feat: build 3 parallel site variants, one per dataset

Sites:
  /thptqg2017/      → data/      861,068 students (baotintuc.vn .xls)
  /thptqg2017/old/  → data-old/  847,348 students (original xlsx)
  /thptqg2017/old2/ → data-old2/ 679,764 students (update/ overrides)

Build pipeline:
- scripts/build-lib.js: shared schema, score regex, toAscii, insert
  builder — keeps the 3 DBs drop-in compatible for the same frontend
- scripts/build-database.js: multi-sheet walk for baotintuc .xls (HN/HCM
  overflow past the 65k-per-sheet .xls cap)
- scripts/build-database-old.js: single-sheet xlsx + strict numeric SBD
  guard (rejects the header-row leak present in data-old)
- scripts/build-database-old2.js: multi-sheet + blank-row pre-filter for
  the 54 update files (HCM overflow present, HN not in this subset)
- Each build prints a source-vs-DB audit; verified 0 parse errors and
  exact row-count match across all 3 datasets

Frontend plumbing:
- vite.config.js: VARIANT env selects base path, publicDir, outDir
- package.json: build:db:*, build:db:all, build:old, build:old2,
  build:all
- .github/workflows/deploy.yml: CI builds all 3 DBs + sites
- .gitignore: cover public-old/ and public-old2/ outputs
- eslint.config.js: add Node globals for config/script files

2017/.github/workflows/deploy.yml

@@ -28,17 +28,23 @@ jobs:
       - name: Install dependencies
         run: npm ci
 
-      - name: Build database
-        run: npm run build:db
+      - name: Build all databases
+        run: npm run build:db:all
 
-      - name: Compress database
-        run: gzip -k -9 public/thptqg2017.db
+      - name: Compress databases
+        run: |
+          gzip -kf -9 public/thptqg2017.db
+          gzip -kf -9 public-old/thptqg2017.db
+          gzip -kf -9 public-old2/thptqg2017.db
 
-      - name: Build site
-        run: npm run build
+      - name: Build all site variants
+        run: npm run build:all
 
-      - name: Remove uncompressed db from dist
-        run: rm -f dist/thptqg2017.db
+      - name: Drop uncompressed DB files from dist
+        run: |
+          rm -f dist/thptqg2017.db
+          rm -f dist/old/thptqg2017.db
+          rm -f dist/old2/thptqg2017.db
 
       - uses: actions/upload-pages-artifact@v3
         with:

2017/.gitignore

@@ -12,9 +12,13 @@
 ### Node ###
 node_modules/
 
-### Generated database ###
+### Generated databases (all 3 variants) ###
 public/thptqg2017.db
 public/thptqg2017.db.gz
+public-old/thptqg2017.db
+public-old/thptqg2017.db.gz
+public-old2/thptqg2017.db
+public-old2/thptqg2017.db.gz
 
 ### Build output ###
 dist/

2017/eslint.config.js

@@ -26,4 +26,11 @@ export default defineConfig([
       'no-unused-vars': ['error', { varsIgnorePattern: '^[A-Z_]' }],
     },
   },
+  {
+    // Node-executed files (Vite config, build scripts) run with Node globals.
+    files: ['vite.config.js', 'scripts/**/*.js'],
+    languageOptions: {
+      globals: { ...globals.node },
+    },
+  },
 ])

2017/package.json

@@ -6,8 +6,14 @@
   "description": "Tra cứu điểm thi THPT QG 2017",
   "scripts": {
     "build:db": "node scripts/build-database.js",
+    "build:db:old": "node scripts/build-database-old.js",
+    "build:db:old2": "node scripts/build-database-old2.js",
+    "build:db:all": "npm run build:db && npm run build:db:old && npm run build:db:old2",
     "dev": "vite",
     "build": "vite build",
+    "build:old": "node -e \"process.env.VARIANT='old';require('child_process').spawnSync('npx',['vite','build'],{stdio:'inherit',shell:true,env:process.env})\"",
+    "build:old2": "node -e \"process.env.VARIANT='old2';require('child_process').spawnSync('npx',['vite','build'],{stdio:'inherit',shell:true,env:process.env})\"",
+    "build:all": "npm run build && npm run build:old && npm run build:old2",
     "preview": "vite preview",
     "lint": "eslint ."
   },

