diff --git a/2017/plans/260412-2303-static-score-lookup/phase-01-project-scaffolding.md b/2017/plans/260412-2303-static-score-lookup/phase-01-project-scaffolding.md
deleted file mode 100644
index 10777ee..0000000
--- a/2017/plans/260412-2303-static-score-lookup/phase-01-project-scaffolding.md
+++ /dev/null
@@ -1,117 +0,0 @@
-# Phase 01 — Project Scaffolding
-
-## Context Links
-- [plan.md](./plan.md)
-- [Converter.java](../../src/main/java/dev/miti99/thptqg2017/Converter.java) — reference parsing logic
-- [Student.java](../../src/main/java/dev/miti99/thptqg2017/entity/Student.java) — reference data model
-
-## Overview
-- **Priority**: P1 (blocker for all other phases)
-- **Status**: Pending
-- **Description**: Initialize Node.js project, Vite+React app, and folder structure
-
-## Key Decisions
-
-- **Monorepo with single package.json** at root — project is small, no need for workspaces
-- **better-sqlite3** for build script (fast native writes), **sql.js** for browser runtime
-- **Vite + React + TypeScript** for the static site
-- Keep Java source intact — no deletions in this phase
-
-## Architecture
-
-```
-thptqg2017/
-├── scripts/                    # Node.js build scripts
-│   └── build-database.js       # Excel -> SQLite converter
-├── web/                        # Vite + React app
-│   ├── src/
-│   │   ├── main.tsx
-│   │   ├── app.tsx
-│   │   └── components/
-│   ├── public/                 # .db file goes here after build
-│   ├── index.html
-│   └── vite.config.ts
-├── package.json                # Root: scripts + web deps
-├── src/main/resources/raw/     # Existing Excel files (unchanged)
-└── ... (existing Java files unchanged)
-```
-
-## Related Code Files
-
-**Create:**
-- `package.json` — root package with workspaces or flat deps
-- `web/index.html` — Vite entry
-- `web/vite.config.ts`
-- `web/src/main.tsx`
-- `web/src/app.tsx`
-- `web/tsconfig.json`
-- `scripts/` directory (empty, placeholder)
-
-**No modifications to existing files.**
-
-## Implementation Steps
-
-1. Initialize `package.json` at project root
-   ```bash
-   npm init -y
-   ```
-2. Install build-script dependencies
-   ```bash
-   npm install --save-dev better-sqlite3 xlsx
-   ```
-3. Install web dependencies
-   ```bash
-   npm install react react-dom sql.js
-   npm install --save-dev @types/react @types/react-dom typescript vite @vitejs/plugin-react
-   ```
-4. Create `web/` directory with Vite scaffold
-   - `web/index.html` — minimal HTML shell
-   - `web/vite.config.ts` — configure root, public dir, build output
-   - `web/src/main.tsx` — React entry
-   - `web/src/app.tsx` — placeholder App component
-   - `web/tsconfig.json` — strict TS config
-5. Create `scripts/` directory
-6. Add npm scripts to `package.json`:
-   ```json
-   {
-     "scripts": {
-       "build:db": "node scripts/build-database.js",
-       "dev": "vite --config web/vite.config.ts",
-       "build": "vite build --config web/vite.config.ts",
-       "preview": "vite preview --config web/vite.config.ts"
-     }
-   }
-   ```
-7. Update `.gitignore` to add:
-   ```
-   node_modules/
-   web/dist/
-   web/public/thptqg2017.db
-   ```
-8. Verify `npm run dev` starts without errors
-
-## Todo List
-
-- [ ] Create package.json with all dependencies
-- [ ] Create web/ directory with Vite + React scaffold
-- [ ] Create scripts/ directory
-- [ ] Update .gitignore
-- [ ] Verify `npm run dev` starts clean
-
-## Success Criteria
-
-- `npm install` completes without errors
-- `npm run dev` opens a blank React page in browser
-- No existing files modified (Java code untouched)
-- All new files use kebab-case naming
-
-## Risk Assessment
-
-| Risk | Mitigation |
-|------|------------|
-| better-sqlite3 native build fails on Windows | Use prebuild binaries (default); fallback to sql.js for build script too |
