diff --git a/.github/workflows/archive.yml b/.github/workflows/archive.yml
new file mode 100644
index 0000000..a4334d6
--- /dev/null
+++ b/.github/workflows/archive.yml
@@ -0,0 +1,53 @@
+name: Archive MAX seats
+
+# The SNCF feed only shows today's seat state. This keeps one snapshot a day on the public
+# `data` branch (archive/max/YYYY/YYYY-MM-DD.json.gz), so seat odds can be computed later.
+# A day already archived is never overwritten.
+
+on:
+  # After the feed's midday refresh and the 13:30 publish.
+  schedule:
+    - cron: "0 14 * * *"
+  workflow_dispatch:
+
+permissions:
+  contents: write
+
+concurrency:
+  group: archive
+  cancel-in-progress: false
+
+jobs:
+  archive:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 22
+          cache: npm
+      - run: npm ci
+      - run: npm run fetch
+      - name: Check out the data branch
+        run: |
+          if git ls-remote --exit-code --heads origin data >/dev/null; then
+            git fetch --depth=1 origin data
+            git worktree add data-branch FETCH_HEAD
+          else
+            git worktree add --detach data-branch
+            git -C data-branch checkout --orphan data
+            git -C data-branch rm -rfq --ignore-unmatch .
+          fi
+      - run: npx tsx scripts/archive-max.ts data/raw/sncf-tgvmax.json data-branch/archive/max
+      - name: Commit and push
+        working-directory: data-branch
+        run: |
+          git config user.name "github-actions[bot]"
+          git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
+          git add archive
+          if git diff --cached --quiet; then
+            echo "Nothing new to archive."
+            exit 0
+          fi
+          git commit -q -m "Archive MAX seats for $(date -u +%F)"
+          git push origin HEAD:data
diff --git a/docs/archive.md b/docs/archive.md
new file mode 100644
index 0000000..e5c507c
--- /dev/null
+++ b/docs/archive.md
@@ -0,0 +1,24 @@
+# MAX seat archive
+
+The SNCF feed only shows today's MAX seats. The **Archive MAX seats** workflow saves one
+snapshot a day on the public `data` branch, so seat odds (by weekday, departure time and
+days before travel) can be computed from the history.
+
+- Path: `archive/max/YYYY/YYYY-MM-DD.json.gz`, named by the day the feed was read (UTC).
+  A day already archived is never overwritten.
+- Content (gzipped JSON):
+
+```json
+{
+  "snapshot": "2026-10-05",
+  "start": "2026-10-05",
+  "days": 31,
+  "services": [["PARIS (intramuros)", "LYON (intramuros)", "6601", "08:00", "11.0..."]]
+}
+```
+
+Each service (origin, destination, train number, departure) is stored once. Its last field
+has one character per day from `start`: `1` a free MAX seat, `0` no MAX seat, `.` not
+running that day.
+
+Data: SNCF Open Data, Licence Ouverte.
diff --git a/scripts/archive-max.ts b/scripts/archive-max.ts
new file mode 100644
index 0000000..f8d78fa
--- /dev/null
+++ b/scripts/archive-max.ts
@@ -0,0 +1,99 @@
+/**
+ * Archive today's MAX seat flags from the SNCF feed, so seat odds (by weekday, departure
+ * time and days before travel) can be computed later. The feed only shows the current
+ * state; anything not archived on the day is lost.
+ *
+ * One file per day, never overwritten: <out>/YYYY/YYYY-MM-DD.json.gz. Each service
+ * (origin, destination, train, departure) is stored once with one character per day of
+ * the feed's window: "1" a free MAX seat, "0" full, "." not running that day.
+ *
+ * Run: npx tsx scripts/archive-max.ts [feed.json] [outDir]
+ */
+import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
+import { dirname, join, resolve } from "node:path";
+import { pathToFileURL } from "node:url";
+import { gzipSync } from "node:zlib";
+import type { SncfRecord } from "../src/operators/sncf";
+
+export interface MaxArchive {
+  /** Day the feed was read (UTC), "YYYY-MM-DD". */
+  snapshot: string;
+  /** First travel day of the window; character i of each service is start + i days. */
+  start: string;
+  days: number;
+  /** [origin, destination, trainNo, depart "HH:MM", per-day flags]. */
+  services: [string, string, string, string, string][];
+}
+
+const DAY_MS = 86_400_000;
+
+function str(v: unknown): string {
+  return typeof v === "string" ? v.trim() : v == null ? "" : String(v).trim();
+}
+
+function hhmm(v: unknown): string {
+  const m = /^(\d{1,2}):(\d{2})/.exec(str(v));
+  return m ? `${m[1]!.padStart(2, "0")}:${m[2]}` : "";
+}
+
+function dayIndex(date: string): number {
+  return Math.round(Date.parse(`${date}T00:00:00Z`) / DAY_MS);
+}
+
+export function buildArchive(records: SncfRecord[], snapshot: string): MaxArchive {
+  const rows: { key: string; day: number; free: boolean }[] = [];
+  for (const r of records) {
+    const date = str(r.date);
+    const depart = hhmm(r.heure_depart);
+    const origin = str(r.origine);
+    const destination = str(r.destination);
+    if (!/^\d{4}-\d{2}-\d{2}$/.test(date) || !depart || !origin || !destination) continue;
+    const key = [origin, destination, str(r.train_no), depart].join("\u0000");
+    rows.push({ key, day: dayIndex(date), free: str(r.od_happy_card).toUpperCase() === "OUI" });
+  }
+  if (rows.length === 0) return { snapshot, start: snapshot, days: 0, services: [] };
+
+  // A loop, not Math.min(...rows): the feed has ~360k rows, too many for spread arguments.
