Readable signals from the AI capital allocation benchmark

Confidence calibration helps readers judge whether model self-reported confidence carries useful information about realized benchmark performance.

Behavior profiles help readers separate model style from short-term score noise: some models seek more risk, some concentrate harder, and some change portfolios less between rounds.

Live alpha is provisional, but it shows how open model portfolios are moving before the final official score.

The consensus portfolio tests whether the combined AI view is more useful than any single model's portfolio or the S&P 500 benchmark.

Benchmark difficulty matters because model scores should be interpreted against the opportunity set and the market window they faced.

This shows whether models identified the eventual best asset before scoring, even when portfolio weights were too small to fully capture the oracle return.

Attribution turns a model score into an explanation of which holdings actually helped or hurt the frozen portfolio.

Leadership that changes with the broad-market backdrop shows why a single all-history ranking can hide meaningful model strengths and weaknesses.

Environment-specific returns show whether a model's all-history result is broad or depends on a particular market direction.

Environment-specific returns show whether a model's all-history result is broad or depends on a particular market direction.

A stronger floor identifies models whose benchmark performance is less dependent on one favorable market direction.

A large directional split identifies models whose benchmark behavior is especially sensitive to the market backdrop.

Environment-specific returns show whether a model's all-history result is broad or depends on a particular market direction.

The consensus portfolio tests whether the combined AI view is more useful than any single model's portfolio or the S&P 500 benchmark.

Benchmark difficulty matters because model scores should be interpreted against the opportunity set and the market window they faced.

This shows whether models identified the eventual best asset before scoring, even when portfolio weights were too small to fully capture the oracle return.

Attribution turns a model score into an explanation of which holdings actually helped or hurt the frozen portfolio.

Leadership that changes with the broad-market backdrop shows why a single all-history ranking can hide meaningful model strengths and weaknesses.

Environment-specific returns show whether a model's all-history result is broad or depends on a particular market direction.

Environment-specific returns show whether a model's all-history result is broad or depends on a particular market direction.

A stronger floor identifies models whose benchmark performance is less dependent on one favorable market direction.

A large directional split identifies models whose benchmark behavior is especially sensitive to the market backdrop.

Environment-specific returns show whether a model's all-history result is broad or depends on a particular market direction.

This compares how strongly current model portfolios favor assets that had already outperformed. It describes the allocation and does not infer why the model chose it.

This compares how strongly current model portfolios favor assets that had already outperformed. It describes the allocation and does not infer why the model chose it.

The score translates allocations into a common risk scale, so readers can see whether models are collectively leaning defensive, balanced, or aggressive.

A horizon split helps readers separate short-window positioning from the longer one-month model view.

This shows how far model portfolios differ from the group. It does not say that being different produced a better return.

This shows the current crowding point in model capital allocation, before the open rounds receive their final market scores.