As of October 2026｜Remembering order and remembering positions: two score scales are not interchangeable
0
October 17, 2026
Daniel LuFull-Stack Engineer | Content Creator
Scoring fields, repeated submissions and extreme values in sequence and covered-number memory tasks.
Categories：Data & Insights
As of October 2026
The stored-score medians are five for sequence memory and nine for covered-number memory. The larger number does not establish greater memory capacity or an easier task: these are different scoring systems and overlapping but unequal participant groups.
Two separate distributions
This article uses a fixed snapshot read on 17 October 2026 at 03:54:35 Beijing time. Only human_benchmark_records is counted: 248024 records, dated from 23 December 2025 to 17 October 2026, with the latest at 01:23:55 that day. The legacy human_scores table's 6507 rows are not added. These are site submissions, not a random population sample.
| Task | Records | Median score | 25th–75th percentile | 90th percentile |
|---|---|---|---|---|
| Sequence memory | 7,450 | 5 | 3—8 | 10 |
| Covered-number memory | 4,310 | 9 | 7—10 | 11 |
What does the score store?
Among sequence records, 7198 have levelsPassed and all those scores equal that field. The other 252 records lack it; they remain in the overall distribution. All 4310 covered-number records have levelsCompleted, and 4273 satisfy score = 4 + levelsCompleted; 37 do not. This reveals different scales, not the exact behaviour of every historical version.
One first record per label
Records are grouped by account ID when present, otherwise by guest name, with the two namespaces kept separate. A label is not a verified person: people may change labels and labels may be shared. Counts from different tools overlap and cannot be added as a total population.
| Task | Labels | Median first score | Median best score |
|---|---|---|---|
| Sequence memory | 3,475 | 5 | 7 |
| Covered-number memory | 2,271 | 8 | 9 |
Taking one first record per label changes covered-number memory's median from nine to eight. First means first visible in this snapshot. All submissions and equal label weighting answer different questions; neither establishes which memory task is superior.
First-to-last changes among repeaters
| Task | Repeated labels | Higher last score | Unchanged | Lower last score |
|---|---|---|---|---|
| Sequence memory | 1,564 | 1,173 | 140 | 251 |
| Covered-number memory | 834 | 605 | 114 | 115 |
Among labels with at least two records, the last score is higher for 75.00% of sequence labels and 72.54% of covered-number labels. Median first-to-last changes are +3 and +2. Lower last scores also occur. There is no random assignment, fixed training duration or control for strategy, device and state. People who tried only once are absent from this comparison, so these rates do not establish causal training effects.
Treat the tail as something to check
The highest stored scores are 66 and 27. Scores above 20 occur in 17 and 11 records respectively. They are retained, rather than deleted for looking high or accepted as verified records. Interpreting them needs version, scoring and record-quality checks.
Follow change within one task
Within one task, distinguish first, latest and best while keeping conditions consistent. Do not subtract scores across tasks or turn the distributions into IQ or medical conclusions. The 90th percentile is a position among these records, not a global human limit.
🔗 Sequence memory
Repeat the flashing sequence and see how many steps you recall.
→
🔗 Covered-number memory
Click the remembered number positions after they are covered.
→
Overall distributions retain all stored scores. First and last use timestamp then record ID; best uses the highest stored score per label. No raw record labels are shown.