diff --git a/Dockerfile b/Dockerfile
index a4b4fce..26bb3e8 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -4,7 +4,10 @@ WORKDIR /app
 COPY package.json bun.lock* bun.lockb* ./
 RUN bun install
 COPY . .
-RUN bun run data:build      # icons + fixtures.json + ratings.json → public/
+# Quick backtest in image builds: score the shipped config only — the published
+# numbers are identical; the hyperparam grid search is a local-dev activity.
+ENV BACKTEST_QUICK=1
+RUN bun run data:build      # icons + fixtures.json + ratings.json + backtest → public/
 RUN bun run build           # tsc -b && vite build → /app/dist
 RUN bun run build:server    # esbuild → /app/server/dist/index.js
 
diff --git a/scripts/buildBacktest.ts b/scripts/buildBacktest.ts
index 3813459..ef882cf 100644
--- a/scripts/buildBacktest.ts
+++ b/scripts/buildBacktest.ts
@@ -40,6 +40,12 @@ const FORM_WINDOW = 10; // last N non-friendly matches feeding the form signal
 const MIN_VAL_GAIN = 0.0005; // candidate must beat the previous config by this much
 /** the previously shipped v3 config — the bar every candidate has to clear */
 const PREV = { xi: 0.5, k: 5, w: 0.75, m: 2, beta: 0, gamma: 0 };
+/** the v4 winner (from the full grid run) — keep in sync with buildRatings.ts */
+const SHIPPED = { xi: 0.25, k: 3, w: 0.45, m: 3, beta: 0.3, gamma: 0 };
+/** BACKTEST_QUICK=1 (Docker builds): skip the grid search and just score the
+ *  shipped + previous configs — the published test numbers are identical, the
+ *  hyperparam hunt only happens locally. */
+const QUICK = process.env.BACKTEST_QUICK === '1';
 
 interface Row { date: string; home: string; away: string; hs: number; as: number; tournament: string; neutral: boolean; t: number }
 type Probs = { h: number; d: number; a: number };
@@ -241,42 +247,55 @@ function main(): void {
     return W_STEPS.map((w, wi) => ({ w, rps: acc[wi]! / preds.length }));
   };
 
-  // ---------- validation stage 1: tune ξ, k, w (β=0, γ=0) ----------
-  console.log('validation stage 1 — ξ/k/w grid (β=0, γ=0)…');
-  const valWalks = new Map<string, Pred[]>();
   let best = { xi: PREV.xi, k: PREV.k, w: PREV.w, m: PREV.m, beta: 0, gamma: 0, rps: Infinity };
   let prevValRps = Infinity; // validation RPS of the previously shipped config
-  for (const xi of XI_GRID) {
-    for (const k of K_GRID) {
-      const preds = walkTeamDc(xi, k, tValFrom, tTestFrom);
-      valWalks.set(`${xi}|${k}`, preds);
-      for (const { w, rps: r } of sweepW(preds, 0, PREV.m, 0)) {
-        if (xi === PREV.xi && k === PREV.k && w === PREV.w) prevValRps = r;
-        if (r < best.rps) best = { xi, k, w, m: PREV.m, beta: 0, gamma: 0, rps: r };
-      }
-      console.log(`  ξ=${xi} k=${k} done (best so far: ξ=${best.xi} k=${best.k} w=${best.w} rps=${best.rps.toFixed(4)})`);
-    }
-  }
 
