Hard Interval Calculation in First Learning Step

Anki Suggestions
1

Current Implementation Details
In learning steps, Hard interval is calculated as:

if learning_step == 1:
    hard_interval = round((again_interval + good_interval) / 2)
else:
    hard_interval = previous_step_interval

Technical Analysis
The current system handles Hard intervals differently depending on the learning phase. In the first learning step, the interval is calculated as an average between Again and Good intervals, which forces an unnecessary coupling of parameters. For later steps, the system returns to the previous step interval, which works as intended and requires no modification.

Core Problem
The averaging calculation in the first step creates two significant issues:

  1. Creates illogical spacing for unknown cards requiring immediate review
  2. Forces coupling between first and second learning steps parameters:
    • Second parameter must be carefully chosen considering the first
    • Limits flexibility in setting optimal intervals
    • Restricts ability to set longer second steps without affecting first step timing

User Impact
This change is particularly crucial for Auto advance users who:

  1. Use only Hard/Good buttons for efficiency
  2. Rely on intuitive and objective criteria for quick decision-making
  3. Need consistent and logical review intervals

Proposed Solution
Modify only the first learning step’s Hard interval behavior:

if learning_step == 1:
    hard_interval = again_interval  # Match Again interval for first step
else:
    hard_interval = previous_step_interval  # Keep current behavior

This modification would:

  1. Create logical progression for initial learning
  2. Allow independent parameter setting
  3. Enable more flexible learning step configurations
  4. Support efficient Auto-Advance workflow
  5. Maintain consistent review patterns for unknown cards
2

Auto advance users who:

  1. Use only Hard/Good buttons for efficiency

You should be using Again/Good button for rating cards and not Hard/Good. Hard is supposed to be used as Pass and so it makes sense for it’s interval to depends upon what you have as the second step (which the Good button uses).

The solution doesn’t make sense to me. Did you write it yourself? Why should Again and Hard intervals be the same?

if learning_step == 1:
    hard_interval = again_interval  # Match Again interval for first step
else:
    hard_interval = previous_step_interval  # Keep current behavior
3

According to research, memory consists of storage strength and retrieval strength. While storage strength remains constant, retrieval strength decreases over time.

Benefits of Hard/Good usage:

  1. Partial recall is more effective for learning than complete reset
  2. Spaced practice is 74% more effective than massed practice
  3. Active recall is 51% more effective than passive review

Therefore, for users processing large volumes of cards:

  • Hard maintains storage strength while adjusting retrieval difficulty
  • Again creates inefficient massed repetition
  • Hard provides optimal spacing for partial recall

The “Forget to Learn” theory suggests that partial decay in retrieval strength can enhance long-term retention.

4

I don’t get what you’re trying to say. You have suggested making Again/Hard the same for first step. Now, you say Again interval is not good therefore make Hard the same as Again?

I will ask with all due respect, are you writing this yourself? There seems to be influx of AI-generated content to various communities.

5

Learning steps should follow user-defined parameters consistently. While every other step does this, the first step’s Hard interval creates an unexpected average, breaking this logical pattern.

I write these explanations from direct Anki experience, using AI as a supplementary tool for refinement. This reflects today’s reality where AI-human collaboration has become essential - not to replace human thinking, but to enhance our problem-solving capabilities.

6

Now, I don’t know what you want and what the AI wants. I believe that code was written by the LLM which basically makes Hard interval no different from Again. That’s less useful than what Anki does currently.

Also, if you want an explanation for the buttons’ working, see Deck Options - Anki Manual.

7

You don’t understand the simple logic and are calling this an AI problem. If Anki continues to operate like this, I will be building my own app and it will completely replace Anki within 10 years.

8

I support to optimize the learning step module. But my proposal is not more flexible learning step configurations. Instead, a ML-powered short-term memory model is the game changer.

However, short-term memory is far more complex than I initially thought.

I’ve already conducted extensive research on this topic：

I don’t think average user could do better than me. So it’s not a good solution to enable more flexible learning step configurations.

2 Likes