Skip to content

Glossary

Starting set, pulled from the three seed documents behind this repo's Feature Sets (the Nova Codex Curator PRD v1.5, the Nova Policy Advisor PRD v1.5, and the Agent Accountability Framework concept paper — all dated June–July 2026). Add to this as new terms need pinning down.

Nova Codex / Codex Curator

Two related but distinct things share the "Codex" name. Codex (sometimes "the Codex corpus") is the structured regulatory database itself — the raw material Nova Policy Advisor's answers are grounded in. Codex Curator is the pipeline and internal tool that populates and maintains that database (discover sources → SME curates → crawl → SME reviews documents → sync to the vector DB). The PRD's own title uses "Curator" for the product, but several passages say "Codex" alone when talking about the data ("Codex's coverage and freshness," "Codex-grounded" as a grounding tier). When reading either PRD, check whether "Codex" refers to the data store or is shorthand for the whole Curator product before assuming which.

Nova Policy Advisor / Nova Advisor Agent

Same ambiguity pattern as specs-superagent's Super Agent / Nova HR Concierge entry. Nova Policy Advisor is the product name for the thread-based, HR-Operations-only conversational compliance advisor described in its PRD. Nova Advisor Agent is the same system referred to as a "system consumer" persona in the Codex Curator PRD, and also shows up mid-document in the Policy Advisor PRD itself as the thing whose retrieval index re-syncs on publish and whose WebSocket layer relays collaborative edits. Treat "Nova Advisor Agent" as the engineering/runtime name for the same system "Nova Policy Advisor" names as a product — worth confirming with Ravindra Kumar's team rather than assuming, since (unlike the Super Agent / HR Concierge pairing) neither PRD states outright that they're the same thing.

Three-Tier Grounding Model

The framework governing how confidently Nova Policy Advisor can answer for a given jurisdiction, defined in the Codex Curator PRD (Section 3) and referenced by the Policy Advisor PRD's fallback strategy:

  • Tier 1 — Codex-grounded: fully crawled, SME-reviewed, ingested; answers come from curated, versioned Codex content.
  • Tier 2 — configured-source fallback: approved source websites exist for the jurisdiction but haven't been fully crawled/curated yet; the advisor pulls directly from those known-good sites.
  • Tier 3 — open web: no sources configured at all yet; the advisor falls back to general web search with no curation behind it.

Per the Codex Curator PRD, which tier grounded a given answer is not currently surfaced to the end user in the Advisor chat surface — that's an open item assigned to the companion Policy Advisor PRD, since a Tier 3 answer should probably carry lower displayed confidence than a Tier 1 one.

Four-State Per-Country Curation Lifecycle

The Codex Curator PRD's maturity model for tracking every country Codex might ever support — distinct from, but closely related to, the three-tier grounding model above (state maps roughly to tier, but is a per-country curation-pipeline status rather than an answer-time grounding label):

  • State A — no source list configured.
  • State B — sources configured, not yet crawled.
  • State C — crawled, review pending (documents exist, haven't cleared SME review).
  • State D — crawled and validated (fully live in Codex).

A Codex Curator attribute captured on every ingested document, distinguishing a USC statute (Track 1) from the CFR regulation that implements it (Track 2). Assigned automatically at crawl time based on source feed (USC vs. CFR), confirmable/correctable by the SME reviewer. Matters because, without it, the advisor risks telling a user it's citing "the law" itself when it's actually citing the implementing regulation — a meaningful distinction in a compliance context, not a pedantic one.

Knowledge Base (KB) article vs. Codex document

Two separate content stores both feed Nova Policy Advisor, and the two PRDs name them differently enough to cause confusion. A KB article is company-authored internal policy content, created and versioned in ServiceNow via the Policy Editor (Nova Policy Advisor PRD) — this is what HR Policy Owners write and publish. A Codex document is externally sourced government regulatory text, crawled and SME-reviewed through the Codex Curator pipeline (Codex Curator PRD) — Novaworks doesn't author this content, only curates it. The advisor retrieves from both: KB articles for "what does our policy say," Codex/tiered-grounding sources for "what does the regulation say." Confirm this split holds as both products mature, since some passages describe the advisor's "retrieval index" as a single thing without always specifying which store a given answer came from.

SME / SME Reviewer

Short for Subject Matter Expert. In the Codex Curator PRD this is a named, proposed role ("SME · Reviewer") — an internal (and possibly eventually contracted) reviewer who curates candidate source lists per country and accepts/rejects/flags individual crawled documents. Distinct from the HR Compliance / Policy Analyst persona in the Policy Advisor PRD, who consumes the advisor's answers rather than reviewing Codex's raw source material — don't conflate the two even though both sit inside HR/Compliance function.

Agent Worker

From the Agent Accountability Framework concept paper: a lightweight worker identity given to an AI agent inside Novaworks (an "Agent Worker" worker type), paired with a machine identity in ServiceNow, a Job Profile, and a named human supervisor. This is a proposed governance concept, not a shipped Nova platform concept — don't confuse it with Functional Agent (FA), the existing specs-superagent term for a Python service handling one domain of work (Core HCM, Time & Attendance, etc.). The Accountability Framework's premise is that Functional Agents (and any other agent acting inside the enterprise) would each be represented by an Agent Worker identity for governance purposes — the Agent Worker is an accountability wrapper, not a replacement runtime.

Job Profile

Per the Accountability Framework: the record defining an Agent Worker's role, scope, permitted actions, and guardrails, authored collaboratively by business leaders and HRBPs (with Novaworks AI's policy editor assisting). Explicitly called out in the source document as a concept still needing a schema and an authoring-workflow spec — see this repo's feature-sets/agent-accountability-framework/schema-changes.md and the open TODOs in the source paper before treating this as settled.

Human-in-the-loop gate (hard gate)

Used across all three source documents to mean a point where an autonomous process cannot proceed without an explicit human decision — not a soft review or an optional check. Codex Curator has two: adding a source to a country's approved list, and accepting/rejecting/flagging an individual crawled document (nothing reaches the vector database without both). Policy Advisor has one: an HR Policy Owner must explicitly publish any policy change, since the advisor never edits or publishes on its own. The Accountability Framework proposes a further one at the agent-identity level (the "kill switch"). Worth using this term precisely — a "gate" that's advisory rather than blocking should be called a review step instead.