Flashcards and Review

Edit: src/content/docs/learning-materials/flashcards-and-review.md

How Anki-style recall fits into the platform.

Flashcards are the retention layer of the platform.

They are not a replacement for lessons. They are the mechanism that makes the lesson come back later.

Card Types

The platform should support several kinds of cards.

Definition Cards

Ask for a concise definition, such as what a KV cache is or what an observation means in an agent loop.

Mechanism Cards

Ask how a process works, such as masked softmax, RoPE application, tool execution, or controller dispatch.

Comparison Cards

Ask for differences between similar ideas, such as MHA vs MQA vs GQA or action vs observation vs event.

Trace Cards

Ask the learner to walk through a sequence, such as prefill to decode or user request to agent runtime action.

What Makes a Good Card

A good card is:

atomic
answerable without rereading the article
tied to a lesson
clear about the expected answer
useful when reviewed days later

The goal is not to memorize trivia. The goal is to make important mechanisms retrievable.

Scoring

The first version can use simple scoring:

remembered
missed
skipped
hidden

From these events, the platform can compute a recall score per roadmap and per topic. A later version can add spaced repetition intervals and due dates.

Why Not Just Use Anki

External Anki decks are powerful, but they are disconnected from the lesson, roadmap, and platform progress.

ML Learning Lab keeps the cards attached to the content:

the card links back to the lesson
the lesson knows which cards it produced
the roadmap knows which topics are weak
the dashboard can choose a balanced daily review set

That integration is the product advantage.