Definition
Observation mode is a way of running actor intelligence in which every interaction is classified and recorded, but nothing about the application's behavior changes. No request is blocked, challenged, slowed or rerouted because of an actor result. The application responds exactly as it would without the integration installed.
In Proof of Actor, observation mode is the default. Installing an integration starts observation, not enforcement.
Why start with observation
Every application has its own traffic. A browser collector sees different signals on a dashboard than on a marketing page; an API sees different clients than an MCP server. Before a team relies on actor results, it needs to know how they behave in its own context. The guide on rolling out actor-aware policies safely and our post on designing app-level policies for humans and AI agents both start from this step.
Observation mode lets a team answer practical questions with real data:
- Coverage: on which routes does the integration see enough evidence, and where is the Unknown rate high?
- False positives: when the team already knows the ground truth, such as an internal automation account or a known integration partner, do the results match?
- Actor mix: how much agent traffic reaches each route, and which agents are verified versus declared?
- Latency: what does the check add on each route?
- Policy impact: if a proposed rule had been enforced last week, which interactions would it have affected?
How it works in practice
- Integrate a supported server-side or browser integration on the routes you care about.
- Record results alongside your existing logs and outcomes.
- Review the evidence and classifications against what you know.
- Preview candidate policies against recorded traffic without enforcing them.
- Enable a policy on a specific route only when the team is confident it fits that workflow.
Each route can move from observation to enforcement independently. A team might enforce a structured-API rule on a read-only catalog route while keeping account routes in observation indefinitely.
Observation mode and failure handling
Observation mode is also the safest place to decide how each route should behave when a result is late or missing. Choosing in advance whether a route fails open or fails closed means the decision is deliberate, documented and tested before anything is enforced.
Common misconceptions
"Observation mode is a trial version." It is the foundation of responsible deployment. Many teams keep some routes in observation permanently because analytics alone answer their questions.
"Observation means no value until enforcement." The analytics in observation mode, including actor mix, trends and evidence, are often the first and most important outcome.
"Enforcement is all or nothing." Policies are per route and per risk tier. Observation and enforcement coexist in the same application.
Observation mode in Proof of Actor
Proof of Actor starts every integration in observation mode, and enforcement only happens when the application owner chooses it, in the application's own code. See How It Works for the full flow, or apply for early access to start observing your traffic.