Caveats of card NOT being atomic?

Hello everyone,I sometimes make cards that are NOT atomic. They sometimes contain multiple cloze deletions (simultaneous) within a single cloze card and test them simultaneously at once
and also in basic front back cards, I put like multiple points at back which means even if i fail one of 4/5 points, I click again.
Example:
Front: Mention Thyroid hormone Functions
Back:

  • Helps in maintaining BMR
  • Helps in production of RBC
  • Helps in digestion of carbs, fats and proteins
  • Helps in water and electrolyte balance

As everyone points out these do not align well with 20 rules by piotr wozniak.
But i wanted a holistic perspective such that i can connect with information too. is this method detrimental for long term??

1 Like

I use both atomic cards and big cards. So for your example, I’d have 5 cards: one for each point, and one for the whole list.

The “big cards” are in their own deck with a 70% and I rate them leniently (so getting only 3 of the points right would be a hard for the big card).

I haven’t used this “long term”, it’s a recent-ish strategy, but I like it so far.

[Also, you should change the category to “learning effectively”.]

3 Likes

But, lets use the example i sent, functions of thyroid hormones. How would i “separate” them?
Wouldn’t be like Front: “Mention the first function of thyroid” …Back: “Regulation of BMR”
Or what else?

I make a chain out of the cards, kinda like “cloze one by one”. The first point is special, but after that I just show a point and ask for the next.

1 Like

Sorry for spelling errors, I’m writing this on the edge of a waterfall in Yosemite. Views are great here BTW.

Your instinct that holistic cards have value isn’t wrong, but the way you’re describing them creates a real scheduling problem if you don’t isolate them. Here’s the mechanism:

FSRS optimizes its 19 weights against your actual review history within a given preset. When a multi-point basic card or a multi-cloze card sits in the same preset as atomic cards, your “Again” rate is structurally inflated for the non-atomic notes - you only need to miss 1 of 5 points to fail the whole card, while an atomic card requires missing the only atom.

FSRS can’t see that distinction; it just sees more lapses.

The optimizer then lowers stability gains and shortens intervals to compensate, pulling your atomic cards toward “shorter-than-necessary” intervals while still under-fitting the non-atomic ones. You end up with a compromise that schedules neither type well.

The fix is structural, not behavioral.

Anki assigns deck presets per deck, and FSRS optimizes weights per preset.

Here’s a real example that I use. I use mainly 3 different note types: 1. Cloze Deletion 2. Basic and 3. Multiple Choice Questions.

I’ll end up publishing my note type templates in the future once I can’t find any other way to improve on them.

So organize like this:

Parent Deck
├── Cloze   → FSRS Preset: Cloze
├── Basic   → FSRS Preset: Basic
└── MCQ     → FSRS Preset: MCQ

Now FSRS trains on homogeneous review data inside each preset.

Your multi-point basic notes get weights calibrated to their elevated lapse profile; your atomic clozes get weights that reflect their actual difficulty. This is what I do — MCQ, Cloze, and Basic each have their own preset — and the resulting intervals look noticeably different across note types.

That difference is the feature, not a bug. People sometimes see “Cloze interval = 30 days, MCQ interval = 60 days” and assume something’s broken. It’s not. The note types are doing genuinely different cognitive work with genuinely different forgetting curves; correctly tuned FSRS should produce different schedules.

Two honest caveats:

The atomic-card critique isn’t only about FSRS. When you fail 1 of 5 points and re-review all 5, you’re spending time on the 4 you knew and not isolating the weak point. You can mitigate that with image occlusion or sub-clozes where possible, but the time cost is real.

Also, you will only be able to optimize once you have enough reviews for that preset. So if I were you, I’d do that first, if you don’t have enough reviews, just use the optimized FSRS preset you already have (copy the preset) and this will over time (as you recalibrate FSRS, using the (optimize all presets button)), better adjust to your needs.

The added complexity is worth it, at least imo, especially as your note type variety grows. The alternative is letting one note type’s noise pollute scheduling for everything else.

Hope that helps, best.