-| Vite config path issues with nested web/ dir | Set `root: 'web'` in vite.config.ts explicitly |
-
-## Next Steps
-- Phase 02 depends on `scripts/` dir and `better-sqlite3` being available
-- Phase 03 depends on `web/` scaffold and React being configured
diff --git a/2017/plans/260412-2303-static-score-lookup/phase-02-excel-parser-db-builder.md b/2017/plans/260412-2303-static-score-lookup/phase-02-excel-parser-db-builder.md
deleted file mode 100644
index 953560c..0000000
--- a/2017/plans/260412-2303-static-score-lookup/phase-02-excel-parser-db-builder.md
+++ /dev/null
@@ -1,201 +0,0 @@
-# Phase 02 — Excel Parser + DB Builder
-
-## Context Links
-- [plan.md](./plan.md)
-- [phase-01](./phase-01-project-scaffolding.md) — prerequisite
-- [Converter.java](../../src/main/java/dev/miti99/thptqg2017/Converter.java) — source parsing logic to port
-
-## Overview
-- **Priority**: P1 (produces the .db file needed by Phase 03)
-- **Status**: Pending
-- **Description**: Node.js script that reads ~119 Excel files, regex-parses scores, writes SQLite .db
-
-## Key Insights
-
-From Converter.java analysis:
-- 4 columns per row: hoTen (0), ngaySinh (1), soBaoDanh (2), diemThi (3)
-- Column 3 is a text blob with scores like `"Toán: 8.5 Ngữ văn: 7.0 Vật lí: 6.25 ..."`
-- 11 subject patterns extracted via regex; not all subjects present for every student
-- Some files have no header row (comment in Java: "Một số file lỗi nên không chắc có header")
-- `session.merge()` = upsert behavior; update folder files should overwrite raw folder data
-- ngaySinh format: `dd/MM/yyyy`
-
-**Format edge cases to handle:**
-- Row 0 might be header OR data — detect by checking if cell 2 (soBaoDanh) looks like an ID pattern
-- Some cells may be numeric instead of string (Excel auto-detection)
-- Duplicate .xlsx files exist: `10_LamDong_GNFT (1).xls.xlsx` and `10_LamDong_GNFT.xls.xlsx`
-
-## Requirements
-
-### Functional
-- Parse all .xlsx from `src/main/resources/raw/` and `src/main/resources/raw/(update)/`
-- Process `raw/` first, then `(update)/` so updates overwrite via INSERT OR REPLACE
-- Extract all 11 subject scores via regex (matching Java patterns exactly)
-- Store ngaySinh as text string (no date conversion)
-- Output `web/public/thptqg2017.db`
-
-### Non-Functional
-- Complete in < 60 seconds on modern hardware
-- Log progress: file count, row count, error count
-- Skip malformed rows gracefully (log, don't crash)
-
-## Architecture
-
-```
-scripts/build-database.js
-  ├── reads: src/main/resources/raw/**/*.xlsx
-  ├── reads: src/main/resources/raw/(update)/**/*.xlsx
-  ├── uses: xlsx (SheetJS) for Excel parsing
-  ├── uses: better-sqlite3 for fast SQLite writes
-  └── outputs: web/public/thptqg2017.db
-```
-
-### Data Flow Per File
-
-```
-.xlsx file
-  → xlsx.readFile()
-  → sheet_to_json({ header: 1, raw: false })  // array of arrays, all strings
-  → for each row:
-      → validate: row[2] matches soBaoDanh pattern (skip headers/junk)
-      → extract: hoTen, ngaySinh, soBaoDanh from columns 0-2
-      → regex match: diemThi (column 3) against 11 subject patterns
-      → INSERT OR REPLACE into student table
-```
-
-## Related Code Files
-
-**Create:**
-- `scripts/build-database.js` — main parser script (~120 lines)
-
-**Read (reference only):**
-- `src/main/java/dev/miti99/thptqg2017/Converter.java`
-
-**No modifications to existing files.**
-
-## Implementation Steps
-
-### 1. Create score-parsing utility
-
-Port the 11 regex patterns from Converter.java:
-
-```javascript
-const SCORE_PATTERNS = {
-  toan:      /Toán:\s*(\d*\.\d*)/,
-  ngu_van:   /Ngữ văn:\s*(\d*\.\d*)/,
-  vat_ly:    /Vật lí:\s*(\d*\.\d*)/,
-  hoa_hoc:   /Hóa học:\s*(\d*\.\d*)/,
-  sinh_hoc:  /Sinh học:\s*(\d*\.\d*)/,
