A working reference for the terminology used across Project Catalyst — spanning program governance, AI/ML and MLOps engineering, healthcare and regulatory context, delivery method, and program finance. Terms are grouped by domain; use the search box to filter across every category at once.
Program & Governance
| AI Governance BoardAIGB | One of the program's three standing governance bodies; owns model-risk and responsible-AI decisions, approves models for pilot and production, and signs phase gates on AI-specific criteria. |
| Center of ExcellenceCoE | Hub-and-spoke operating model that centralizes AI standards, reusable assets, and governance while embedding practitioners in delivery teams. The AI CoE is a cross-cutting workstream that transitions to ACME steady-state operations in Year 3. |
| Enterprise Architecture Review BoardEARB | Standing board that reviews technical architecture, security, and integration; owns the Platform Foundation gate and any architecture-affecting change. |
| Executive Steering BoardESB | Top governance body for business, budget, and risk decisions; authorizes the cost baseline, releases contingency, and approves re-baselining. |
| Joint Application DesignJAD | Facilitated requirements workshop series run per BRD in Phases 0–1, with mandatory Legal, Compliance, Enterprise Architecture, Cybersecurity, and business-SME participation. |
| Phase Gate | A formal go/no-go decision point between phases requiring sign-off from the three boards against defined exit criteria before the next phase is funded to proceed. |
| Two Lines of Defense | Model-risk control structure separating the first line (AI Governance & CoE, who build and self-assess) from an independent second line (Independent Model Validation) that must clear models before production. |
AI / ML & MLOps
| Agent-Assist | An AI capability that supports a human worker in real time (e.g., suggesting responses to a service agent) rather than acting autonomously — a core pattern in BRD-02. |
| Drift (data / concept) | Degradation of model performance over time as live data diverges from training data (data drift) or as the underlying relationship changes (concept drift); monitored continuously in the MLOps pipeline. |
| Fairness / Bias Testing | Evaluation of a model's outputs across protected and sensitive groups to detect disparate impact; a gating requirement before any underwriting or claims model reaches production. |
| Ground Truth | The verified correct outcome used to train and evaluate a model; for claims and prior-auth AI, established from adjudicated historical decisions. |
| Hallucination | A confident but false or fabricated output from a generative model; mitigated in member-facing tools through retrieval grounding, confidence thresholds, and human-in-the-loop review. |
| Human-in-the-LoopHITL | A control design that keeps a qualified person in the decision path for high-stakes or low-confidence AI outputs, rather than allowing fully autonomous action. |
| Independent Model ValidationIMV | The program's second line of defense: a team separate from delivery that independently validates model performance, fairness, and documentation before production. Operates on a 5–10 business-day review queue. |
| Large Language ModelLLM | A generative model trained on large text corpora, used in the program for member/provider conversational assistance and agent-assist, always within governed guardrails. |
| MLOps | The engineering discipline and toolchain for building, deploying, monitoring, and retraining models reliably in production — including versioning, CI/CD for models, drift monitoring, and rollback. |
| Model Card | Standardized documentation of a model's purpose, data, performance, limitations, and validation status; required governance artifact for every production model. |
| Retrieval-Augmented GenerationRAG | A pattern that grounds an LLM's responses in retrieved, authoritative source content to reduce hallucination and keep answers traceable to policy. |
| Shadow Mode | Running a model against live inputs without acting on its outputs, to measure real-world performance safely before go-live — used for the underwriting risk model in BRD-03. |
Healthcare & Regulatory
| CMS-0057-F | The CMS Interoperability and Prior Authorization final rule; its January 2027 requirements are a genuine regulatory driver for the program's context. ACME's baseline compliance is handled by a separate project and is explicitly out of Project Catalyst's scope. |
| ISO/IEC 42001 | The international management-system standard for artificial intelligence; one of the program's two primary governance anchors, providing the AI management-system frame. |
