Illustrative demo data. The numbers below are synthetic examples that
demonstrate the evaluation methodology and reporting format. They are not measured or
independently verified results. Measured values will replace them when benchmark runs
are published.
Hallucination rate by intervention (illustrative)
| Domain | Base | Self-consistency | Judge gate | M-Stack |
|---|
Calibration (illustrative)
Risk-controlling coverage curve (illustrative)
Lower risk at equal coverage is better; the M-Stack curve shows the intended shape of the improvement.