-  khtn:      /KHTN:\s*(\d*\.\d*)/,
-  lich_su:   /Lịch sử:\s*(\d*\.\d*)/,
-  dia_ly:    /Địa lí:\s*(\d*\.\d*)/,
-  gdcd:      /GDCD:\s*(\d*\.\d*)/,
-  khxh:      /KHXH:\s*(\d*\.\d*)/,
-  tieng_anh: /Tiếng Anh:\s*(\d*\.\d*)/,
-};
-```
-
-### 2. Create database schema
-
-```javascript
-db.exec(`
-  CREATE TABLE IF NOT EXISTS student (
-    so_bao_danh TEXT PRIMARY KEY,
-    ho_ten      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
-  );
-`);
-```
-
-### 3. Build the main parsing loop
-
-```
-for each folder in [raw/, raw/(update)/]:
-  for each .xlsx file:
-    workbook = xlsx.readFile(filePath)
-    sheet = workbook.Sheets[workbook.SheetNames[0]]
-    rows = xlsx.utils.sheet_to_json(sheet, { header: 1, raw: false })
-    for each row:
-      if row[2] doesn't look like a valid soBaoDanh → skip (header detection)
-      parse scores from row[3]
-      INSERT OR REPLACE
-```
-
-### 4. Header detection heuristic
-
-```javascript
-// soBaoDanh is typically a numeric string like "02000001"
-function isDataRow(row) {
-  return row[2] && /^\d{6,}$/.test(String(row[2]).trim());
-}
-```
-
-### 5. Wrap inserts in a transaction
-
-```javascript
-const insert = db.prepare(`INSERT OR REPLACE INTO student (...) VALUES (...)`);
-const insertMany = db.transaction((rows) => {
-  for (const row of rows) insert.run(row);
-});
-```
-
-### 6. Add indexes after all inserts
-
-```javascript
-db.exec('CREATE INDEX IF NOT EXISTS idx_ho_ten ON student(ho_ten)');
-```
-
-### 7. Add summary logging
-
-```
-console.log(`Processed ${fileCount} files, ${rowCount} rows, ${errorCount} errors`);
-```
-
-### 8. Ensure output directory exists
-
-```javascript
-fs.mkdirSync('web/public', { recursive: true });
-```
-
-## Todo List
-
-- [ ] Create `scripts/build-database.js`
-- [ ] Port all 11 regex patterns from Converter.java
-- [ ] Implement header-detection heuristic
-- [ ] Process raw/ folder first, then (update)/ folder
-- [ ] Wrap all inserts in single transaction for speed
-- [ ] Add index on ho_ten after insert
-- [ ] Log file count, row count, error count
-- [ ] Run script and verify row count matches Java output
-- [ ] Verify .db file size is reasonable (< 60MB expected)
-
-## Success Criteria
-
-1. `node scripts/build-database.js` completes without unhandled errors
-2. Output .db contains same number of unique students as existing database.sqlite
-3. Spot-check: query 5 random soBaoDanh values → scores match between old and new DB
-4. Script completes in < 60 seconds
-5. Skipped/errored rows are logged with file name and row number
-
-## Risk Assessment
-
-| Risk | Likelihood | Impact | Mitigation |
-|------|-----------|--------|------------|
-| xlsx library misreads Vietnamese characters | Low | High | Use `{ raw: false }` to get string values; test with known file |
-| Regex patterns don't match all score formats | Medium | Medium | Test against sample rows; add `\d+\.?\d*` fallback if needed |
-| better-sqlite3 won't install on Windows | Low | Medium | Prebuild binaries exist; fallback: use sql.js for build too |
-| Memory pressure with 119 files open | Low | Low | Process files sequentially, one at a time |
-| Duplicate students from overlapping files | Medium | Low | INSERT OR REPLACE handles this; (update) processed last wins |
-
-## Security Considerations
-- Public exam data, no PII concerns beyond names (already public)
-- No network access needed during build
-- Output .db is read-only in production
-
-## Next Steps
-- Phase 03 consumes the `web/public/thptqg2017.db` file produced here
-- Verify row count against existing database before proceeding
diff --git a/2017/plans/260412-2303-static-score-lookup/phase-03-react-static-site.md b/2017/plans/260412-2303-static-score-lookup/phase-03-react-static-site.md
