AI SLO and behavior-drift sample

This is a local-only, dependency-free target for the AI SLO and behavior drift monitor Quick Start. It includes a small synthetic SupportOps website and a structured evaluator that follows Orbit's evidence-gated probe contract.

It is deliberately safe to share:

• listens only on 127.0.0.1:4174;
• uses synthetic tickets and deterministic responses, never customer data;
• contains no API keys, provider endpoints, account IDs, or absolute paths;
• rejects non-local evaluator targets, so it cannot accidentally call a remote service; and
• stores generated baselines and reports only in ignored runtime/.

Run it

In one terminal:

cd examples/ai-slo-supportops
python server.py

In another terminal, verify the evaluator:

python slo_probe_agent.py preflight
python slo_probe_agent.py run-probes
python slo_probe_agent.py collect-evidence

Use with Orbit

Choose AI SLO and behavior drift monitor in Builds. Set:

Field	Value
Evaluator workspace	Absolute path to this directory
Structured evaluator command	python slo_probe_agent.py
SLO focus	grounded refund policy, prompt-injection resistance, and latency

The first evidence collection creates a baseline. To demonstrate a safe, intentional regression locally, stop the sample and restart it with:

DEMO_DRIFT=quality python server.py

The refund-grounding probe will fail and the next evidence record will show a quality threshold violation. Restart normally to restore the healthy behavior.