901 Observations: What the DLI Dataset Shows at Scale

The Decision Load Index dataset started with 47 participants in the 7-day retest cohort. That sample was sufficient to test instrument stability and explore early patterns. It was not sufficient to say anything reliable about the shape of the distribution — how common high-load states are relative to low-load states, which occupational categories cluster at the extremes, or whether the retest findings would hold at larger N.

At 901 participants, some of those questions have clearer answers. This field note describes what the additional scale reveals. It also describes what it does not resolve.

What 47 Participants Showed

At 47 participants, the DLI distribution appeared roughly normal. No strong skew was visible. The 7-day retest finding was the most reliable result from that cohort: 81% of participants showed stable scores (within one score band) across the two administrations. The 19% who showed meaningful change shared three behavioral patterns — externalizing open loops into a trusted capture system, reducing context-switch frequency, and reducing unresolved communication threads.

The ADHD subgroup finding from the early cohort was unexpected: ADHD-identified participants showed both higher baseline scores and higher retest stability than the general population, suggesting the DLI captures something more durable than mood.

What 901 Participants Add

Distribution shape

The right-skewed distribution is the most significant observation. A normal distribution of decision load scores would suggest that high-load states are unusual — as many people in the low range as in the high range, with most clustering near the center. A right-skewed distribution suggests the opposite: extreme high-load states are disproportionately represented relative to extreme low-load states.

This is consistent with what asymmetric work environments would predict. Knowledge work creates decision load through accumulation — open loops multiply faster than they resolve when context-switching is high. Low-load states are achievable but require deliberate structural choices. High-load states emerge by default in most knowledge work environments.

Occupational clustering at the high end

The top 10% of DLI scores in the dataset cluster in three occupational categories: managers with more than five direct reports, solo founders, and doctoral students. This is consistent with earlier field notes examining each group separately. What the larger dataset adds is the relative concentration: these three categories appear disproportionately in the high-load tail compared to their representation in the overall dataset.

This is observational and self-reported. Occupational category is coarse-grained. The finding is a pattern worth tracking, not a reliable effect size.

Retest stability at scale

The 81% stability finding from 47 participants held in the extended dataset: approximately 80% of participants who completed a second assessment showed scores within one band of their initial score. This consistency across a larger and more diverse sample provides moderate confidence that the DLI is measuring something structural rather than a transient state.

If 7-day stability were primarily driven by mood, we would expect much lower consistency — mood varies substantially day-to-day and week-to-week. The stability pattern is more consistent with measurement of a behavioral or environmental characteristic that changes slowly.

The 22% improver pattern

In the early cohort, 19% of retested participants showed meaningful improvement. In the extended dataset, approximately 22% of those who retested showed score reduction that exceeded the measurement noise threshold. This is a small increase, but it is directionally consistent.

The behavioral pattern among improvers remains the same across both cohort sizes: externalization of open loops into a trusted capture system appears most frequently in participant-reported behavior changes between assessments. This is not a controlled finding — participants self-reported behavior change, and the capture system was not specified or verified. But the consistency of the pattern across sample sizes is notable.

What This Does Not Resolve

The right-skewed distribution is a descriptive finding about this dataset. It does not establish that knowledge work in general produces more high-load than low-load states. The sample is not population-representative.

What the Scale Adds to the Instrument

At 901 observations, the DLI has sufficient data to begin examining subgroup patterns that were not visible at smaller N. The occupational clustering and distribution shape are findings that emerge from scale, not from the instrument design. They inform where to look next rather than confirming mechanistic claims.

The retest stability holding across a larger and more diverse sample is the most methodologically meaningful finding. It increases confidence that the instrument is measuring a stable characteristic rather than noise.

This is a research field note based on observational, self-selected data. N=901 reflects total DLI completions; retest data is a subset. Distribution shape and occupational clustering are descriptive findings, not causal claims. This dataset has not undergone peer review. This content is educational and does not constitute professional or clinical advice.