deleted file mode 100644
index 16b9721..0000000
--- a/2017/plans/260412-2303-static-score-lookup/phase-03-react-static-site.md
+++ /dev/null
@@ -1,201 +0,0 @@
-# Phase 03 — React Static Site
-
-## Context Links
-- [plan.md](./plan.md)
-- [phase-01](./phase-01-project-scaffolding.md) — provides Vite+React scaffold
-- [phase-02](./phase-02-excel-parser-db-builder.md) — provides .db file
-
-## Overview
-- **Priority**: P1
-- **Status**: Pending
-- **Description**: Vite+React+TypeScript site that loads SQLite .db client-side via sql.js for score lookup
-
-## Key Insights
-
-- .db file is ~50-60MB — must show loading progress to user
-- sql.js requires WASM file — load from CDN (cdnjs) or bundle in public/
-- All queries are readonly SELECT — no write operations
-- Vietnamese UI — all labels, placeholders, messages in Vietnamese
-- Two search modes: by name (ho_ten LIKE) and by exam ID (so_bao_danh exact match)
-
-## Requirements
-
-### Functional
-- Load .db file on page load with progress indicator
-- Search by số báo danh (exact match) or họ tên (partial match, case-insensitive)
-- Display results in a clean table with all score columns
-- Show "không tìm thấy" when no results
-- Limit results to 50 rows (prevent rendering 800K rows)
-
-### Non-Functional
-- First meaningful paint < 2s (before DB loads)
-- Search response < 100ms after DB is loaded
-- Mobile-responsive layout
-- Works offline after initial load (static site + cached .db)
-
-## Architecture
-
-```
-Browser
-  ├── index.html (Vite entry)
-  ├── main.tsx → App
-  │     ├── useSqlite hook (loads .db, exposes query fn)
-  │     ├── SearchBar component (input + mode toggle)
-  │     └── ResultTable component (score display)
-  └── sql.js WASM (from CDN)
-```
-
-### Data Flow
-
-```
-Page Load
-  → fetch('/thptqg2017.db') with progress tracking
-  → initSqlJs({ locateFile: cdnjs url })
-  → new SQL.Database(arrayBuffer)
-  → DB ready, enable search
-
-User Types Query
-  → debounce 300ms
-  → if mode=id:  SELECT * FROM student WHERE so_bao_danh = ?
-  → if mode=name: SELECT * FROM student WHERE ho_ten LIKE ? LIMIT 50
-  → render ResultTable with rows
-```
-
-## Related Code Files
-
-**Create:**
-- `web/src/app.tsx` — main app layout (~60 lines)
-- `web/src/hooks/use-sqlite.ts` — sql.js loading + query hook (~70 lines)
-- `web/src/components/search-bar.tsx` — search input + mode toggle (~40 lines)
-- `web/src/components/result-table.tsx` — score results table (~60 lines)
-- `web/src/types/student.ts` — TypeScript interface (~20 lines)
-- `web/src/index.css` — minimal styling (~50 lines)
-
-**Modify:**
-- `web/src/main.tsx` — import App + CSS
-
-## Implementation Steps
-
-### 1. Define Student type
-
-```typescript
-// web/src/types/student.ts
-export interface Student {
-  so_bao_danh: string;
-  ho_ten: string;
-  ngay_sinh: string | null;
-  toan: number | null;
-  ngu_van: number | null;
-  vat_ly: number | null;
-  hoa_hoc: number | null;
-  sinh_hoc: number | null;
-  khtn: number | null;
-  lich_su: number | null;
-  dia_ly: number | null;
-  gdcd: number | null;
-  khxh: number | null;
-  tieng_anh: number | null;
-}
-```
-
-### 2. Implement use-sqlite hook
-
-```typescript
-// web/src/hooks/use-sqlite.ts
-// States: loading (with progress %), ready, error
-// On mount: fetch .db file → init sql.js → create Database instance
-// Expose: { db, loading, progress, error, query(sql, params) }
-```
-
-Key details:
-- Use `fetch()` with `response.body.getReader()` for progress tracking
-- sql.js WASM from: `https://cdnjs.cloudflare.com/ajax/libs/sql.js/1.11.0/sql-wasm.wasm`
-- Memoize the Database instance with useRef
-
-### 3. Implement search-bar component
-
-- Single text input with placeholder "Nhập số báo danh hoặc họ tên..."
