Build this tool: Curiosity Title Grader - Does Your Title Actually Hook?
Product folder name: Curiosity Caption Grader
Display title: Curiosity Title Grader

PRODUCT BRIEF
Desktop app scores a title on two axes: "curiosity pull" (question words, ambiguity) vs "searchability" (keyword presence), gives a blended recommendation.

CONTEXT
Difficulty: Easy
Category: Howto & Style

TREND HOOK
Titles like "Who can relate?" and "Which was your favorite?" rely entirely on curiosity with no keywords.

VIDEO ANGLE
Short: "I graded 20 viral titles — most have ZERO searchable words." Long: build the scorer in Cursor, run it against trending list.

WHY THIS TOOL
New spin on title auditing not covered by prior "Big Number Title" tools.

STACK (standalone desktop app)
- Python 3 + tkinter GUI — standalone desktop tool (not a website).
- Even if the brief says "web app" / "site", ship a Python desktop GUI for v1.
- Include: main .py, requirements.txt, run_*.bat, build_exe.bat (PyInstaller), and an exports/ folder for saved output.
- Smoke-test core logic offline without requiring live API keys when possible.

GUI RULES
- Clean, readable dark or light UI — pick one consistent palette (no brand lock-in required).
- Clear app title / hero name derived from the product folder name.
- Short muted "How to use: …" line under the title (one sentence: what to enter → what button → what you get).
- Prefer real inputs (paste URL / fields / text) over fake demos.
- No sample dropdowns or canned demo datasets unless explicitly requested.
- Dropdowns: NEVER ttk.Combobox (unreadable on Windows dark UI). Use DarkSelect
  (preferred, no grab_set) OR dark-styled tk.OptionMenu (BG_ELEVATED / TEXT /
  active #243056 / BORDER+TEAL highlight; style the popup menu the same way).
  Never use grab_set (breaks close behavior on Windows).
- Handle window close with WM_DELETE_WINDOW.
- One job per tool; working copy/export; avoid overbuilt dashboards.

APIs & KEYS
- Never hardcode API keys or secrets into source or the packaged EXE.
- Store keys in a local user settings JSON under %LOCALAPPDATA%\<AppName>\ (or a single shared settings file if multiple tools share a key).
- Label each key field clearly in the UI (e.g. YouTube API key vs Gemini API key).
- Local/offline logic must work first; APIs are optional enhancements where possible.
- Call network APIs in a background thread; update the UI with after(...). On failure, keep local results and show a clear error.

YouTube Data API (v3)
- Add a shared YouTube Data API key field (these tools do not use OAuth).
- Use for: video/channel/playlist lookups, titles, descriptions, tags, statistics, search as needed by the brief.
- Prefer paste-a-URL / channel-ID inputs; parse video IDs from common YouTube URL formats.
- If captions/transcripts are needed, use the captions endpoint or a documented fallback (manual paste) when captions are disabled.
- Respect quotas; cache responses during a session when useful.

Gemini API (optional copy / polish)
- Add Gemini (AI-first for paste-ready copy) when the tool WRITES language creators will paste online
  (titles, captions, descriptions, pinned comments, end-cards, disclaimers, SEO templates).
- UI: API key field + Model text field + checkbox like "Use Gemini" / "Polish with Gemini".
- Default model: gemini-2.5-flash-lite. Also accept typed model ids (e.g. gemini-3.1-flash-lite).
- Do not hardcode the model URL/path; pass the Model field value into the generate call.
- Pattern: Gemini writes paste-ready copy first when enabled. Local templates/scores are the fallback if Gemini is off or fails.
- On Gemini failure: keep local output, show error, do not wipe results.
- Flash Lite is enough for short creative jobs — do not default to Pro.

Other APIs (only if the product brief needs them)
- Roblox: if the brief needs group/game data, use public Roblox APIs or documented endpoints; allow manual ID paste if API is blocked.

IMPLEMENTATION NOTES
- Match the product brief closely; keep v1 focused on one job.
- Support Copy and/or Export of results where useful (TXT/CSV/JSON as fits the tool).
- Validate inputs; show empty/error states clearly.
- If scoring/grading: show score + short explanation of what failed/passed.
- If generating copy: show multiple variants when the brief asks for them.

DEMO INPUT → OUTPUT (from live scan)
- Field Title: You won't believe what happened...
- Primary button name: Grade
- Example result lines:
  RESULT Title graded | You won't believe what happened...
  RESULT Curiosity pull | 42/100
  RESULT Searchability | 40/100
  RESULT Blended score | 41/100
  RESULT Recommendation | LEAN_CURIOSITY
  RESULT Verdict | Search ahead of curiosity - add a question, gap, or stakes phrase.

OUTPUT
- Runnable standalone Python app
- requirements.txt + run bat + optional EXE build script
- In-app how-to line
- Example of typical input → output (in README or comments, not fake UI samples)