EvidenceStrong architectural inferencev1.22.1
In plain English
This page shows what kind of support exists for each claim: real systems, experiments, early evidence, architectural reasoning, open questions, or speculative scenarios.
- Why this matters: AI risk can come from the whole arrangement, not one obvious model.
- What to look for: data, memory, routes, adapters, tools, evaluators, updates, and rollback paths.
- Technical version below: the expert terminology remains available and is linked through the glossary.
Evaluator independence
Evidence card
- Claim
- Candidate artifacts should not be able to edit tests, thresholds, policies, or evidence records.
- Evidence level
- Architectural inference
- Source
- https://modelbreeder.com/theory/evaluator-independence
- Publication date
- 2026-06-26
- Authors or institution
- ModelBreeder.com
- System tested
- Evaluator boundary and evidence-store pattern.
- Limitations
- Control-plane design principle, not evidence that evaluator monoculture is solved.
- What the evidence does show
- Candidate artifacts should not be able to edit tests, thresholds, policies, or evidence records.
- What the evidence does not show
- That model-based judges are independent merely because they are outside the candidate process.
- Date last reviewed in UTC
- 2026-06-26T00:00:00Z
Site use
This source supports Cognivirus.com pages related to evaluatorA system that judges whether an AI output or candidate is acceptable. Open glossary definition independence, hidden tests, append-only evidence. Its role is bounded by the limitations listed above.