-- Auto-detect mode: if input is all digits → search by soBaoDanh; else → search by hoTen
-- Debounce input by 300ms before triggering query
-- Minimum 2 characters to trigger name search
-
-### 4. Implement result-table component
-
-- Responsive HTML table
-- Columns: STT, Số báo danh, Họ tên, Ngày sinh, Toán, Ngữ văn, Vật lí, Hóa học, Sinh học, KHTN, Lịch sử, Địa lí, GDCD, KHXH, Tiếng Anh
-- Show null scores as "-" (not 0)
-- Highlight search match in name column (bold)
-- "Không tìm thấy kết quả" message when empty
-
-### 5. Implement app.tsx layout
-
-```
-<header>
-  <h1>Tra cứu điểm thi THPT QG 2017</h1>
-</header>
-<main>
-  {loading ? <ProgressBar /> : <SearchBar />}
-  <ResultTable />
-</main>
-<footer>
-  Dữ liệu sưu tầm, chỉ mang tính tham khảo
-</footer>
-```
-
-### 6. Styling (index.css)
-
-- Clean, minimal CSS — no framework needed for this scope
-- CSS variables for colors
-- Mobile-first responsive: table scrolls horizontally on small screens
-- Loading bar animation
-
-### 7. SQL query construction
-
-```sql
--- By exam ID (exact)
-SELECT * FROM student WHERE so_bao_danh = ?
-
--- By name (partial, accent-insensitive not feasible in SQLite, use LIKE)
-SELECT * FROM student WHERE ho_ten LIKE ? LIMIT 50
--- param: `%${query}%`
-```
-
-Note: SQLite LIKE is case-insensitive for ASCII only. Vietnamese diacritics mean users must type exact accents. This is acceptable — Vietnamese users expect this.
-
-## Todo List
-
-- [ ] Create Student TypeScript interface
-- [ ] Implement use-sqlite hook with progress tracking
-- [ ] Implement search-bar with auto-detect mode + debounce
-- [ ] Implement result-table with all score columns
-- [ ] Implement app.tsx layout with loading state
-- [ ] Add minimal CSS styling (mobile-responsive)
-- [ ] Test with actual .db file from Phase 02
-- [ ] Verify search by soBaoDanh returns exact match
-- [ ] Verify search by hoTen returns partial matches (limit 50)
-
-## Success Criteria
-
-1. `npm run dev` shows app with loading indicator while .db fetches
-2. Search by exact soBaoDanh returns single student with correct scores
-3. Search by partial name returns up to 50 matching students
-4. No results shows Vietnamese "không tìm thấy" message
-5. Table is readable on mobile (horizontal scroll)
-6. No console errors in Chrome/Firefox
-
-## Risk Assessment
-
-| Risk | Likelihood | Impact | Mitigation |
-|------|-----------|--------|------------|
-| .db fetch takes too long (50MB+) | High | Medium | Show progress bar; consider gzip in Phase 04 |
-| sql.js WASM CDN unreachable | Low | High | Bundle WASM in public/ as fallback |
-| Vietnamese text search limitations | Medium | Low | Document that exact diacritics required; acceptable UX |
-| 800K row render if query too broad | Medium | Medium | LIMIT 50 on all queries; show "refine search" message |
-
-## Security Considerations
-- All data is public exam scores — no auth needed
-- sql.js runs client-side only — no server attack surface
-- No user data collected or stored
-
-## Next Steps
-- Phase 04 adds build optimization, gzip, and deploy configuration
diff --git a/2017/plans/260412-2303-static-score-lookup/phase-04-integration-deploy.md b/2017/plans/260412-2303-static-score-lookup/phase-04-integration-deploy.md
