{exampleScenario.context}
{exampleScenario.prompt}
+Gates run in order. Inside each gate, criteria form a dependency graph. Here, both Test checks depend on "Tests execute"; if it fails, both are skipped.
+Illustrative demo only. No agents are running. All outcomes, token counts, and durations are invented, not measured benchmarks or agent rankings. Planned agents are not available today and are not included in the sample runs. Select any step to explore.
+ + ++ How do coding agents really handle your product? Give them the same + task, define what good looks like, and see the evidence. That's Scope. +
+ +Your tasks. Your criteria. Repeatable experiments.
+From a prompt to a picture
Beyond a pass or fail
A score is a starting point. Understand the task, the setup, and the path an agent took.
+Define success
+Model what matters as a criteria graph, with dependencies that make the evaluation meaningful for your product.
+ Build your criteria graph + + + +Keep the context
Save agent configuration in versioned profiles, so the next experiment starts from a known setup.
+ Explore profiles + + + +Follow the work
Follow live logs and inspect the run's iterations and criterion feedback to see where things went right or wrong.
+ Explore the workflow + +Make your next move
{path.description}
+ {path.label} + + ))} +Bring your own question
+Start with one task you care about. Scope helps you see the rest.
+ Get started + You'll need access to a Scope deployment. No account is needed for this demo. +From the community
What the community has learned evaluating agentic product experiences with Scope, from the blog and the stage.
- Run one scenario across GitHub Copilot, Claude Code, and VS Code. - Score every run against criteria you define, and see how your - agentic experience holds up across all of them — at scale. -
- -Scope submits your task to each agent, scores every run against the criteria you define, and lines the results up side by side.
-Hand one coding task to GitHub Copilot, Claude Code, or VS Code with the Copilot driver extension — with the profile you want.
-Each run is evaluated against your criteria DAG. Gate child criteria on their parents, or leave them as independent roots.
-See not just whether it worked, but how each agent got there — the tools it reached for and where it got stuck.
-Scope drives existing agents against tasks you control, then captures what they asked for, what tools they reached for, and where they got stuck.
-Hand a task to GitHub Copilot, Claude Code, or VS Code with the Copilot driver extension.
-Bundle worker, model, version, MCP servers, skills, and extensions to re-run a setup consistently.
-Describe what “good” means. Gate child criteria on their parents, or leave them as independent roots.
-Stream logs straight from the worker as each run executes, attempt by attempt.
-Scope detects characteristics on your prompts so you can compare across heterogeneous tasks.
-Submit runs, manage profiles and criteria, and fetch results through the REST API or the CLI.
-Submit requests, configure profiles and criteria, and watch runs stream — all from the browser.
- Submit from the Portal → -Wire Scope into your pipelines and tooling with a single authenticated request.
- Read the API guide → -Run benchmarks from the terminal with real-time log streaming.
- Install the CLI → -Spin up your first benchmark in minutes — from the Portal, the API, or the CLI.
-