-  // ---------- validation stage 2: fast-Elo blend (m, β) at the stage-1 ξ/k ----------
-  console.log('validation stage 2 — fast-Elo blend (m, β)…');
-  const stagePreds = valWalks.get(`${best.xi}|${best.k}`)!;
-  for (const m of M_GRID) {
-    for (const beta of BETA_GRID) {
-      if (beta === 0) continue; // control already covered by stage 1
-      for (const { w, rps: r } of sweepW(stagePreds, beta, m, 0)) {
-        if (r < best.rps) best = { ...best, m, beta, w, rps: r };
+  if (QUICK) {
+    // ---------- quick mode: score only shipped + previous on validation ----------
+    console.log('quick mode — scoring shipped + previous configs on validation…');
+    const predsShipped = walkTeamDc(SHIPPED.xi, SHIPPED.k, tValFrom, tTestFrom);
+    const predsPrev = SHIPPED.xi === PREV.xi && SHIPPED.k === PREV.k
+      ? predsShipped
+      : walkTeamDc(PREV.xi, PREV.k, tValFrom, tTestFrom);
+    const shippedRps = sweepW(predsShipped, SHIPPED.beta, SHIPPED.m, SHIPPED.gamma).find((x) => x.w === SHIPPED.w)!.rps;
+    prevValRps = sweepW(predsPrev, PREV.beta, PREV.m, PREV.gamma).find((x) => x.w === PREV.w)!.rps;
+    best = { ...SHIPPED, rps: shippedRps };
+  } else {
+    // ---------- validation stage 1: tune ξ, k, w (β=0, γ=0) ----------
+    console.log('validation stage 1 — ξ/k/w grid (β=0, γ=0)…');
+    const valWalks = new Map<string, Pred[]>();
+    for (const xi of XI_GRID) {
+      for (const k of K_GRID) {
+        const preds = walkTeamDc(xi, k, tValFrom, tTestFrom);
+        valWalks.set(`${xi}|${k}`, preds);
+        for (const { w, rps: r } of sweepW(preds, 0, PREV.m, 0)) {
+          if (xi === PREV.xi && k === PREV.k && w === PREV.w) prevValRps = r;
+          if (r < best.rps) best = { xi, k, w, m: PREV.m, beta: 0, gamma: 0, rps: r };
+        }
+        console.log(`  ξ=${xi} k=${k} done (best so far: ξ=${best.xi} k=${best.k} w=${best.w} rps=${best.rps.toFixed(4)})`);
       }
     }
-    console.log(`  m=${m} swept (best: β=${best.beta} m=${best.m} w=${best.w} rps=${best.rps.toFixed(4)})`);
-  }
 
-  // ---------- validation stage 3: recent-form multiplier γ at the winners ----------
-  console.log('validation stage 3 — form multiplier γ…');
-  for (const gamma of GAMMA_GRID) {
-    if (gamma === 0) continue; // control covered above
-    for (const { w, rps: r } of sweepW(stagePreds, best.beta, best.m, gamma)) {
-      if (r < best.rps) best = { ...best, gamma, w, rps: r };
+    // ---------- validation stage 2: fast-Elo blend (m, β) at the stage-1 ξ/k ----------
+    console.log('validation stage 2 — fast-Elo blend (m, β)…');
+    const stagePreds = valWalks.get(`${best.xi}|${best.k}`)!;
+    for (const m of M_GRID) {
+      for (const beta of BETA_GRID) {
+        if (beta === 0) continue; // control already covered by stage 1
+        for (const { w, rps: r } of sweepW(stagePreds, beta, m, 0)) {
+          if (r < best.rps) best = { ...best, m, beta, w, rps: r };
+        }
+      }
+      console.log(`  m=${m} swept (best: β=${best.beta} m=${best.m} w=${best.w} rps=${best.rps.toFixed(4)})`);
+    }
+
+    // ---------- validation stage 3: recent-form multiplier γ at the winners ----------
+    console.log('validation stage 3 — form multiplier γ…');
+    for (const gamma of GAMMA_GRID) {
+      if (gamma === 0) continue; // control covered above
+      for (const { w, rps: r } of sweepW(stagePreds, best.beta, best.m, gamma)) {
+        if (r < best.rps) best = { ...best, gamma, w, rps: r };
+      }
     }
   }
   console.log(`  champion: ξ=${best.xi} k=${best.k} w=${best.w} β=${best.beta} m=${best.m} γ=${best.gamma} rps=${best.rps.toFixed(4)}`);