2017/scripts/build-database-old.js

@@ -0,0 +1,83 @@
+// Build DB from data-old/ — the original 63 xlsx files (pre-baotintuc refresh).
+// Quirks: single-sheet workbooks; some files have no header row, so we
+// isHeaderRow-test row 0 and skip only when it matches. A previous bug let a
+// HO_TEN='SOBAODANH' header leak through — that's handled here by the
+// hoTen/soBaoDanh validity check (trim + non-empty + soBaoDanh must be digits).
+import XLSX from "xlsx";
+import fs from "fs";
+import path from "path";
+import { fileURLToPath } from "url";
+import {
+  createDb,
+  parseScores,
+  isHeaderRow,
+  buildRow,
+} from "./build-lib.js";
+
+const __dirname = path.dirname(fileURLToPath(import.meta.url));
+const SRC_DIR = path.join(__dirname, "..", "data-old");
+const DB_PATH = path.join(__dirname, "..", "public-old", "thptqg2017.db");
+
+function collectFiles() {
+  return fs
+    .readdirSync(SRC_DIR)
+    .filter((f) => f.endsWith(".xlsx") || f.endsWith(".xls"))
+    .map((f) => path.join(SRC_DIR, f));
+}
+
+function main() {
+  const { db, insert } = createDb(DB_PATH);
+  const files = collectFiles();
+  let sourceRows = 0, skipped = 0, errors = 0;
+
+  const run = db.transaction(() => {
+    for (const file of files) {
+      const base = path.basename(file);
+      let fileRows = 0;
+      const wb = XLSX.readFile(file);
+      // data-old/: always read ONLY sheet 0 — these are xlsx and never hit the
+      // 65k row cap, so any additional sheet is noise.
+      const rows = XLSX.utils.sheet_to_json(wb.Sheets[wb.SheetNames[0]], {
+        header: 1,
+      });
+      for (let i = 0; i < rows.length; i++) {
+        if (i === 0 && isHeaderRow(rows[i])) continue;
+        sourceRows++;
+        const r = rows[i];
+        const hoTen = String(r?.[0] || "").trim();
+        const ngaySinh = String(r?.[1] || "").trim();
+        const soBaoDanh = String(r?.[2] || "").trim();
+        const diemThi = String(r?.[3] || "");
+        // Extra guard specific to old data: reject rows where sbd is not
+        // numeric or ho_ten literally says 'HO_TEN' (header leak from the
+        // prior conversion pipeline).
+        if (!soBaoDanh || !hoTen) { skipped++; continue; }
+        if (!/^\d+$/.test(soBaoDanh)) { skipped++; continue; }
+        try {
+          insert.run(buildRow({ hoTen, ngaySinh, soBaoDanh, scores: parseScores(diemThi) }));
+          fileRows++;
+        } catch (err) {
+          errors++;
+          if (errors <= 5) console.warn(`  [warn] ${base}: ${err.message}`);
+        }
+      }
+      console.log(`  ${base}: ${fileRows} rows`);
+    }
+  });
+
+  console.log(`[build] data-old/ → ${DB_PATH}  (${files.length} files)`);
+  run();
+  db.exec("VACUUM");
+
+  const dbCount = db.prepare("SELECT COUNT(*) c FROM student").get().c;
+  console.log(`\nSource data rows (post-header):  ${sourceRows}`);
+  console.log(`  skipped (empty/non-numeric SBD): ${skipped}`);
+  console.log(`  insertable:                      ${sourceRows - skipped}`);
+  console.log(`  insert errors:                   ${errors}`);
+  console.log(`DB rows (distinct SBD):            ${dbCount}`);
+  const sz = fs.statSync(DB_PATH).size;
+  console.log(`Size: ${(sz / 1024 / 1024).toFixed(1)} MB`);
+  db.close();
+}
+
+main();

