Start from the setup you approved
The approved version lists the model, agent, instructions, tools, permissions, routing, workflow, checks and runtime your team meant to test.
Product / Runtime verification
How do you verify which model and harness actually ran?
A coding agent run can change after it starts. A model gets rerouted, a plugin loads, a subagent takes over the hardest step. Where a supported agent connection is installed, Fullbeam records enough about the run to tell whether the result still belongs to the setup you tested.
In practice
What this looks like on your code
Every result links back to the setup, the tasks and the commit it came from. If something wasn't measured, you see that too, instead of a number that looks complete.
Follow the run
Fullbeam records model routes, tool and plugin identities, schema changes, permissions, subagent handoffs and runtime changes as the agent goes from task to finished change.
What is the difference between coding agent tracing and evaluation?
Tracing records what happened. Evaluation asks whether the result was good and which setup earned it. Fullbeam marks each run as verified, partial, drifted or unknown, so a copied version label can't hide a changed tool schema, a fallback model or a gap in what was recorded.
Keep only what the result needs
Fullbeam keeps an allowlisted record: hashes, lineage, coverage, measured cost and outcomes. By default it does not keep source checkouts, provider credentials, prompts, patches, hidden test contents, raw tool output or model reasoning.
Got a coding agent change coming up?
A cheaper model gets credit only for the work it did itself. If an expensive fallback model or an untested subagent finished the job, the result says so.
Your current setupThe change you want to tryPRs you already mergedAn answer for each kind of work