Do you have a
question spaced repetition or SuperMemo Algorithm? See: SuperMemo Algorithm |

Last modified: |

**Huge interval discrepancies between Algorithm SM-15 and SM-17**

**Question**

How can SM16 and SM17 differ by 800% in their estimation of the next optimum interval. I have made a repetition (cloze) and the discrepancy now stands at 7 vs. 61 days.

**Answer**

SuperMemo 16 assumes the current interval is optimum even if you set it by hand. SuperMemo 17 will look at the estimated stability, which depends on the entire repetition history. If you set the interval to 7 and stability is 61, the next optimum interval will be:

- SuperMemo 16: 7 days exactly (repeating the item on the same day is assumed to have no impact on stability and the interval due to the spacing effect)
- SuperMemo 17: stability will be computed from the SInc[] matrix and may change even if the repetition takes place on the same day. In theory, it should not be far from 61 days, but it all depends on the statistics of past performance of similar items in similar contexts

800% difference is not the worse you can see. If you take an item with stability 10,000 and give it an interval of 1 day (e.g. because of its importance or fear of forgetting), the difference in optimum next interval will be 1,000,000%. Once you decide on a new interval, SuperMemo 16 takes your word for it, SuperMemo 17 does not.

If you tend to pay attention to intervals, in SuperMemo 17 you may need to hold your nose and accept the new reality. If you care about high speed, high retention and efficiency, you may need to accept that 10% forgetting is still painful and holding it off manually contradicts the principles of efficient learning.