deleted file mode 100644
index 27e2bab..0000000
--- a/2017/plans/260412-2303-static-score-lookup/phase-04-integration-deploy.md
+++ /dev/null
@@ -1,222 +0,0 @@
-# Phase 04 — Integration + Deploy Config
-
-## Context Links
-- [plan.md](./plan.md)
-- [phase-02](./phase-02-excel-parser-db-builder.md) — produces .db file
-- [phase-03](./phase-03-react-static-site.md) — produces React app
-
-## Overview
-- **Priority**: P2
-- **Status**: Pending
-- **Description**: Wire everything together, optimize .db delivery, configure GitHub Pages deploy
-
-## Key Insights
-
-- .db file is ~50-60MB — gzip reduces SQLite files by ~70% typically → ~15-20MB
-- GitHub Pages has 100MB file size limit; raw .db is borderline, gzipped is safe
-- Vite build output goes to `web/dist/` — GitHub Actions deploys this folder
-- No server needed — fully static
-
-## Requirements
-
-### Functional
-- `npm run build:db` → `npm run build` produces deployable `web/dist/`
-- .db file served with gzip (either pre-compressed or server-side)
-- GitHub Actions workflow for automated deploy on push to main
-
-### Non-Functional
-- Total deploy artifact < 25MB (gzipped .db + app bundle)
-- Build completes in CI in < 5 minutes
-
-## Architecture
-
-```
-GitHub Push (main)
-  → Actions workflow
-    → npm ci
-    → node scripts/build-database.js  (produces web/public/thptqg2017.db)
-    → npm run build                    (Vite builds web/dist/)
-    → deploy web/dist/ to GitHub Pages
-```
-
-### .db Delivery Optimization
-
-**Option A — Pre-gzip (recommended):**
-- Build script outputs `thptqg2017.db`
-- Post-build step: `gzip -k web/public/thptqg2017.db` → produces `.db.gz`
-- Vite copies `.db.gz` to `dist/`
-- Client fetches `.db.gz`, decompresses with `DecompressionStream` or pako
-- Pros: works on any static host, no server config needed
-
-**Option B — Rely on server gzip:**
-- Serve raw .db, let CDN/server gzip on-the-fly
-- Pros: simpler client code
-- Cons: GitHub Pages may not gzip .db extension; unreliable
-
-**Decision: Option A** — pre-gzip with client-side decompress. Use browser-native `DecompressionStream` (supported in all modern browsers).
-
-## Related Code Files
-
-**Create:**
-- `.github/workflows/deploy.yml` — GitHub Actions workflow (~40 lines)
-
-**Modify:**
-- `scripts/build-database.js` — add gzip step after DB creation
-- `web/src/hooks/use-sqlite.ts` — fetch .db.gz, decompress, then init sql.js
-- `package.json` — add `build:all` script combining db + vite build
-- `.gitignore` — ensure web/dist/ and .db files are excluded
-
-## Implementation Steps
-
-### 1. Add gzip post-processing to build script
-
-```javascript
-// At end of scripts/build-database.js
-const zlib = require('zlib');
-const dbBuffer = fs.readFileSync('web/public/thptqg2017.db');
-const gzipped = zlib.gzipSync(dbBuffer);
-fs.writeFileSync('web/public/thptqg2017.db.gz', gzipped);
-console.log(`Compressed: ${(gzipped.length / 1024 / 1024).toFixed(1)}MB`);
-```
-
-### 2. Update use-sqlite hook for gzip fetch
-
-```typescript
-// Fetch .db.gz instead of .db
-const response = await fetch('/thptqg2017.db.gz');
-const reader = response.body.getReader();
-// ... read chunks with progress ...
