Ask a plain-language policy question with a clear, single-KB-article
answer; confirm the response includes an inline citation reference to
that article and that the article/version cited is the currently
published one.
Ask a follow-up question in the same thread; confirm the thread resumes
with prior context rather than starting a new thread, and both turns
persist to ServiceNow (cross-check against auditable-conversation-history
test plan).
Ask a question spanning two related policies (e.g. a state-transfer
question that also touches a remote-work policy); confirm the answer
either synthesizes both citations or flags the conflict rather than
silently answering from only one.
Ask a question about a jurisdiction with only Tier 2 (configured-source
fallback) coverage; confirm the advisor answers from the known-good web
source and cites it as a web source, not as a Codex-grounded citation.
Ask a question about a jurisdiction with no configured sources at all
(Tier 3); confirm the advisor still answers via open web search rather
than refusing, and that nothing in the response implies Codex-grade
confidence.
Ask a question citing a federal regulation (e.g. FLSA); confirm the
response includes a reference number and does NOT present a clickable
link (until Codex Curator's source-URL tracking ships) — a link
appearing here unexpectedly would indicate scope creep into the Codex
Curator dependency.
Attempt to reach the advisor chat surface as a non-HR-Operations user
(e.g. a general employee or people-manager role); confirm access is
denied — this surface is HR-Operations-only per scope.
Attempt to get the advisor to edit or publish a policy directly through
conversation (e.g. "please update this policy for me"); confirm it
declines/redirects rather than silently mutating KB content, since the
advisor cannot edit or publish today.
Close and reopen the advisor surface mid-conversation (simulating an
interruption); confirm the thread and prior turns are resumable from
ServiceNow-persisted state, not lost.