2 Likes

Gonna drop my 2 cents here, this is how I would make a card for those factoids:

The {{c1::Thyroid}} {{c2::hormone}} helps in

<ol>
  <li>{{c3::maintaining}} {{c4::BMR}}</li>
  <li>{{c5::production}} of {{c6::RBC}}</li>
  <li>{{c7::digestion}} of {{c8::carbs}}, {{c8::fats}} and {{c8::proteins}}</li>
  <li>{{c9::water}} and {{c9::electrolyte}} {{c10::balance}}</li>
</ol>

I used to do Front and Back cards way at the beginning (around 4/5 years ago), then I found that remembering lists of things is not something that was that useful to me, and switched to clozes where each list item would be a cloze. Then I found out that leaving blanks that big left too much space for a correct answer, and that it was becoming more of a matter of “remembering the order and which facts I included in my list” more than “remembering the fact”. That’s why I started breaking them down further within the list items, as in the example above. A nice bonus of this is that I can also use the same cloze number for the same word in multiple lines, so that I have more “context” to narrow down, exactly, what I should complete the blanks with.

1 Like

This sounds very reasonable and makes me again wonder why we have those fsrs settings per deck not per note type?

2 Likes

Im very hesitant to use clozes, because then I feel like I do image recognition rather understand the text. How easy is it for you to recall this info later on, without anki in front of you?

If a definition, city name, and so on are hidden, how can you recognize anything from the surrounding text? You’re forming connections, which is very useful. When we read, we also guess what will happen next, and practicing this skill in advance makes it easier for us to process information (it all works for auditory processing). But if we hide part of the image, we’re essentially also forming connections. It’s like a device: just looking at it from the outside makes it very difficult to disassemble it and understand all the connections inside. But once you figure out what’s inside, you’ll never forget the external appearance, because you’re held by all the connections (all the information) about the internal structure. I call this information redundancy. That is, to remember one concept, learn four or five connections between it and what you already know, and then you won’t forget it.

And why is Anki necessary for hiding text? You make a very narrow text, three words or seven, depending on your vision, and read it one word at a time. Okay, so you don’t have a computer, so hide it with a ruler and go. This whole thing was invented long ago, and some polyglots even now learn languages ​​this way: they pick up a book in a foreign language and underline the words they don’t know yet. Then they quickly skim through it again (that’s what spaced repetition is). Anki is good, but it’s just a tool, and everyone uses it as they can and if they want.

You can make 4 cards instead of 1.

Card 1: Thyroid hormone regulates which energy expenditure? ; Basal metabolic rate (BMR)
Card 2: Thyroid hormone stimulates the production of which blood cell? ; Red blood cells
Card 3: Thyroid hormone assists in the metabolism of which macronutrients? ; Carbohydrates, fats, and proteins
Card 4: Thyroid hormone regulates which bodily balance besides energy? ; Water and electrolyte balance

3 Likes

You might mean “pattern recognition”, which means you won’t have to read the (entire) question but can still answer in anki what word was occluded. Of course, real life performance without that exact structure will be poor in that case.

You could learn to make your cards more uniform in structure, preventing pattern recognition. Or you could use Basic cards. They, too, might cause pattern recognition though, though in my experience basic cards are way better than cloze cards.


Edit: This is a great example of good basic cards that are more uniform in structure:

2 Likes

Because you can easily make separate decks for each note type.

On the other hand, imagine that you’re a “simple user” who likes to only use the default note types. If you make the FSRS options per note type then you’d have to duplicate the basic note type multiple times just for the sake of separating some of your decks.

Yes. Pattern recognition. By saying image I just wanted to emphasise that it has nothing to do with the meaning of the text (unlike other patterns which we have in language).

And yes, im tending to avoid clozes in favour of basics cards. Thats why Im asking for second opinion. Many people use clozes all the time.

I can give you a brief overview on how I do it.

Personally, I only use two note types: a heavily modified Basic note type and image occlusion (from the image occlusion enhanced addon).

I use the following principles

  1. Understand card content first, where possible[1]
  2. Active Recall
  3. Spaced Repitition
  4. Interleaving (all my cards are in the same deck)
  5. Atomic cards (A card asks for a single thing only. I mostly do it for motivational purposes though, I get demotivated if I have to recall a wall of stuff on the same card)

I have a streak of almost 650 days and study multiple topics (all in one deck of course): psychology, programmimg and IT, some medical stuff, language learning, some others (like NATO Alphabet). This workflow works great for me.

