Journal — Method
Choosing what to measure next
An experimental planning tool compares candidate measurements by the explanations they could distinguish.
When two explanations fit the available evidence, collecting more of the same data may not resolve the question. The useful next observation is one on which the explanations make different predictions.
In summer 2026 I developed an experimental tool called Futures Sidewalk to organize that decision. It accepts evidence records, competing hypotheses, candidate measurements, and their estimated costs. It then looks for a set of measurements that distinguishes the hypotheses at low total cost.
What the output means
The tool simulates how the hypotheses would appear under the proposed measurements. Its output is a plan: which measurements to consider and which pairs of explanations each would separate under the supplied assumptions.
The quality of that recommendation depends on the candidate explanations, the measurement model, and the cost estimates. The output belongs with observation-design work. Taking the recommended measurement and interpreting its result are subsequent tasks.
What a failed check changed
The August working record describes a frozen run that exceeded a manually chosen acceptance ceiling: ten flagged items against a ceiling of nine. Rather than discard the run, the work examined the basis for the ceiling.
The record reports 200 null-model runs, used to estimate how the statistic behaved under the test's no-effect assumptions. It describes a 95th-percentile value of nine and a later run that passed. Those calculations gave the threshold a stated basis; they did not erase the original failed check.
The record also describes a second assistant reproducing the failing configuration and identifying gaps in run documentation. That account is part of the development history, rather than a new independent audit performed for this article.
Where the tool stands
The reported evaluation used synthetic cases. The next useful application would be a real ambiguity for which alternative explanations, feasible observations, and costs can be specified. Until then, the prototype provides a method to investigate, not a demonstrated improvement in field decisions.
For the map, this opens a practical possibility: a coverage gap could eventually carry a reasoned suggestion for the observation most likely to help resolve it.