Moving between Claude Code, Codex, OpenCode, or an internal client changes the harness, tools, permissions, context policy, and checks around the model, so Fullbeam compares the complete current and candidate setups before your Platform team widens access.
Each result stays tied to the Stack Release, workload, repository state, and coverage that produced it. If evidence is missing, the gap remains visible in the release call.
Freeze the model route, harness, instructions, skills, tools, permissions, context policy, workflow, verification, and runtime for both sides before running the first comparison case from a reconstructed repository state.
Run maintenance, migrations, incident fixes, feature work, review corrections, and follow-up changes from the codebases where the candidate may become the default.
Repeat the critical cases. Track cost spread, tail failures, candidate-only regressions, and evidence gaps so one polished demo does not decide the rollout.
The candidate can take routine test repair while the current release keeps infrastructure and authentication. Fullbeam records that split as the release decision.
Promote the candidate where it wins, preserve the baseline where it falls short, and keep uncertain workloads out of the rollout.
