General baseline
HELM
Holistic evaluation design: accuracy, calibration, robustness, fairness, and efficiency measured together rather than one headline score.
As of 13 August 2026. Conditional methodology reference. Stanford CRFM HELM. A framework, not a single dataset.
HELM measures accuracy, calibration, robustness, fairness, and efficiency as separate dimensions. That design is why it is listed. Scenarios pull their own data and licenses.
Antidote already refuses unvalidated composites. HELM is cited as evaluation design, not installed as a winner table.
- Scale
- Stanford CRFM HELM scenarios. A runnable framework, not a single download.
- Provenance
- Stanford CRFM HELM. Methodology reference, not a winner table.
- Contamination
- Varies by scenario.
- Cannot establish
- A diligence-workflow pass rate, or an overall HELM winner used as a moat.
Metrics, kept separate
- Scenario-specific official metrics
- Calibration
- Robustness
No cited public figure is installed for this pack. Method, scale, and gates still apply. A missing chart is not a pass rate.
What the figures bound
Read HELM for how to keep dimensions apart. Do not import a HELM overall as a moat.
No chart is installed.
What this page does not conclude
- A diligence-workflow pass rate
- An overall HELM winner
- GPT-5.6 or Fable 5 HELM without a cited scenario cell
- Vendor economics from a framework score
Cited as calibration on these memos — not a score of those products.
Not attached to a software, advisory, or automation memo yet.