-const compressed = new Uint8Array(allChunks);
-
-// Decompress using DecompressionStream
-const ds = new DecompressionStream('gzip');
-const writer = ds.writable.getWriter();
-writer.write(compressed);
-writer.close();
-const decompressed = await new Response(ds.readable).arrayBuffer();
-
-// Init sql.js with decompressed buffer
-const db = new SQL.Database(new Uint8Array(decompressed));
-```
-
-### 3. Add combined build script
-
-```json
-{
-  "scripts": {
-    "build:db": "node scripts/build-database.js",
-    "build:web": "vite build --config web/vite.config.ts",
-    "build:all": "npm run build:db && npm run build:web",
-    "dev": "vite --config web/vite.config.ts",
-    "preview": "vite preview --config web/vite.config.ts"
-  }
-}
-```
-
-### 4. Create GitHub Actions deploy workflow
-
-```yaml
-# .github/workflows/deploy.yml
-name: Deploy to GitHub Pages
-
-on:
-  push:
-    branches: [main]
-  workflow_dispatch:
-
-permissions:
-  contents: read
-  pages: write
-  id-token: write
-
-jobs:
-  build-and-deploy:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          lfs: true   # in case Excel files are in LFS
-      - uses: actions/setup-node@v4
-        with:
-          node-version: 20
-          cache: npm
-      - run: npm ci
-      - run: npm run build:all
-      - uses: actions/upload-pages-artifact@v3
-        with:
-          path: web/dist
-      - uses: actions/deploy-pages@v4
-```
-
-### 5. Configure Vite base path for GitHub Pages
-
-```typescript
-// web/vite.config.ts
-export default defineConfig({
-  base: '/thptqg2017/',  // matches GitHub repo name
-  // ...
-});
-```
-
-### 6. Update .gitignore
-
-```
-node_modules/
-web/dist/
-web/public/thptqg2017.db
-web/public/thptqg2017.db.gz
-```
-
-### 7. End-to-end verification
-
-1. Run `npm run build:all`
-2. Run `npm run preview`
-3. Open in browser, search for known student
-4. Verify scores display correctly
-5. Check network tab: .db.gz transfer size < 25MB
-
-## Todo List
-
-- [ ] Add gzip step to build-database.js
-- [ ] Update use-sqlite hook for .db.gz fetch + decompress
-- [ ] Add build:all npm script
-- [ ] Configure Vite base path for GitHub Pages
-- [ ] Create .github/workflows/deploy.yml
-- [ ] Update .gitignore
-- [ ] End-to-end test: build:all → preview → search
-- [ ] Verify GitHub Actions workflow passes
-
-## Success Criteria
-
-1. `npm run build:all` produces `web/dist/` with all assets + .db.gz
-2. .db.gz file < 25MB
-3. `npm run preview` serves working site from dist/
-4. GitHub Actions workflow deploys successfully
-5. Live site at `https://{username}.github.io/thptqg2017/` is functional
-
-## Risk Assessment
-
-| Risk | Likelihood | Impact | Mitigation |
-|------|-----------|--------|------------|
-| Excel files not in git (too large for checkout) | Medium | High | Check if LFS needed; or commit .db.gz directly and skip build:db in CI |
-| DecompressionStream not supported in old browsers | Low | Low | 95%+ browser support; show "update browser" message for others |
-| GitHub Pages deploy fails on large artifact | Low | Medium | Gzipped artifact should be well under limits |
-| CI build timeout (Excel parsing slow) | Low | Low | 119 files in < 60s locally; CI has 6h limit |
-
-## Backwards Compatibility
-
-- Existing `database.sqlite` (97MB) preserved — not deleted or modified
-- Java source code preserved — can be removed in separate cleanup PR
-- New site deployed to GitHub Pages — no impact on existing setup
-
-## Rollback Plan
-
-- Revert the deploy.yml workflow → Pages stops updating
-- Previous Pages deployment (if any) is preserved in GitHub Pages history
-- Local: `git revert` the integration commit; Phases 1-3 artifacts still work standalone
-
-## Next Steps (Future, out of scope)
-
-- Remove Java source code and Gradle files (cleanup PR)
-- Add statistics page (average scores per province)
-- PWA support for offline access
diff --git a/2017/plans/260412-2303-static-score-lookup/plan.md b/2017/plans/260412-2303-static-score-lookup/plan.md
deleted file mode 100644
index b8222b5..0000000
--- a/2017/plans/260412-2303-static-score-lookup/plan.md
+++ /dev/null
@@ -1,105 +0,0 @@
----
-title: "Static Score Lookup - Excel to SQLite + Vite React Site"
-description: "Replace Java/Hibernate pipeline with Node.js parser + client-side SQLite lookup site"
-status: pending
-priority: P1
-effort: 6h
-branch: main
-tags: [node, vite, react, sql.js, sqlite, migration]
-created: 2026-04-12
----
-
-# Static Score Lookup
-
-## Goal
-
-Replace the Java-based Excel-to-SQLite pipeline with a Node.js script, and build a static Vite+React site that loads the .db file client-side via sql.js for readonly score queries.
