diff --git a/apps/backend/src/repositories/worker-platform/work-order-query-repo.ts b/apps/backend/src/repositories/worker-platform/work-order-query-repo.ts
index 3cf59912..2e5be9d3 100644
--- a/apps/backend/src/repositories/worker-platform/work-order-query-repo.ts
+++ b/apps/backend/src/repositories/worker-platform/work-order-query-repo.ts
@@ -295,6 +295,37 @@ export async function listWorkOrders({
       : 'wo.id DESC'
   const effectiveOrderBy = pinnedFirst ? `wo.pinned_at DESC NULLS LAST, ${orderBy}` : orderBy
   params.push(pageSize, offset)
+
+  // 打手端「我的订单」的排序是表达式（依赖打手参数与拼单表），**没有索引能支撑**，
+  // 规划器只能把该打手全部可见单连同宽列（含 1.2KB 的 material_json）物化后排序。
+  // 生产实测：名下 6600 单的打手首访 1112ms、每次读约 60MB（列表缓存 2 秒一失效，
+  // 所以这个代价每分钟会重复付好几次）。这里改成"先只取当前页的 id，再按 id 取详情"：
+  // 排序阶段只带窄列，宽列只回表 20 行。语义（行集、顺序、总数）完全不变。
+  if (sort === 'worker_claimed_at_desc' && workerSharingId && !pinnedFirst) {
+    const idsResult = await query<{ id: number }>(
+      `SELECT wo.id
+       FROM work_orders wo
+       ${joins}
+       ${whereClause}
+       ORDER BY ${effectiveOrderBy}
+       LIMIT $${params.length - 1} OFFSET $${params.length}`,
+      params,
+    )
+    const orderedIds = idsResult.rows.map((row) => Number(row.id))
+    const [itemsResult, total] = await Promise.all([
+      orderedIds.length === 0
+        ? Promise.resolve({ rows: [] as WorkOrderRow[] })
+        : query<WorkOrderRow>(`${WORK_ORDER_SELECT} WHERE wo.id = ANY($1::bigint[])`, [orderedIds]),
+      includeTotal ? countWorkOrdersWithWhere(joins, whereClause, countParams) : Promise.resolve(0),
+    ])
+    // ANY() 不保证顺序，按 id 顺序还原分页结果
+    const rowsById = new Map(itemsResult.rows.map((row) => [Number(row.id), row]))
+    const items = orderedIds
+      .map((id) => rowsById.get(id))
+      .filter((row): row is WorkOrderRow => Boolean(row))
+    return { items, total }
+  }
+
   // 宽详情只查当前页；不将 COUNT 窗口放进详情查询，避免 LIMIT 前处理全量 JSON。
   const [itemsResult, total] = await Promise.all([
     query<WorkOrderRow>(
diff --git a/apps/backend/src/services/worker-platform/worker-platform-service.test.ts b/apps/backend/src/services/worker-platform/worker-platform-service.test.ts
index 4793a55a..ae70f176 100644
--- a/apps/backend/src/services/worker-platform/worker-platform-service.test.ts
+++ b/apps/backend/src/services/worker-platform/worker-platform-service.test.ts
@@ -2787,3 +2787,24 @@ test('统计的「状态+日期范围」组合必须有 created_at 键的覆盖
     '迁移在事务内执行，不能使用 CONCURRENTLY',
   )
 })
+
+test('打手端「我的订单」必须走"先取 id 再取详情"的延迟分页', async () => {
+  const { readFile } = await import('node:fs/promises')
+  const src = await readFile(
+    new URL('../../repositories/worker-platform/work-order-query-repo.ts', import.meta.url),
+    'utf8',
+  )
+  assert.ok(
+    src.includes("sort === 'worker_claimed_at_desc' && workerSharingId && !pinnedFirst"),
+    '缺少打手排序的延迟分页分支：该排序是表达式，无索引可支撑，会物化全部可见单的宽列',
+  )
+  assert.ok(
+    src.includes('SELECT wo.id'),
+    '第一阶段必须只取 id（窄投影），否则排序仍会读 material_json',
+  )
+  assert.ok(
+    src.includes('WHERE wo.id = ANY($1::bigint[])'),
+    '第二阶段必须按 id 批量取详情，而不是重新跑一遍带排序的宽查询',
+  )
+  assert.ok(src.includes('rowsById.get(id)'), 'ANY() 不保证顺序，必须按 id 顺序还原分页结果')
+})