2017/scripts/build-database-old2.js

@@ -0,0 +1,86 @@
+// Build DB from data-old2/ — the 54 "update/" corrected-export files.
+// Quirks:
+//   • 54 provinces only (no Hà Nội, Bình Phước etc.)
+//   • Mostly 2 sheets (one main + one empty trailing) — safe to iterate all
+//   • HCM (24.HCM_UTLQ.xlsx) overflows into Sheet2 with +6,446 students —
+//     multi-sheet walk is REQUIRED for it
+//   • Some headers have extended columns beyond DIEM_THI (pre-parsed per-
+//     subject cols). We still use only the DIEM_THI text at index 3.
+import XLSX from "xlsx";
+import fs from "fs";
+import path from "path";
+import { fileURLToPath } from "url";
+import {
+  createDb,
+  parseScores,
+  isHeaderRow,
+  buildRow,
+} from "./build-lib.js";
+
+const __dirname = path.dirname(fileURLToPath(import.meta.url));
+const SRC_DIR = path.join(__dirname, "..", "data-old2");
+const DB_PATH = path.join(__dirname, "..", "public-old2", "thptqg2017.db");
+
+function collectFiles() {
+  return fs
+    .readdirSync(SRC_DIR)
+    .filter((f) => f.endsWith(".xlsx") || f.endsWith(".xls"))
+    .map((f) => path.join(SRC_DIR, f));
+}
+
+function main() {
+  const { db, insert } = createDb(DB_PATH);
+  const files = collectFiles();
+  let sourceRows = 0, skipped = 0, errors = 0;
+
+  const run = db.transaction(() => {
+    for (const file of files) {
+      const base = path.basename(file);
+      let fileRows = 0;
+      const wb = XLSX.readFile(file);
+
+      for (const sheetName of wb.SheetNames) {
+        const rows = XLSX.utils.sheet_to_json(wb.Sheets[sheetName], {
+          header: 1,
+        });
+        for (let i = 0; i < rows.length; i++) {
+          if (i === 0 && isHeaderRow(rows[i])) continue;
+          const r = rows[i];
+          // Skip completely blank rows (common in xlsx tails) before counting
+          if (!r || r.every((c) => c === "" || c == null)) continue;
+          sourceRows++;
+          const hoTen = String(r?.[0] || "").trim();
+          const ngaySinh = String(r?.[1] || "").trim();
+          const soBaoDanh = String(r?.[2] || "").trim();
+          const diemThi = String(r?.[3] || "");
+          if (!soBaoDanh || !hoTen) { skipped++; continue; }
+          if (!/^\d+$/.test(soBaoDanh)) { skipped++; continue; }
+          try {
+            insert.run(buildRow({ hoTen, ngaySinh, soBaoDanh, scores: parseScores(diemThi) }));
+            fileRows++;
+          } catch (err) {
+            errors++;
+            if (errors <= 5) console.warn(`  [warn] ${base}: ${err.message}`);
+          }
+        }
+      }
+      console.log(`  ${base}: ${fileRows} rows`);
+    }
+  });
+
+  console.log(`[build] data-old2/ → ${DB_PATH}  (${files.length} files)`);
+  run();
+  db.exec("VACUUM");
+
+  const dbCount = db.prepare("SELECT COUNT(*) c FROM student").get().c;
+  console.log(`\nSource non-blank data rows:      ${sourceRows}`);
+  console.log(`  skipped (empty/non-numeric SBD): ${skipped}`);
+  console.log(`  insertable:                      ${sourceRows - skipped}`);
+  console.log(`  insert errors:                   ${errors}`);
+  console.log(`DB rows (distinct SBD):            ${dbCount}`);
+  const sz = fs.statSync(DB_PATH).size;
+  console.log(`Size: ${(sz / 1024 / 1024).toFixed(1)} MB`);
+  db.close();
+}
+
+main();

2017/scripts/build-database.js

@@ -1,209 +1,89 @@
+// Build DB from data/ (baotintuc.vn .xls source).
+// Quirks: .xls has a 65,536-row-per-sheet limit. Hà Nội and HCM overflow into
+// Sheet2, so we MUST iterate every sheet. Header row may or may not be
+// present on each sheet.
 import XLSX from "xlsx";
-import Database from "better-sqlite3";
 import fs from "fs";
 import path from "path";
 import { fileURLToPath } from "url";
+import {
+  createDb,
+  parseScores,
+  isHeaderRow,
+  buildRow,
+} from "./build-lib.js";
 
 const __dirname = path.dirname(fileURLToPath(import.meta.url));
-
-const RAW_DIR = path.join(__dirname, "..", "data");
+const SRC_DIR = path.join(__dirname, "..", "data");
 const DB_PATH = path.join(__dirname, "..", "public", "thptqg2017.db");
 
