Start with the approved release
The declared Stack Release holds the model, harness, instructions, tools, permissions, routing, workflow, verification, and runtime your team meant to evaluate.
Product / Runtime Verification
How do you verify which model and harness actually ran?
A coding-agent run can change shape after it starts. Models get rerouted. Plugins can appear while a subagent takes over the hardest step. Fullbeam records enough runtime identity to tell whether the accepted result still belongs to the release under evaluation.
What the release gate checks
The evidence behind the decision
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.
Follow the execution graph
Observe model routes, tool and plugin identities, schema changes, permission state, subagent handoffs, and runtime transitions as the session moves from task to accepted change.
What is the difference between coding-agent tracing and evaluation?
Fullbeam marks evidence as verified, partial, drifted, or unknown. A copied release label cannot hide a changed tool schema, a fallback model, or a gap in observation.
Persist only bounded evidence
Fullbeam persists an allowlisted evidence envelope containing digests, lineage, coverage, measured cost, and outcomes. Source checkouts, provider credentials, prompt bodies, patch bodies, hidden grader bodies, raw tool output, and model reasoning are not retained as default evidence.
Bring us the next stack change
A cheaper route gets credit for the work it completed itself. Expensive fallbacks and unqualified subagents stay visible in the result.
One current releaseOne candidatePrivate repository workA workload-level decision