AgentSeism · Fig. 1 · Regression explorer

What the average hides

An agent is stochastic: run it twice and some tasks come out differently. Below, every mark is one run. Pick a pull request and play it.

What this is. These are the three seism demo scenarios. Every run's outcome is written out in the scenario, not drawn at random, and every number and verdict here comes from AgentSeism's own decision engine. The trace is drawn from the same runs. It shows how the check reasons. It is not evidence that it works. For that, see the pre-registered real-agent study and the two public example PRs.
github.com/lyr-ai/agentseism · Why agent CI can't be treated like deterministic tests