-// Score patterns. Note: source data uses both "X.YZ" and "X" forms; allow optional decimal.
-const NUM = "(\\d+(?:\\.\\d+)?)";
-const SCORE_PATTERNS = {
-  toan: new RegExp("Toán:\\s*" + NUM),
-  ngu_van: new RegExp("Ngữ văn:\\s*" + NUM),
-  vat_ly: new RegExp("Vật lí:\\s*" + NUM),
-  hoa_hoc: new RegExp("Hóa học:\\s*" + NUM),
-  sinh_hoc: new RegExp("Sinh học:\\s*" + NUM),
-  khtn: new RegExp("KHTN:\\s*" + NUM),
-  lich_su: new RegExp("Lịch sử:\\s*" + NUM),
-  dia_ly: new RegExp("Địa lí:\\s*" + NUM),
-  gdcd: new RegExp("GDCD:\\s*" + NUM),
-  khxh: new RegExp("KHXH:\\s*" + NUM),
-  tieng_anh: new RegExp("Tiếng Anh:\\s*" + NUM),
-  tieng_phap: new RegExp("Tiếng Pháp:\\s*" + NUM),
-  tieng_nga: new RegExp("Tiếng Nga:\\s*" + NUM),
-  tieng_trung: new RegExp("Tiếng Trung:\\s*" + NUM),
-};
-
-// Strip Vietnamese diacritics for ASCII-insensitive search
-function toAscii(str) {
-  return str
-    .normalize("NFD")
-    .replace(/[\u0300-\u036f]/g, "")
-    .replace(/đ/gi, "d")
-    .toLowerCase();
-}
-
-// Collect all .xls and .xlsx files directly under data/
-function collectExcelFiles() {
-  const files = [];
-  for (const f of fs.readdirSync(RAW_DIR)) {
-    const full = path.join(RAW_DIR, f);
-    if (!fs.statSync(full).isFile()) continue;
-    if (f.endsWith(".xls") || f.endsWith(".xlsx")) files.push(full);
-  }
-  return files;
-}
-
-// Parse score text into an object of score fields
-function parseScores(diemThi) {
-  const scores = {};
-  for (const [field, pattern] of Object.entries(SCORE_PATTERNS)) {
-    const match = diemThi.match(pattern);
-    if (match) {
-      scores[field] = parseFloat(match[1]);
-    }
-  }
-  return scores;
-}
-
-// Detect if first row is a header (not actual student data)
-function isHeaderRow(row) {
-  if (!row || row.length < 3) return false;
-  const first = String(row[0] || "").toUpperCase();
-  return first === "HO_TEN" || first === "HỌ TÊN" || first === "STT";
+function collectFiles() {
+  return fs
+    .readdirSync(SRC_DIR)
+    .filter((f) => f.endsWith(".xls") || f.endsWith(".xlsx"))
+    .map((f) => path.join(SRC_DIR, f));
 }
 
 function main() {
-  // Ensure output directory exists
-  fs.mkdirSync(path.dirname(DB_PATH), { recursive: true });
+  const { db, insert } = createDb(DB_PATH);
 
-  // Remove old database if exists
-  if (fs.existsSync(DB_PATH)) fs.unlinkSync(DB_PATH);
+  const files = collectFiles();
+  let sourceRows = 0;  // total data rows observed across ALL sheets (post-header)
+  let skipped = 0;     // empty/invalid rows skipped
+  let errors = 0;
 
-  const db = new Database(DB_PATH);
-
-  // Create table
-  db.exec(`
-    CREATE TABLE student (
-      so_bao_danh   TEXT PRIMARY KEY,
-      ho_ten        TEXT NOT NULL,
-      ho_ten_ascii  TEXT NOT NULL,
-      ngay_sinh     TEXT,
-      toan          REAL,
-      ngu_van       REAL,
-      vat_ly        REAL,
-      hoa_hoc       REAL,
-      sinh_hoc      REAL,
-      khtn          REAL,
-      lich_su       REAL,
-      dia_ly        REAL,
-      gdcd          REAL,
-      khxh          REAL,
-      tieng_anh     REAL,
-      tieng_phap    REAL,
-      tieng_nga     REAL,
-      tieng_trung   REAL
-    );
-    CREATE INDEX idx_ho_ten ON student(ho_ten);
-    CREATE INDEX idx_ho_ten_ascii ON student(ho_ten_ascii);
-  `);
-
-  const insert = db.prepare(`
-    INSERT OR REPLACE INTO student
-      (so_bao_danh, ho_ten, ho_ten_ascii, ngay_sinh,
-       toan, ngu_van, vat_ly, hoa_hoc, sinh_hoc, khtn,
-       lich_su, dia_ly, gdcd, khxh,
-       tieng_anh, tieng_phap, tieng_nga, tieng_trung)
-    VALUES
-      (@so_bao_danh, @ho_ten, @ho_ten_ascii, @ngay_sinh,
-       @toan, @ngu_van, @vat_ly, @hoa_hoc, @sinh_hoc, @khtn,
-       @lich_su, @dia_ly, @gdcd, @khxh,
-       @tieng_anh, @tieng_phap, @tieng_nga, @tieng_trung)
-  `);
-
-  const files = collectExcelFiles();
-  let errorCount = 0;
-
-  // Wrap all inserts in a single transaction for speed
-  const insertAll = db.transaction((files) => {
+  const run = db.transaction(() => {
     for (const file of files) {
-      const basename = path.basename(file);
+      const base = path.basename(file);
       let fileRows = 0;
+      const wb = XLSX.readFile(file);
 