And I avoid cloze at all costs. I did use them in the beginning but found that I wasn’t really able to recall clozed info in the real world. But of course, I might have just used them very ineffiently. Even my basic cards were bad in the beginning compared to what I have now.


  1. if it’s e.g. the greek alphabet, then understanding it might be difficult. But you should understand e.g. things like models, theories, ect. ↩︎

1 Like

I think because sometimes you might want to use the same note in multiple decks and have different presets for each.

The example use here can be handled with subdecks. The complex cards go in a subdeck of the main deck, and the subdeck has its own preset. They still show up mixed in with other cards of the main deck when you study the main deck, but they get their subdeck preset’s FSRS settings applied to them.

I wouldn’t call this non-atomic, I’d just say your atoms are always found together as a molecule.

I think it makes sense to always study this information together as a unit. Your exam is not going to ask you “what is the third function of thyroid hormone”, it’s going to ask you to name the functions of thyroid hormone. So you should practicing the unit of information that you will actually be asked to recall.

I would also recommend against single-word answers, they give you too much of a hint. “Thyroid hormone helps in maintaining ____” gives you the information that it one of its functions is to help in maintaining something. You’re not going to get that hint on an exam, you’ll just be asked to list its functions.

So if you were going to chunk this into a cloze card, I’d make each bullet point a cloze, eg. {{c1::Helps in maintaining BMR}} rather than Helps in maintaining {{c1::BMR}}.

But even then, Anki’s sibling-burying behavior is going to keep the cards apart, and that’s probably not what you want if the times you actually need to recall this information you’ll need to recall it together.

The way you will actually be asked to recall the information comes first. Practice it in the way you will need to recall it. Keeping it optimal for spaced repetition comes second. If the two are in conflict, the way actually be asked to recall the information takes priority, even if it leads to more reviews than you need for the easier parts.

(And as others have suggested, separate deck presets can help to make sure the fact that your “molecule of atoms” cards are harder is handled appropriately by FSRS.)

1 Like

Also, you could probably shorten it to focus on the parts you actually need to remember.

Thyroid hormone affects BMR, RBC production, digestion, and water/electrolyte balance. You can drop:

  • “maintaining” BMR, the verb is irrelevant
  • the “carbs, fats, and proteins” part of the digestion point. If it was specific to one thing like fats you’d want to include that. But since it’s all three, if that’s just everything you would normally expect to be covered by digestion there’s probably no need to list them.
  • All of the “helps with”

Dropping the unnecessary parts will help keep the focus on the parts you actually need to remember.

You might also be able to do some sort of nested cloze thing to practice the information both together and separately:

Thyroid hormone affects {{c5::{{c1::BMR}}, {{c2::RBC production}}, {{c3::digestion}}, and {{c4::water/electrolyte balance}}}}.

I’ve seen that discussed before, but I haven’t done it myself so I’m not sure if it’s a base Anki feature or something people do using an addon. But it might let you have it both ways - practice the atoms and the molecule.

Yeah, that’s a very real risk, and I do sometimes run into it. Overall, the recall without anki in front of me is mostly fine for my purposes, and I feel is way better than having a back card with 4 loosely connected items. Might not be optimal, but it’s at least 70% good, which is plenty for most of my studying.

Also, think of the following: that particular list of facts, in that order, in that level of detail, is something that made almost obvious sense to you the day you created the card. However, imagine you need to take a break of a few months on that deck, and come back to it again: given the same front card, it might be the case that the number of facts that apply to it, their order of importance, and the level of details you need to remember have changed significantly in the meantime. So now the possible answers to the front have widened, and the chance that you will guess the exact one you wrote months prior gets lower.

On the other hand, breaking down those factoids (either with a cloze like I showed you, or with atomic basic notes like others mentioned in this thread), will help narrow down the number of possible answers for each card, and keep it narrow for longer (e.g. a blank before “carbs, fats, and proteins” probably has very few answers that make sense, so you will have an easier time remembering “digestion” for that).

In general, regardless of the specific format you’ll use, you’ll want to make flashcards that are as “stable” over time as possible, i.e. with as narrow an unchanging a set of possible answers as you can get it, and you want to avoid All-or-Nothing cards (where, like you mentioned, if you have 5 points and you fail 1, you fail the whole card. That might lead you to over-repeat the other 4 points unnecessarily)

2 Likes