How It Works
One frozen brief, one scored decision, no model tools, and real market prices.
Audit
Audit And Data
Check how the tests work, how scores are calculated, and where the source files live.
Start here
Open test rounds How To Read CapitalBench
Start with how it works, then scoring. Open a test round when you want the prompt, model portfolios, prices, and audit hashes.
Scoring
How returns, S&P context, and CapitalBench Score are calculated.
Fairness
Why models get the same information and why old scores are not rewritten.
Limitations
Noise, sample size, hosted model changes, and non-advice boundaries.
Insights
Readable signals generated from model positioning, benchmark results, risk appetite, and scoring windows.
Model Behavior Benchmarks
Side-by-side model behavior profiles covering risk appetite, concentration, turnover, Portfolio Difference, and sample caveats.
Insights Engine Plan
Research and implementation plan for deterministic and LLM-assisted benchmark insights.
Asset List
The public asset choices available to models in each test.
Round 1
Model portfolios, starting prices, status, and audit hashes.
Changelog
Major public changes to the benchmark, site, methodology, data, and operations.
Repository
Source code, schemas, local files, and implementation history.