Corta Reports: Analysis of the Relationship Between User Age in the Corta Brain Assessment App and Their Cognitive Metrics
Internal Analytical Report — Corta, April–June 2026
1. Data and Sample Selection
| Parameter | Value |
|---|---|
| Total participants | 258 people |
| Inclusion criteria | Minimum of 3 completed sessions in the Corta brain assessment app during the study period (April–June 2026) |
| Data collection period | April–June 2026 (3 months) |
| Gender | 100% women |
| Group 1 (younger) | 129 people, age 25–40 |
| Group 2 (older) | 129 people, age 45–65 |
| Available user data | Age (in ranges) and gender (no other demographic data) |
| Cognitive metrics | Reaction speed, accuracy, working memory, attention; composite score; variability |
2. Key Findings
High cognitive status — a score in the top quartile of the full sample (>75th percentile, threshold = 86 points).
| Group | Total | High scores | Share |
|---|---|---|---|
| 25–40 years | 129 | 53 | 41% |
| 45–65 years | 129 | 71 | 55% |
In absolute terms:
- Younger group: 53 out of 129 scored ≥86
- Older group: 71 out of 129 scored ≥86
- Difference: +18 people, or +14 percentage points
3. Score Variability Within Groups
| Group | Mean composite score | Median | Standard deviation | Coefficient of variation | Minimum | Maximum |
|---|---|---|---|---|---|---|
| 25–40 years | 74.2 | 73.8 | 11.9 | 16.0% | 48 | 95 |
| 45–65 years | 76.8 | 77.5 | 18.4 | 24.0% | 41 | 98 |
Interpretation:
- The mean score in the older group is 2.6 points higher;
- The median is also higher (77.5 vs 73.8), indicating that the upward shift is not driven by outliers alone;
- Spread in the older group is substantially wider: standard deviation 18.4 vs 11.9, coefficient of variation 24% vs 16%;
- The minimum score in the older group is lower (41 vs 48), while the maximum is higher (98 vs 95).
This supports polarization in the older group: among 45–65-year-olds, we see both the weakest and the strongest performers in the entire sample.
4. Statistical Significance of Differences
To test whether these differences could be due to chance, we ran two tests:
A. Proportion comparison (chi-square test)
| Metric | Value |
|---|---|
| χ² | 5.14 |
| p-value | 0.023 |
| Difference in shares | 14% |
Conclusion: The difference in the share of high scores between groups is statistically significant at p < 0.05. The probability that this difference arose by chance is just over 2%. We can conclude that the older group objectively reaches the top 25% more often.
B. Mean comparison (t-test for independent samples)
| Metric | Value |
|---|---|
| t-statistic | 1.29 |
| p-value | 0.20 |
Conclusion: The difference in mean scores (74.2 vs 76.8) is not statistically significant at p < 0.05. This means the averages are statistically similar, although a small advantage for the older group is visible.
Why is that? Because greater variability in the older group “dilutes” the significance of mean differences. Although the share of high scores is higher, the mean does not differ — since the older group includes weaker users who pull the average down.
This is an important nuance: averages mask the real picture. Looking at means alone, we would say “there is no difference.” Looking at distribution (share of high scores + spread), we see that the older group is simply more heterogeneous.
5. Breakdown by Cognitive Metrics
| Metric | 25–40 years (mean) | 45–65 years (mean) | Difference | p-value (t-test) | Correlation with age |
|---|---|---|---|---|---|
| Reaction speed (ms) | 410 | 436 | −26 ms | 0.04 | −0.23 |
| Accuracy (%) | 86.0 | 87.5 | +1.5% | 0.31 | +0.04 |
| Working memory (correct answers) | 7.0 | 7.3 | +0.3 | 0.27 | +0.03 |
| Attention (switching, ms) | 520 | 554 | −34 ms | 0.06 | −0.19 |
| Composite score | 74.2 | 76.8 | +2.6 | 0.20 | +0.05 |
What matters here:
- Reaction speed — the only metric where the difference is statistically significant (p = 0.04) and shows a weak negative correlation with age. Older users react more slowly on average by 26 ms.
- Attention — close to significance (p = 0.06) but does not reach it. Same pattern of mild slowing.
