Validated psychometric assessments should produce distributions that look like real human variability, clustered around means, normally distributed across populations, with opposing traits anti-correlating where the design says they should. Here's what our production dataset shows.
The distributions below come from our v1 item set, the 800+ assessments collected before this year's rebuild. The rebuilt 2026 instrument is now live with question-level response logging, and we'll publish its distributions here as enough assessments accumulate.
Discrimination, not ceiling-stacking
Alongside the 10 personality dimensions, the assessment measures 10 soft skills using scenario-based items. The figures below reflect 300 assessments from the v1 item set (before this year's rebuild). In practice respondents score in roughly a 6-25 range, with a practical midpoint near 15.5. A well-designed self-report soft-skill scale should show real variance and means that sit close to the practical midpoint, not piled at the ceiling, the classic failure mode of self-report.
| Soft skill | Mean | Std. dev. | Range |
|---|---|---|---|
| Adaptability | 16.32 | 2.90 | 7–25 |
| Conflict resolution | 18.43 | 3.29 | 7–25 |
| Critical thinking | 17.06 | 2.86 | 9–24 |
| Empathy | 16.74 | 3.04 | 6–23 |
| Leadership | 16.81 | 3.37 | 8–25 |
| Self-awareness | 16.48 | 3.34 | 7–23 |
| Teamwork | 16.27 | 3.74 | 6–24 |
| Time management | 16.70 | 3.57 | 6–25 |
| Verbal communication | 16.31 | 3.30 | 8–23 |
| Written communication | 16.68 | 3.34 | 7–25 |
n = 300 assessments, v1 item set. Means rounded to 2 decimals.
Three things stand out. First, standard deviations around 3.0-3.7 confirm the instrument is discriminating, respondents are not clustering at any single value. Second, means sit at 16.27-18.43, an average of just ~0.4 standard deviations above the practical midpoint, a moderate lift, well within published norms for self-report assessment (cf. Donovan, Dwight & Hurtz, 2003, which found ~0.3-0.4 SD shifts in high-stakes contexts like job applications, and Tellstone is a lower-stakes self-administered context where smaller shifts are expected). Some of the observed lift also reflects genuine selection: our respondents skew toward founders, operators, and executives who plausibly score above population midpoint on leadership, critical thinking, and adaptability in reality. Third, the soft skills most visible to outside observers, teamwork (16.27) and verbal communication (16.31), score slightly lower than internally-judged skills like conflict resolution (18.43) and critical thinking (17.06). That's a small but real signal that respondents calibrate differently on externally-evaluated dimensions, which is what well-designed soft-skill scales should produce.
What we expect the v2 data to show
Good science commits to its predictions before the data arrives. The rebuilt v2 items will produce different distributions than the v1 tables above, so rather than explain that shift after the fact, here's what we expect, registered now (2026), with the detailed per-option version logged internally:
- Means drift toward the midpoint (~15.5). The v2 items use more honest low options and flaw-admitting top options, so the ceiling lean should ease, and conflict resolution, our highest v1 mean at 18.43, should drop the most.
- Spreads hold or widen. As more respondents pick the genuine low options, standard deviations should stay near their v1 range (≈3.0-3.7) or increase, not collapse.
- Rewritten options redistribute as designed. The floors we made less flattering should draw the people who genuinely sit there, rather than whoever found the old wording appealing.
If the data lands where we predicted, the v1→v2 shift reads as a confirmed hypothesis, not instability. If it doesn't, we'll say so and adjust, which is the whole point of committing first.