33 memories went in. 33 came back. The harder part was proving they still answered the same questions.

We ran a real GitHub issue + 31 comments through Memanto's OKF migration path. Result: 0 skipped, 0 removed, 0 changed—and 5/5 golden-question recall before and after.

74-second proof: https://youtu.be/E_r7tzmHtq0 Code, tests and one-command reproduction: https://github.com/moorcheh-ai/memanto/pull/1813

If you build agent-memory systems, what evidence would you require before deleting the source?

AI-assisted editing and synthetic narration; every result shown is from the reproducible command.

QuizDash is a full-stack trivia quiz app I built end-to-end with Flask and SQLite, from database design through deployment. Players select a category and difficulty level, then race a countdown timer that scales with difficulty — harder questions are worth more points but allow less time to answer. I designed the SQLite schema for questions and scores, built out the backend routes for fetching randomized questions, submitting scores, and aggregating leaderboard rankings via SQL, and implemented session-based authentication for a password-protected admin panel where questions can be added or deleted without touching the database directly.

On the frontend, I used vanilla JavaScript and Jinja2 templates (no framework) to handle the quiz flow, live countdown bar, and a post-quiz review screen showing right vs. wrong answers per question. The layout is fully responsive, adapting from a stacked single-column view on mobile to a side-by-side quiz-and-leaderboard layout on desktop. The app is deployed live on Render with secrets (session key, admin password) managed through environment variables rather than hardcoded in the codebase.