- Accuracy and memory — differences are not statistically significant; correlation with age is effectively zero. On accuracy, older users even slightly outperform.
- Composite score — difference is not significant; no correlation with age.
Conclusion: If we assessed cognitive status by reaction speed alone, we would see “age-related decline.” But on other metrics, older users do not fall behind — they compensate for slower responses with accuracy and strategy.
6. Summary: What Correlates and What Does Not
| Relationship | Correlation coefficient (r) | Interpretation |
|---|---|---|
| Age ↔ Composite score | +0.05 | No correlation |
| Age ↔ Accuracy | +0.04 | No correlation |
| Age ↔ Working memory | +0.03 | No correlation |
| Age ↔ Reaction speed | −0.23 | Weak negative |
| Age ↔ Attention | −0.19 | Weak negative |
| Age ↔ Variability | present | 1.5× wider spread in the older group |
7. Distribution Analysis: Medians, Quartiles, and Asymmetry
To better understand how scores are distributed, we examined quartiles and distribution shape.
| Group | 25th percentile | Median | 75th percentile | Skewness coeff. |
|---|---|---|---|---|
| 25–40 years | 65.2 | 73.8 | 82.5 | −0.21 (mild left skew) |
| 45–65 years | 62.1 | 77.5 | 89.8 | −0.38 (moderate left skew) |
Interpretation:
- In the younger group, scores are tightly clustered: from the 25th to the 75th percentile — a range of 17.3 points.
- In the older group, the interquartile range is wider: 27.7 points.
- Both groups are left-skewed (a longer tail on the low end), but this is more pronounced in the older group.
- At the same time, the median in the older group is higher than in the younger group (77.5 vs 73.8), indicating that most older Corta users perform well, while weaker performers form a separate tail.
Visualization (descriptive):
- Younger group histogram — bell-shaped, compact, peak around 72–76 points.
- Older group histogram — flatter, with two “humps”: one around 74–78 points (main mass), another around 88–92 points (strong performers), and a long tail down to 40 points.
8. Analysis of Low Performers (Below the 50th Percentile)
We also looked at users who scored below the median (below the 50th percentile, threshold = 74 points).
| Group | Total | Below median | Share |
|---|---|---|---|
| 25–40 years | 129 | 64 | 50% |
| 45–65 years | 129 | 61 | 47% |
The difference is small (3 percentage points) and not statistically significant (χ² = 0.18, p = 0.67). In other words, the share of weaker Corta users is roughly the same in both groups.
But if we look at the depth of underperformance (mean score among those below the median):
| Group | Mean score among “weak” users | Gap from median |
|---|---|---|
| 25–40 years | 64.1 | −9.7 points |
| 45–65 years | 59.3 | −18.2 points |
Important difference: Weak performers in the older group are significantly weaker than weak performers in the younger group. They are not just slightly below average — they fall substantially lower (18 points below the median vs 10 points in the younger group).
This means the issue is not that there are more weak Corta users in the older group (the share is about the same), but that among older users, a subset shows a deep decline that we do not see in the younger group.
9. Cognitive Metrics Correlation Matrix
We examined how metrics relate to each other within each group to understand the structure of cognitive profiles.
| Speed | Accuracy | Memory | Attention | Composite | |
|---|---|---|---|---|---|
| Speed | 1 | −0.12 | −0.08 | +0.41 | 0.31 |
| Accuracy | −0.12 | 1 | +0.35 | −0.09 | +0.52 |
| Memory | −0.08 | +0.35 | 1 | −0.05 | +0.48 |
| Attention | +0.41 | −0.09 | −0.05 | 1 | 0.29 |
| Composite | 0.31 | +0.52 | +0.48 | 0.29 | 1 |
Correlations are reported for the full sample.
What we see:
- Composite score is most strongly linked to accuracy (r = 0.52) and memory (r = 0.48).
- Speed and attention are more strongly linked to each other (r = 0.41) than to other metrics.
- Accuracy and memory are also linked (r = 0.35).
This suggests that the app captures two relatively independent cognitive “clusters”: a speed cluster (speed + attention) and a quality cluster (accuracy + memory). Overall results are largely driven by the second cluster.
Applied to the age analysis: older users lose a little ground in the speed cluster but do not fall behind (and may even gain) in the quality cluster. That is why the composite score does not decline.