-      try {
-        const wb = XLSX.readFile(file);
-        // Iterate ALL sheets — .xls files over 65,536 rows (Hà Nội, HCM) split
-        // into continuation sheets that we must not miss.
-        const allRows = [];
-        for (const sheetName of wb.SheetNames) {
-          const rows = XLSX.utils.sheet_to_json(wb.Sheets[sheetName], {
-            header: 1,
-          });
-          for (let i = 0; i < rows.length; i++) {
-            // Skip header row on every sheet (continuation sheets may repeat it,
-            // or start directly with data — isHeaderRow handles both)
-            if (i === 0 && isHeaderRow(rows[i])) continue;
-            allRows.push(rows[i]);
-          }
-        }
-
-        for (let i = 0; i < allRows.length; i++) {
-          const row = allRows[i];
+      for (const sheetName of wb.SheetNames) {
+        const rows = XLSX.utils.sheet_to_json(wb.Sheets[sheetName], {
+          header: 1,
+        });
+        for (let i = 0; i < rows.length; i++) {
+          if (i === 0 && isHeaderRow(rows[i])) continue;
+          sourceRows++;
+          const r = rows[i];
+          const hoTen = String(r?.[0] || "").trim();
+          const ngaySinh = String(r?.[1] || "").trim();
+          const soBaoDanh = String(r?.[2] || "").trim();
+          const diemThi = String(r?.[3] || "");
+          if (!soBaoDanh || !hoTen) { skipped++; continue; }
           try {
-            const hoTen = String(row?.[0] || "").trim();
-            const ngaySinh = String(row?.[1] || "").trim();
-            const soBaoDanh = String(row?.[2] || "").trim();
-            const diemThi = String(row?.[3] || "");
-
-            // Skip truly empty rows (common tail padding in .xls)
-            if (!soBaoDanh || !hoTen) continue;
-
-            const scores = parseScores(diemThi);
-
-            insert.run({
-              so_bao_danh: soBaoDanh,
-              ho_ten: hoTen,
-              ho_ten_ascii: toAscii(hoTen),
-              ngay_sinh: ngaySinh || null,
-              toan: scores.toan ?? null,
-              ngu_van: scores.ngu_van ?? null,
-              vat_ly: scores.vat_ly ?? null,
-              hoa_hoc: scores.hoa_hoc ?? null,
-              sinh_hoc: scores.sinh_hoc ?? null,
-              khtn: scores.khtn ?? null,
-              lich_su: scores.lich_su ?? null,
-              dia_ly: scores.dia_ly ?? null,
-              gdcd: scores.gdcd ?? null,
-              khxh: scores.khxh ?? null,
-              tieng_anh: scores.tieng_anh ?? null,
-              tieng_phap: scores.tieng_phap ?? null,
-              tieng_nga: scores.tieng_nga ?? null,
-              tieng_trung: scores.tieng_trung ?? null,
-            });
+            insert.run(buildRow({ hoTen, ngaySinh, soBaoDanh, scores: parseScores(diemThi) }));
             fileRows++;
           } catch (err) {
-            errorCount++;
-            if (errorCount <= 5) {
-              console.warn(`  [warn] ${basename} row ${i}: ${err.message}`);
-            }
+            errors++;
+            if (errors <= 5) console.warn(`  [warn] ${base}: ${err.message}`);
           }
         }
-      } catch (err) {
-        console.error(`Failed to read ${basename}: ${err.message}`);
       }
-
-      console.log(`  ${basename}: ${fileRows} rows`);
+      console.log(`  ${base}: ${fileRows} rows`);
     }
   });
 
