[DSRP Evidence](https://dsrpevidence.org/)

# Measuring Context as Structural Organization

Cabrera D, Cabrera L, Journal of Systems Thinking 6(1), 2026

## Abstract

A test of whether context can be measured as structural organization rather than as surrounding text. 600 items from two standard language benchmarks were coded on five features under a protocol fixed in advance, with scores computed arithmetically from the features so no judgement entered at the scoring stage.

## What they found (results)

Correlation of 0.347 (p < 0.001) between structural context and interpretation accuracy. Accuracy rose from 4.7% in the lowest-context band to 64.0% in the mid-to-high band, a difference of nearly fourteen times. Information quantity alone predicted nothing significant.

## Response

Evidence against the additive form from an experiment entirely separate from the machine-learning one, which matters because the cognitive and reality laws otherwise rest on a single study. It also yields a by-product useful to researchers who reject DSRP outright: low structural context means an underconstrained interpretation space, so the score doubles as a measure of benchmark ambiguity. Limits: AI raters with agreement from κ 0.42 to 0.83, and the two datasets occupy different context regions so the bands partly confound with source.

[Read the paper](https://doi.org/10.54120/jost.v25i1.dc6mcnl6)