10. Extended Conclusions
Conclusion #1. There is no direct correlation between age and cognitive status.
Data from 258 active Corta users in April–June 2026 show that age is not a predictor of cognitive results. The older group (45–65 years) is not worse than the younger group (25–40 years), and on the share of high scores — it is statistically better.
Conclusion #2. The older group is polarized.
Among Corta users aged 45–65, results are less predictable. At the same time, this group contains:
- the strongest performers in the full sample (55% high scores vs 41% among younger users);
- the weakest performers (the gap from the median among “weak” users in the older group is nearly twice as large as among “weak” users in the younger group).
This is not “age-related decline.” This is “age-related divergence” — trajectories split apart.
Conclusion #3. Averages are misleading.
The difference in mean scores between groups is not statistically significant. If we relied on averages alone, we would conclude “no difference.” But distribution and share analysis show that a difference exists — it is just bidirectional: older users are simultaneously better and worse, depending on who we look at.
Conclusion #4. A compensatory effect.
Older users are slightly slower on reaction speed and attention switching but compensate with accuracy and working memory. The composite score does not decline. This suggests that cognitive aging in our sample appears not as uniform decline, but as a profile shift: less emphasis on speed, more on accuracy.
Conclusion #5. Age affects variability, not level.
If age predicts anything, it is not cognitive status itself but its spread. The older the group, the wider the spectrum of results. For the product, this means the older audience holds both the greatest upside (strong performers) and the greatest risk zone (weak performers).
11. Product Hypotheses (For Future Validation)
Based on the data, we formulated several hypotheses that can be tested with additional information:
Hypothesis 1. Strong performers in the older group are those who train regularly (session frequency above average). If we add frequency data, we will most likely find that regularity — not age — drives their high scores.
Hypothesis 2. Weak performers in the older group are recent joiners or users with a low baseline. They may catch up with sufficient regularity, but need more time.
Hypothesis 3. The older Corta user base includes two subtypes: “compensators” (slower but accurate) and “non-compensators” (slower and inaccurate). The first group benefits from untimed tasks; the second needs more frequent short sessions on foundational skills.
12. Product Recommendations
1. Move away from age-based personalization.
Age should not affect task difficulty, training frequency, or recommendations. All users should receive tasks matched to their current level, determined by the dynamics of their own results over the past 2–3 weeks.
2. Introduce profile-based segmentation, not age-based.
Instead of age groups, define cognitive profiles:
- “Speed-oriented” — strong on speed/attention, weaker on accuracy;
- “Precision-oriented” — strong on accuracy/memory, moderate speed;
- “Balanced” — all metrics at average level;
- “Weak” — all metrics below average.
Each profile gets its own training recommendations.
3. For the older group — separate variability monitoring.
Track not only mean score but also stability. If a 45+ user has high and stable results — do not reduce difficulty. If scores drop or fluctuate — offer more frequent short sessions.
4. Collect additional data.
In next phases, gather:
- exact age (or narrower bands: 45–49, 50–54, 55–59, 60–65);
- session frequency (sessions per week);
- optionally: sleep, stress, physical activity (self-reported).
This will enable multivariate analysis and a clearer picture of cognitive performance drivers.
13. Limitations of the Current Analysis
| Limitation | Impact on conclusions |
|---|---|
| No session frequency data (beyond the 3+ threshold) | Cannot assess the role of regularity in results |
| No data on education, health, sleep, or stress | Cannot explain underlying patterns |
| Women only | Results may not generalize to a male audience |
| Active users only (≥3 sessions) | We do not see users who came 1–2 times and left, who may differ systematically |
| Correlational design | We observe associations but cannot confirm causation |
14. Executive Summary
Based on data from 258 active Corta users in April–June 2026, there is no direct correlation between age and cognitive status. Users aged 45–65 are not worse than Corta users aged 25–40, and on the share of high scores they perform even better (55% vs 41%). The key difference is not in level but in variability: among older Corta users, results are less predictable — there are both the strongest and the weakest participants in the sample. Age affects not how well a person performs, but how much results can vary. For the product, this means age-based personalization does not work — focus should be on individual cognitive profiles and each user’s performance dynamics.
Report prepared based on data from April–June 2026.