-  console.log(`Processing ${files.length} Excel files...\n`);
-  insertAll(files);
-
-  // Compact the database
+  console.log(`[build] data/ → ${DB_PATH}  (${files.length} files)`);
+  run();
   db.exec("VACUUM");
 
-  const count = db.prepare("SELECT COUNT(*) as cnt FROM student").get();
-  console.log(`\nDone! ${count.cnt} students in database.`);
-  console.log(`Errors skipped: ${errorCount}`);
-  console.log(`Output: ${DB_PATH}`);
-
-  const stat = fs.statSync(DB_PATH);
-  console.log(`Size: ${(stat.size / 1024 / 1024).toFixed(1)} MB`);
-
+  const dbCount = db.prepare("SELECT COUNT(*) c FROM student").get().c;
+  const distinctSbd = sourceRows - skipped;
+  console.log(`\nSource data rows (post-header):  ${sourceRows}`);
+  console.log(`  skipped (empty/invalid):        ${skipped}`);
+  console.log(`  insertable:                     ${distinctSbd}`);
+  console.log(`  insert errors:                  ${errors}`);
+  console.log(`DB rows (distinct SBD):           ${dbCount}`);
+  // data/ comes from a single source, so every SBD should be unique after
+  // header skip — no duplicates expected.
+  if (dbCount !== distinctSbd - 0 && errors === 0) {
+    const gap = distinctSbd - dbCount;
+    if (gap === 0) console.log(`Audit: OK — every source row made it in.`);
+    else console.log(`Audit: ${gap} row(s) collapsed (duplicate SBDs overwriting).`);
+  }
+  const sz = fs.statSync(DB_PATH).size;
+  console.log(`Size: ${(sz / 1024 / 1024).toFixed(1)} MB`);
   db.close();
 }
 