+  let first = Infinity;
+  let last = -Infinity;
+  for (const r of rows) {
+    first = Math.min(first, r.day);
+    last = Math.max(last, r.day);
+  }
+  const days = last - first + 1;
+  const flags = new Map<string, string[]>();
+  for (const r of rows) {
+    let f = flags.get(r.key);
+    if (!f) flags.set(r.key, (f = Array<string>(days).fill(".")));
+    // A free seat on any duplicate row wins: the feed never lists a seat it doesn't have.
+    if (f[r.day - first] !== "1") f[r.day - first] = r.free ? "1" : "0";
+  }
+
+  const services = [...flags]
+    .map(([key, f]) => [...(key.split("\u0000") as [string, string, string, string]), f.join("")] as MaxArchive["services"][number])
+    .sort((a, b) => a.join("|").localeCompare(b.join("|")));
+  return { snapshot, start: new Date(first * DAY_MS).toISOString().slice(0, 10), days, services };
+}
+
+/** Write the day's archive; returns the path, or null when that day is already archived. */
+export function writeArchive(archive: MaxArchive, outDir: string): string | null {
+  const path = join(outDir, archive.snapshot.slice(0, 4), `${archive.snapshot}.json.gz`);
+  if (existsSync(path)) return null;
+  mkdirSync(dirname(path), { recursive: true });
+  writeFileSync(path, gzipSync(JSON.stringify(archive)));
+  return path;
+}
+
+if (process.argv[1] && import.meta.url === pathToFileURL(resolve(process.argv[1])).href) {
+  const feed = resolve(process.argv[2] ?? "data/raw/sncf-tgvmax.json");
+  const outDir = resolve(process.argv[3] ?? "data/archive");
+  const records = JSON.parse(readFileSync(feed, "utf-8")) as SncfRecord[];
+  if (!Array.isArray(records) || records.length === 0) throw new Error(`no records in ${feed}`);
+  const archive = buildArchive(records, new Date().toISOString().slice(0, 10));
+  const path = writeArchive(archive, outDir);
+  console.log(
+    path
+      ? `[archive] ${archive.services.length} services over ${archive.days} days → ${path}`
+      : `[archive] ${archive.snapshot} is already archived, left as is`,
+  );
+}
diff --git a/tests/archive.test.ts b/tests/archive.test.ts
new file mode 100644
index 0000000..1174f47
--- /dev/null
+++ b/tests/archive.test.ts
@@ -0,0 +1,47 @@
+import { mkdtempSync, readFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { gunzipSync } from "node:zlib";
+import { describe, expect, it } from "vitest";
+import { buildArchive, writeArchive } from "../scripts/archive-max";
+import fixture from "./fixtures/sncf-tgvmax.sample.json";
+
+const row = (date: string, train: string, seat: string) => ({
+  date,
+  origine: "PARIS (intramuros)",
+  destination: "LYON (intramuros)",
+  heure_depart: "8:00",
+  heure_arrivee: "10:00",
+  train_no: train,
+  od_happy_card: seat,
+});
+
+describe("MAX seat archive", () => {
+  it("stores each service once with a flag per day", () => {
+    const a = buildArchive(
+      [row("2026-10-05", "6601", "OUI"), row("2026-10-07", "6601", "NON"), row("2026-10-06", "6605", "OUI")],
+      "2026-10-05",
+    );
+    expect(a.start).toBe("2026-10-05");
+    expect(a.days).toBe(3);
+    expect(a.services).toEqual([
+      ["PARIS (intramuros)", "LYON (intramuros)", "6601", "08:00", "1.0"],
+      ["PARIS (intramuros)", "LYON (intramuros)", "6605", "08:00", ".1."],
+    ]);
+  });
+
+  it("keeps a free seat when the feed lists a service twice on one day", () => {
+    const a = buildArchive([row("2026-10-05", "6601", "NON"), row("2026-10-05", "6601", "OUI")], "2026-10-05");
+    expect(a.services[0]?.[4]).toBe("1");
+  });
+
+  it("archives the whole fixture and never overwrites a day", () => {
+    const a = buildArchive(fixture, "2026-06-25");
+    expect(a.services.length).toBeGreaterThan(0);
+    const dir = mkdtempSync(join(tmpdir(), "archive-"));
+    const path = writeArchive(a, dir);
+    expect(path).toBe(join(dir, "2026", "2026-06-25.json.gz"));
+    expect(JSON.parse(gunzipSync(readFileSync(path!)).toString())).toEqual(a);
+    expect(writeArchive(a, dir)).toBeNull();
+  });
+});