-
-## Architecture Overview
-
-```
-[~119 .xlsx files] --> [Node.js parser script] --> [thptqg2017.db]
-                                                        |
-                                          [Vite build copies to public/]
-                                                        |
-                                          [React app loads via sql.js]
-                                                        |
-                                          [User searches by name/ID]
-```
-
-### Data Flow
-
-1. **Parse**: Node script reads all .xlsx from `src/main/resources/raw/` and `raw/(update)/`
-2. **Transform**: Extract 4 columns per row; regex-parse scores from column 4 text
-3. **Load**: Insert rows into SQLite via better-sqlite3 (Node-native, fast)
-4. **Serve**: Vite copies .db to `public/`; React app fetches + initializes sql.js
-5. **Query**: User input -> SQL WHERE -> render results table
-
-### Database Schema
-
-```sql
-CREATE TABLE student (
-  so_bao_danh TEXT PRIMARY KEY,
-  ho_ten      TEXT NOT NULL,
-  ngay_sinh   TEXT,          -- stored as dd/MM/yyyy string
-  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
-);
-
-CREATE INDEX idx_ho_ten ON student(ho_ten);
-```
-
-**Change from Java version**: `ngay_sinh` stored as TEXT (not DATE) — simpler, no timezone issues, display-only field.
-
-## Phases
-
-| # | Phase | Status | Effort | Details |
-|---|-------|--------|--------|---------|
-| 1 | Project scaffolding | Pending | 30m | [phase-01](./phase-01-project-scaffolding.md) |
-| 2 | Excel parser + DB builder | Pending | 2h | [phase-02](./phase-02-excel-parser-db-builder.md) |
-| 3 | React static site | Pending | 2.5h | [phase-03](./phase-03-react-static-site.md) |
-| 4 | Integration + deploy config | Pending | 1h | [phase-04](./phase-04-integration-deploy.md) |
-
-## Dependencies
-
-```
-Phase 1 --> Phase 2 --> Phase 3 --> Phase 4
-                  |                    ^
-                  +--(produces .db)----+
-```
-
-## Risk Assessment
-
-| Risk | Likelihood | Impact | Mitigation |
-|------|-----------|--------|------------|
-| .db file too large for browser fetch (~97MB) | High | High | Compress with gzip; sql.js supports ArrayBuffer; lazy-load on first query |
-| Excel format variations (missing headers, bad cells) | Medium | Medium | try/catch per row (same as Java); log skipped rows; validate after build |
-| sql.js WASM loading fails on some browsers | Low | Medium | Fallback error message; test Chrome/Firefox/Safari |
-| Duplicate soBaoDanh across files (raw + update) | Medium | Low | Use INSERT OR REPLACE (update folder overwrites raw) |
-
-## Backwards Compatibility
-
-- Existing `database.sqlite` (97MB) preserved until new pipeline verified
-- Java source code left intact — can be removed in future cleanup
-- New .db file placed at `public/thptqg2017.db` (different name/location)
-
-## Rollback Plan
-
-- Phase 1-2: Delete `scripts/` and `package.json`; Java pipeline still works
-- Phase 3-4: Delete `web/` folder; .db file from phase 2 still standalone-usable
-
-## Success Criteria
-
-1. `node scripts/build-database.js` produces valid .db with ~800K+ rows (matching Java output count)
-2. `npm run dev` serves site; search by soBaoDanh returns correct student
-3. `npm run build` produces static dist/ deployable to any static host
-4. .db file size < 60MB (better-sqlite3 is more compact than Hibernate output)