Research program / Fleet evaluation research
Provider-free foundationCore32
Compare method, role, and state-transition behavior with framework-neutral fixtures while retaining the scope and limits of each evaluation.
- Project type
- Fleet evaluation research
- Surface
- Public explanatory projection
- Artifact
- Evaluation fixtures and role comparisons
- Source boundary
- Core32 Fleet Evaluation Framework public report orientation
Compare without flattening
How can method and role performance be compared without collapsing different responsibilities into one score?
Define the task before judging the role
Declare the task, source scope, expected return, criteria, and failure conditions before interpreting a candidate response.
Roles are bounded responsibilities
A role description identifies a responsibility. It does not establish selection, loading, successful task execution, or a professional credential.
Task × role × criterion
Keep the task fixture, selected responsibility, and evaluation criterion separately inspectable. A result belongs to its conditions rather than to a universal ranking.
Preserve disagreements
Compare returns against the declared fixture. Record missing evidence, boundary errors, and useful distinctions individually so an aggregate cannot hide them.
Provider-free research posture
The public report describes a framework-neutral foundation. This page supplies no new fleet benchmark, independent replication, or measured model ranking.
A comparison with a return route
Show the criterion beside its result and source. Give the reviewer a way to inspect the fixture and understand what was not tested.
Evaluation is not activation
A roster or favorable comparison does not select a profile or authorize a provider invocation. Adoption and runtime use require their own scoped decision.