What it measures

Published prototype observations examine economic choices under fixed prompts. Risk consistency is measured separately from patience, generosity, trust, and other stated preferences.

Reading the measurement

Check the source documents, dates, accounting conventions and calculation chain. A financial answer needs evidence and consistent units. Benchmark success does not establish live investment performance or suitability for an individual decision.

Published evaluation

The publisher’s stored risk-consistency error, preserved in native error points; lower is better. These prototype files predate the canonical release protocol and have varying sample sizes and experiment coverage. They do not establish economic reasoning accuracy, a composite rationality score, or a score on Freakonomics or GameTheoryBench.

Where this fits in an evaluation suite

Synthetic research rewards do not establish live investment performance. Preserve transaction costs, held-out regimes, constraints, and the exact policy behind a reported result.

Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.

A reproducible comparison

The publisher’s stored risk-consistency error, preserved in native error points; lower is better. These prototype files predate the canonical release protocol and have varying sample sizes and experiment coverage. They do not establish economic reasoning accuracy, a composite rationality score, or a score on Freakonomics or GameTheoryBench.

Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.