Skip to content

verify-functionality.md

Latest commit

 

History

History
49 lines (36 loc) · 1.38 KB

verify-functionality.md

File metadata and controls

49 lines (36 loc) · 1.38 KB
Raw

Verify Functionality

Run the complete deployment verification from the example root:

bash scripts/verify.sh

It checks:

  • all six bundled payment decisions;
  • Phoenix, the mock rail, and the FinGuard UI;
  • NeMo Relay and Hermes health inside the sandbox;
  • the rendered Relay plugin configuration;
  • denial of payment-rail access from the sandbox.

Offline-only checks

These do not require Docker, OpenShell, or inference credentials:

bash -n scripts/*.sh agents/hermes/start.sh
python3 -m py_compile $(find . -name '*.py' -not -path './.git/*' -print)
python3 scripts/smoke-payment.py
git diff --check

Human checker

python3 scripts/approve_release.py --id WIRE-1007 --approver "Jane Ops"
python3 scripts/approve_release.py --id ACH-2003 --approver "Jane Ops"
curl -fsS http://127.0.0.1:8780/released

Expected: WIRE-1007 is released and ACH-2003 is refused because it is on hold. UI-driven host decisions emit audit spans under service finguard-host-checker, distinct from NeMo Relay's agent spans.

In the UI, exercise screening and Agent: Release before checking Phoenix. Those actions call Hermes and produce Relay evidence. Merely loading the queue or ledger intentionally does not create a trace.

Relay logs and traces

openshell sandbox exec --name payment-ops -- tail -50 /tmp/nemo-relay.log
bash scripts/download-traces.sh