| Moffatt v. Air Canada | A 2024 decision holding an airline liable for its chatbot's incorrect statement; cited in the program as precedent for member-facing conversational AI liability and the basis for human-in-the-loop and disclaimer controls. |
| NAIC Model AI Bulletin | The National Association of Insurance Commissioners' model bulletin on the use of AI by insurers; informs the program's governance of underwriting and risk models. |
| NIST AI Risk Management FrameworkAI RMF | The U.S. National Institute of Standards and Technology framework for managing AI risk (Govern, Map, Measure, Manage); the program's second primary governance anchor. |
| Prior AuthorizationPA | The health-plan process of approving a service or medication before it is delivered; the flagship BRD-01 applies AI to accelerate and improve consistency of prior-auth decisions under clinical oversight. |
| Protected Health InformationPHI | Individually identifiable health information governed under HIPAA; its handling drives onshore-only staffing for claims-decision roles and the Data Privacy Office's controls. |
| Utilization ManagementUM | The clinical review function that evaluates medical necessity and appropriateness of care; the business owner and subject-matter source for prior-authorization AI. |
Delivery & PM Method
| Business Requirements DocumentBRD | The signed requirements baseline for a delivery leg; the program has three (BRD-01 Claims & Prior Auth, BRD-02 Member/Provider UX, BRD-03 Underwriting & Risk). |
| Definition of Done / ReadyDoD / DoR | Agreed quality checklists that a backlog item must meet to enter a sprint (Ready) or be accepted as complete (Done), including governance and validation criteria for AI work. |
| Earned Value ManagementEVM | A cost/schedule control method comparing planned value, earned value, and actual cost to derive SPI and CPI; the program reports EVM to the ESB monthly. |
| Hybrid Delivery | The program's three-layer method: phase-gated governance over agile sprint execution over a governed MLOps lifecycle — detailed in the Methodology Guide. |
| RAIDD | The consolidated register of Risks, Assumptions, Issues, Dependencies, and Decisions used to manage program uncertainty. |
| RACI | A responsibility-assignment matrix mapping who is Responsible, Accountable, Consulted, and Informed for each activity or decision. |
| Cost / Schedule Performance IndexCPI / SPI | EVM efficiency ratios; a value of 1.00 means on-baseline, below 1.00 means over cost or behind schedule. |
| Work Breakdown StructureWBS | The hierarchical decomposition of program scope into elements, groups, and work packages; the backbone to which every dollar and deliverable traces. See the WBS Console. |
Financial & Commercial
| Contingency Reserve | A management reserve (here $9.0M, 9.1% of budget) held outside workstream allocations and released only by ESB approval against a realized risk. |
| Cost-Benefit AnalysisCBA | The financial case comparing program cost to expected benefits; the program's CBA shows a positive NPV (+$2.97M) and simple payback of ~7.6 years. See the CBA. |
| FinOps | The practice of managing and optimizing cloud and AI compute spend; guardrails and budget alerting are a specific control against GenAI token/compute cost overrun. |
| Net Present ValueNPV | The discounted value of future benefits net of costs; used to justify the program's investment. |
| Statement of WorkSOW | The contract defining scope, deliverables, and payment for each program year; three envelopes total $99.0M ($27.72M / $41.58M / $29.70M). |
| Total Cost of OwnershipTCO | The full 10-year cost of a solution including run and maintenance; the program's TCO model compares build vs. status-quo options. See the TCO. |
Acronym Quick-Reference
| Acronym | Expansion |
|---|---|
| AI RMF | (NIST) Artificial Intelligence Risk Management Framework |
| AIGB | AI Governance Board |
| BRD | Business Requirements Document |
| CBA | Cost-Benefit Analysis |
| CMS | Centers for Medicare & Medicaid Services |
| CoE | Center of Excellence |
| CPI / SPI | Cost / Schedule Performance Index |
| DoD / DoR | Definition of Done / Definition of Ready |
| EARB | Enterprise Architecture Review Board |
| ESB | Executive Steering Board |
| EVM | Earned Value Management |
| HITL | Human-in-the-Loop |
| IMV | Independent Model Validation |
| JAD | Joint Application Design |
| LLM | Large Language Model |
| MLOps | Machine Learning Operations |
| NPV | Net Present Value |
| PA | Prior Authorization |
| PHI | Protected Health Information |
| RAG | Retrieval-Augmented Generation |
| RAIDD | Risks, Assumptions, Issues, Dependencies, Decisions |
| SOW | Statement of Work |
| TCO | Total Cost of Ownership |
| UM | Utilization Management |
| WBS | Work Breakdown Structure |
No terms match your filter.