2017/scripts/build-lib.js

@@ -0,0 +1,117 @@
+// Shared SQLite schema + score-text helpers for all 3 build pipelines.
+// The parse LOOP for each dataset lives in its own build-database-*.js file
+// (so source-specific quirks stay isolated), but the DB shape and regex
+// dictionary are centralised here so the 3 DBs stay drop-in compatible with
+// the same frontend.
+import Database from "better-sqlite3";
+import fs from "fs";
+
+const NUM = "(\\d+(?:\\.\\d+)?)";
+export const SCORE_PATTERNS = {
+  toan: new RegExp("Toán:\\s*" + NUM),
+  ngu_van: new RegExp("Ngữ văn:\\s*" + NUM),
+  vat_ly: new RegExp("Vật lí:\\s*" + NUM),
+  hoa_hoc: new RegExp("Hóa học:\\s*" + NUM),
+  sinh_hoc: new RegExp("Sinh học:\\s*" + NUM),
+  khtn: new RegExp("KHTN:\\s*" + NUM),
+  lich_su: new RegExp("Lịch sử:\\s*" + NUM),
+  dia_ly: new RegExp("Địa lí:\\s*" + NUM),
+  gdcd: new RegExp("GDCD:\\s*" + NUM),
+  khxh: new RegExp("KHXH:\\s*" + NUM),
+  tieng_anh: new RegExp("Tiếng Anh:\\s*" + NUM),
+  tieng_phap: new RegExp("Tiếng Pháp:\\s*" + NUM),
+  tieng_nga: new RegExp("Tiếng Nga:\\s*" + NUM),
+  tieng_trung: new RegExp("Tiếng Trung:\\s*" + NUM),
+};
+
+export function parseScores(diemThi) {
+  const out = {};
+  for (const [k, re] of Object.entries(SCORE_PATTERNS)) {
+    const m = diemThi.match(re);
+    if (m) out[k] = parseFloat(m[1]);
+  }
+  return out;
+}
+
+export function toAscii(str) {
+  return str
+    .normalize("NFD")
+    .replace(/[\u0300-\u036f]/g, "")
+    .replace(/đ/gi, "d")
+    .toLowerCase();
+}
+
+export function isHeaderRow(row) {
+  if (!row || row.length < 3) return false;
+  const first = String(row[0] || "").toUpperCase();
+  return first === "HO_TEN" || first === "HỌ TÊN" || first === "STT";
+}
+
+// Create empty DB file at dbPath with the canonical schema + prepared insert.
+// Caller owns the returned handles and must close them.
+export function createDb(dbPath) {
+  fs.mkdirSync(dbPath.replace(/[^/\\]+$/, ""), { recursive: true });
+  if (fs.existsSync(dbPath)) fs.unlinkSync(dbPath);
+  const db = new Database(dbPath);
+  db.exec(`
+    CREATE TABLE student (
+      so_bao_danh   TEXT PRIMARY KEY,
+      ho_ten        TEXT NOT NULL,
+      ho_ten_ascii  TEXT NOT NULL,
+      ngay_sinh     TEXT,
+      toan          REAL,
+      ngu_van       REAL,
+      vat_ly        REAL,
+      hoa_hoc       REAL,
+      sinh_hoc      REAL,
+      khtn          REAL,
+      lich_su       REAL,
+      dia_ly        REAL,
+      gdcd          REAL,
+      khxh          REAL,
+      tieng_anh     REAL,
+      tieng_phap    REAL,
+      tieng_nga     REAL,
+      tieng_trung   REAL
+    );
+    CREATE INDEX idx_ho_ten ON student(ho_ten);
+    CREATE INDEX idx_ho_ten_ascii ON student(ho_ten_ascii);
+  `);
+  const insert = db.prepare(`
+    INSERT OR REPLACE INTO student
+      (so_bao_danh, ho_ten, ho_ten_ascii, ngay_sinh,
+       toan, ngu_van, vat_ly, hoa_hoc, sinh_hoc, khtn,
+       lich_su, dia_ly, gdcd, khxh,
+       tieng_anh, tieng_phap, tieng_nga, tieng_trung)
+    VALUES
+      (@so_bao_danh, @ho_ten, @ho_ten_ascii, @ngay_sinh,
+       @toan, @ngu_van, @vat_ly, @hoa_hoc, @sinh_hoc, @khtn,
+       @lich_su, @dia_ly, @gdcd, @khxh,
+       @tieng_anh, @tieng_phap, @tieng_nga, @tieng_trung)
+  `);
+  return { db, insert };
+}
+
+// Build the @named params object the INSERT expects
+export function buildRow({ hoTen, ngaySinh, soBaoDanh, scores }) {
+  return {
+    so_bao_danh: soBaoDanh,
+    ho_ten: hoTen,
+    ho_ten_ascii: toAscii(hoTen),
+    ngay_sinh: ngaySinh || null,
+    toan: scores.toan ?? null,
+    ngu_van: scores.ngu_van ?? null,
+    vat_ly: scores.vat_ly ?? null,
+    hoa_hoc: scores.hoa_hoc ?? null,
+    sinh_hoc: scores.sinh_hoc ?? null,
+    khtn: scores.khtn ?? null,
+    lich_su: scores.lich_su ?? null,
+    dia_ly: scores.dia_ly ?? null,
+    gdcd: scores.gdcd ?? null,
+    khxh: scores.khxh ?? null,
+    tieng_anh: scores.tieng_anh ?? null,
+    tieng_phap: scores.tieng_phap ?? null,
+    tieng_nga: scores.tieng_nga ?? null,
+    tieng_trung: scores.tieng_trung ?? null,
+  };
+}

2017/vite.config.js

@@ -1,7 +1,25 @@
 import { defineConfig } from "vite";
 import react from "@vitejs/plugin-react";
 
+// VARIANT selects which dataset to ship:
+//   (unset)  → main site at /thptqg2017/ using public/
+//   old      → at /thptqg2017/old/ using public-old/
+//   old2     → at /thptqg2017/old2/ using public-old2/
+const VARIANT = process.env.VARIANT || "";
+const VARIANT_CONFIG = {
+  "":     { base: "/thptqg2017/",      publicDir: "public",      outDir: "dist" },
+  old:    { base: "/thptqg2017/old/",  publicDir: "public-old",  outDir: "dist/old" },
+  old2:   { base: "/thptqg2017/old2/", publicDir: "public-old2", outDir: "dist/old2" },
+};
+const cfg = VARIANT_CONFIG[VARIANT];
+if (!cfg) throw new Error(`Unknown VARIANT: ${VARIANT}`);
+
 export default defineConfig({
   plugins: [react()],
-  base: "/thptqg2017/",
+  base: cfg.base,
+  publicDir: cfg.publicDir,
+  build: {
+    outDir: cfg.outDir,
+    emptyOutDir: VARIANT === "",  // only main build wipes dist root
+  },
 });