Direct comparison
All measurement dimensions match exactly.
PassMeasurement contracts
A score can share a scale only when its model, prompt, retrieval, grader, sampling, control, and runner contract matches—or a declared overlap bridge passes.
All measurement dimensions match exactly.
PassA directed paired study names and calibrates changed dimensions.
1 activeNo adequate bridge means Holdout refuses continuity.
Fail closed| Provider | Requested model | Observed identity | Epoch | Prompt hash | Boundary |
|---|---|---|---|---|---|
| OpenAI | gpt-5.6-sol | gpt-5.6-sol / recorded fixture | openai-2026-07-a | sha256:demo-prompt-set-7f9a | Comparable |
| Anthropic | claude-sonnet snapshot | archived-fixture-2026-06 | anthropic-2026-06-b | sha256:demo-prompt-set-7f9a | Comparable |
| Gemini | gemini snapshot | archived-fixture-2026-07 | google-2026-07-a | sha256:demo-prompt-set-7f9a | Bridged |
| Perplexity | sonar snapshot | archived-fixture-2026-06 | perplexity-2026-06-c | sha256:demo-prompt-set-7f9a | Comparable |