The most remote repetition ever (120 years)
Summary
After a break from learning, user experienced alerts from SuperMemo showing very long intervals. Explanation follows. In language learning, it may happen that some items create an illusion of "permastore" as they may be well consolidated in the daily use of the language or review of similar items.
Problem
After a long break with Supermemo I get back to repetitions. However, almost every day, when I perform Detailed Repair I can see a lot of errors, especially regarding intervals beyond my expected lifetime.
SuperMemo Report Activity: Checking the integrity of the collection Collection: c:\sm19\systems\main Date: Saturday, October 19, 2024, 7:34:02 PM SuperMemo 19 (Build 19.05, Sep 19, 2024) Checking 7,637 element records (ElementInfo.dat) Turning on tasklist verification along element data integrity checks Verifying integrity of element data (7637 elements in ElementInfo.dat) ++++++++++++++++++++++ ERROR #1 ++++++++++++++++++++++ Too long interval (44302 days) before the next repetition on 2/4/2146 for Element #3257 ++++++++++++++++++++++ ERROR #2 ++++++++++++++++++++++ Wrong date of the next repetition at Element #3257 Expected: 10/29/2024 Found: 2/4/2146 ++++++++++++++++++++++ ERROR #3 ++++++++++++++++++++++ Interval beyond the expected lifetime! Interval: 123yr 11mth 25 days Days: 45286 Item #3,403: noun [ U + sing/pl verb ]. top brass ++++++++++++++++++++++ ERROR #4 ++++++++++++++++++++++ Wrong date of the next repetition at Element #3403 Expected: 10/29/2024 Found: 10/15/2148 ++++++++++++++++++++++ ERROR #5 ++++++++++++++++++++++ Too long interval (43576 days) before the next repetition on 2/9/2144 for Element #3821 ++++++++++++++++++++++ ERROR #6 ++++++++++++++++++++++ Wrong date of the next repetition at Element #3821 Expected: 10/29/2024 Found: 2/9/2144 ++++++++++++++++++++++ ERROR #7 ++++++++++++++++++++++ Item Burden has been corrected from 174.47292 to 174.77285 ++++++++++++++++++++++ ERROR #8 ++++++++++++++++++++++ Avg Item Interval has been corrected from 2631.94197 to 2609.35008 Verifying knowledge tree (7,637 elements in Contents.dat) Verifying the priority queue of 6,616 elements Verifying repetition schedule (Workload.dat and ElementInfo.dat) ++++++++++++++++++++++ ERROR #9 ++++++++++++++++++++++ Wrong item count on 10/29/2024 (Day #3653) Expected: 10 Found: 13 ++++++++++++++++++++++ ERROR #10 ++++++++++++++++++++++ Wrong item count on 2/9/2144 (Day #47219) Expected: 1 Found: 0 ++++++++++++++++++++++ ERROR #11 ++++++++++++++++++++++ Wrong item count on 2/4/2146 (Day #47945) Expected: 1 Found: 0 ++++++++++++++++++++++ ERROR #12 ++++++++++++++++++++++ Wrong item count on 10/15/2148 (Day #48929) Expected: 1 Found: 0 Verifying pending queue (Intact.dat) Verifying A-Factor, Lapses and Repetitions distributions ++++++++++++++++++++++ ERROR #13 ++++++++++++++++++++++ Item interval distribution discrepancy in category 5 Items reported: 82 Items found: 85 ++++++++++++++++++++++ ERROR #14 ++++++++++++++++++++++ Item interval distribution discrepancy in category 13 Items reported: 1612 Items found: 1609
Usually one pass is enough to fix all errors, however the next day I can see a new list of errors.
Question to the user
Repetition in the year 2146 sounds plausible to you? probably not (the most remote known repetition is recorder for 2130)
How long was the break. Please send the repetition history for the item scheduled for 2146 (Element #3257)
Any chance to send this collection?
Hypothesis
update: this hypothesis has been falsified
Try F7 and Compute. Perhaps some data got corrupted in memory optimization? Before you do it, you could backup and mail \alg17\ folder of the collection (if this was the case). If you decide to send your data, use bug2024(@)supermemo.org
Answer
Thank you for submitting the collection. All seems to be ok, except SuperMemo warns you of intervals going beyond the expected lifetime. All you need to do is to go to a troublesome element (e.g. Element #2605) and schedule it at times you feel ok about. For example, you can schedule it for when you are 60 or 80 or even 100, or just dismiss it if you conclude the element is unforgettable (none truly are).
For rescheduling you can just Reschedule (Ctrl+J) or better yet Execute repetition (Ctrl+Shift+R). You will notice that Execute is likely to reduce stability due to the spacing effect.
The real "culprit" is your excellent recall for that particular set of memory parameters (DSR) and the unexpectedly long interval of 6-7 years for stability of one year. Despite the predicted R=0.65 you score 5 (perhaps due to using the word in real life or due to excellent actual recall). This is no error. That category amassed 7000 repetition cases, and you recall around 99%.
What looked like a bug is rather a case for celebrating your healthy memory and good item formulation (or actual use of the language). You can say you now hold a record for the most remote repetition (Sep 7, 2136).
Human lifespan is fixed in SuperMemo and set to 122 years (121 is the Guiness Record of longevity).
Technical
The error in Repair collection should rather be called a "notification". However, it should rather stay as it is because it is a useful alert. We might consider a separate optional repetition alert for very long intervals. Currently, the alert is turned on only if Alg SM-18 intervals differ substantially from Alg SM-15 intervals. But even this alert is frequently silenced in options.
Followup
I have encountered top brass in just one book, but it looks to me like the opposite of memory leech. Thank you for the praise, however most of my items require reformulation, and after so many years I am starting to grasp real meaning of "20 rules".
I am afraid if this factor won't affect worse formulated items' schedules?
Effect of great items on bad items
Items are classified by difficulty, stability and retrievability. This should separate difficult items from the effects of easy items. Your "worst" items have only 5 lapses, but they all have already a great deal of repetitions and very long intervals. The one with the lowest retrievability of 37% has been in limbo for 6 years. If you happen to answer correctly, you may end up with 11-year interval. This may be an accidental effect of reviewing something similar recently but it is an unavoidable risk of spaced repetition.
To assess the risk and the effects, you can play this game:
- sort the browser by lapses
- chose items with lapses=5
- execute subset review (Ctrl+L)
This could be an interesting experiment on how your memory copes after a break, and how SuperMemo handles it. Most likely, you should get lots of lapses, but the items that "survive" will get stratospherically long intervals.
If you happen to do that, please report, e.g. browser screenshot showing dates, intervals, reps and lapses.