How-to

Flashcards and review

Cards are generated from your nodes rather than written separately — so a card can never drift from the note it came from.

Cards come from nodes

You do not write a flashcard here. You write a node, and the card is derived from it — its title, its definition, and the links in its body. Edit the node, regenerate, and the card follows.

That is the whole design, and it has one consequence worth internalising: if a card is wrong, the node is wrong. Fix it there. A card you could edit independently would be a second copy of a fact, and a second copy is a fact that will eventually disagree with the first.

Four kinds exist. How many a given node produces depends on what is in its body:

Type	What it asks
Recall	The title on the front, the definition on the back.
Reverse	The definition first — name the thing.
Cloze	A term blanked out of a sentence from the body, shown with the sentence it was cut from. Only from prose — see below.
Relational	Built from a typed link: given this node and this relation, what is at the other end? These test structure rather than vocabulary.

A node with a dozen links does not become a dozen cards. The most context-bearing ones win, so a well-linked node produces a few good cards rather than a heap.

Why a node may produce three kinds and not four

A cloze needs a sentence. If a node’s links sit in a list — a See also line, say — then blanking one out leaves a row of names with nothing to reason from, and every other name in the row is an equally good answer. There is no single right one, so no card is made.

A link inside ordinary prose does produce one. The test is what is left of the sentence once every link name is taken out of it: three or more words of your own and it is a cloze, punctuation and commas and it is a list. Three per node at most, either way.

The starter palace is all lists. Its nodes carry their links under See also, so they produce recall, reverse and relational cards and no clozes at all. That is the guard working rather than a fault — your own nodes will make clozes as soon as you link to something mid-sentence.

What makes a node eligible

Enough to build a card from — a title and a definition at minimum. And it must not be a draft.

A grey node produces no cards. Not fewer, not lower priority — none. Something a model wrote that you have not read cannot enter a system whose whole job is to make things stick. Open it, it settles, and then it is eligible like anything else. Some of my nodes are grey.

The bar is in the database rather than in the code that asks for cards, so it holds even if some future feature forgets to check.

Reviewing

Open Cards and start a session. You see a front, you try to answer, you reveal, and you say how it went:

Grade	Key	Means
Again	1	You did not have it. It comes back soon.
Hard	2	You got there, slowly. A shorter next interval.
Good	3	You knew it. The normal path.
Easy	4	Instant. A longer jump before you see it again.

The schedule is SM-2, the algorithm Anki popularised: each card carries an ease and an interval, and how you grade it moves both. Cards you find hard come back often; cards you know recede. Nothing here is unusual, deliberately — the interesting part of this app is where the cards come from, not the arithmetic that spaces them.

The daily cap on new cards

Twenty new cards a day. It exists because the failure mode of a system like this is not laziness — it is generating four hundred cards in an afternoon and being buried in reviews for a month.

Cards over the cap are not lost; they queue and arrive on later days. Due reviews are never capped, only new intake is.

Per-card controls

Suspend — out of rotation indefinitely, history kept.
Bury — not today; back tomorrow.
Flag — mark it to come back to.
Set due date — put it on a specific day.
Reset progress — back to new, discarding its schedule. The one here that throws work away.

Bringing an Anki deck in

You can import an .apkg or a CSV, and the point is not that this is a nicer card viewer. The rows become nodes. Your deck arrives as a graph you can link, group and explore, with the cards attached to the nodes they came from.

Your schedule comes with it. The review history you have built up is carried over rather than reset, so moving a deck does not cost you the years already spent on it. The import summary reports how many schedules carried, which is the number worth checking.

Near-duplicate rows are grouped before anything is created, so an import that would have produced two nodes for one subject tells you first.

Getting cards out

Cards and their full review state are in the account export as rows in data.json, not just as text — so the schedule survives leaving, which is the part an export usually loses. Getting your data out.

Drafts, and where a node came from — why the gate exists at all.
Typed relations — relational cards are only as good as the edges under them.