championCounts = new Map<string, number>(participantIds.map((id) => [id, 0]));
const finalistCounts = new Map<string, number>(participantIds.map((id) => [id, 0]));
const semiLoserCounts = new Map<string, number>(participantIds.map((id) => [id, 0]));
const r1LoserCounts = new Map<string, number>(participantIds.map((id) => [id, 0]));
// 6. Monte Carlo simulation loop.
for (let s = 0; s < numSimulations; s++) {
const [s1, s2, s3, s4, s5, s6, s7, s8] = buildSeededBracket();
// Round 1 (best-of-3): 1v8, 4v5, 2v7, 3v6
const r1_a = simSeriesN(s1, s8, 2, parityFactor);
const r1_b = simSeriesN(s4, s5, 2, parityFactor);
const r1_c = simSeriesN(s2, s7, 2, parityFactor);
const r1_d = simSeriesN(s3, s6, 2, parityFactor);
r1LoserCounts.set(r1_a.loser.id, (r1LoserCounts.get(r1_a.loser.id) ?? 0) + 1);
r1LoserCounts.set(r1_b.loser.id, (r1LoserCounts.get(r1_b.loser.id) ?? 0) + 1);
r1LoserCounts.set(r1_c.loser.id, (r1LoserCounts.get(r1_c.loser.id) ?? 0) + 1);
r1LoserCounts.set(r1_d.loser.id, (r1LoserCounts.get(r1_d.loser.id) ?? 0) + 1);
// Semifinals (best-of-5): winner(1/8) vs winner(4/5), winner(2/7) vs winner(3/6)
const sf_a = simSeriesN(r1_a.winner, r1_b.winner, 3, parityFactor);
const sf_b = simSeriesN(r1_c.winner, r1_d.winner, 3, parityFactor);
semiLoserCounts.set(sf_a.loser.id, (semiLoserCounts.get(sf_a.loser.id) ?? 0) + 1);
semiLoserCounts.set(sf_b.loser.id, (semiLoserCounts.get(sf_b.loser.id) ?? 0) + 1);
// Finals (best-of-7)
const final = simSeriesN(sf_a.winner, sf_b.winner, 4, parityFactor);
championCounts.set(final.winner.id, (championCounts.get(final.winner.id) ?? 0) + 1);
finalistCounts.set(final.loser.id, (finalistCounts.get(final.loser.id) ?? 0) + 1);
}
// 7. Convert integer counts to probability distributions.
const results: SimulationResult[] = participantIds.map((participantId) => {
const c = championCounts.get(participantId) ?? 0;
const f = finalistCounts.get(participantId) ?? 0;
const sl = semiLoserCounts.get(participantId) ?? 0;
const r1 = r1LoserCounts.get(participantId) ?? 0;
return {
participantId,
probabilities: {
probFirst: c / numSimulations,
probSecond: f / numSimulations,
probThird: sl / (2 * numSimulations),
probFourth: sl / (2 * numSimulations),
probFifth: r1 / (4 * numSimulations),
probSixth: r1 / (4 * numSimulations),
probSeventh: r1 / (4 * numSimulations),
probEighth: r1 / (4 * numSimulations),
},
source,
};
});
// 8. Per-position normalization — belt-and-suspenders guard against floating-point residuals.
const positionKeys: Array<keyof (typeof results)[0]["probabilities"]> = [
"probFirst", "probSecond", "probThird", "probFourth",
"probFifth", "probSixth", "probSeventh", "probEighth",
];
for (const key of positionKeys) {
const colSum = results.reduce((s, r) => s + r.probabilities[key], 0);
const residual = 1.0 - colSum;
if (residual !== 0) {
const maxResult = results.reduce((best, r) =>
r.probabilities[key] > best.probabilities[key] ? r : best
);
maxResult.probabilities[key] += residual;
}
}
return results;
}
}