Mortgage Technology Glossary

A multi-tenant, web-delivered loan origination system that runs the full mortgage lifecycle — application, processing, underwriting, closing, and post-close — on shared infrastructure with continuous updates. Compared to on-premise or hybrid-cloud LOS platforms, a true cloud LOS has no per-tenant deployment work, releases features continuously rather than quarterly, scales elastically, and shifts patching/uptime/disaster recovery to the vendor.

Software that performs a specific origination task autonomously — document classification, income calculation, condition clearing, TRID timing — and explains its work in an audit-traceable way. It differs from a chatbot (which only retrieves) and from rule-based automation (which executes static logic) by combining decision-making with action and explanation. A modern mortgage LOS should ship 5–9 specialized agents in production.

Self-employed borrower income calculated using Fannie Mae 1084 worksheet logic with no LLM in the math path. Same inputs always produce the same output, so investor and regulator audits can reproduce the math on demand. Covers Schedule C, Schedule E, K-1, 1120-S, 1065 add-back rules including depreciation, amortization, business use of home, and the 50% meals deduction. Also covers 2-year trending and declining-income flag (>10% YoY decline).

A workflow timer that enforces TRID 3-business-day Closing Disclosure delivery and Loan Estimate timing rules using a durable execution engine (Confer uses Temporal). Timer state persists at every transition, so server restarts, deploys, and worker crashes cannot lose the clock. Distinguishes from cron-based or queue-worker timers, which fail in failure scenarios. Eliminates the highest-severity TRID timing-violation risk per 12 CFR § 1026.19(e)–(f).

Per-loan verification of the eight Ability-to-Repay factors required under 12 CFR § 1026.43(c): (1) current/expected income, (2) employment, (3) other income relied upon, (4) monthly mortgage payment, (5) payments on simultaneous loans, (6) other obligations, (7) DTI/residual income, (8) credit history. Modern compliance architecture verifies all 8 factors per loan with stored evidence rather than sampling. Workflow-enforced ATR blocks close until factors are documented.

An open protocol for exposing tools and data to LLMs in a standard interface. Lets any LLM — Claude, GPT, Llama, or purpose-built models — interact with a software platform through a common contract. For mortgage LOS, MCP exposes underwriting, document AI, and compliance capabilities so downstream BI, customer service, and executive dashboards can connect without per-vendor SDK work. Confer ships 32+ MCP tools across underwriting, document AI, and compliance domains.

Mortgage Industry Standards Maintenance Organization version 3.4 — the XML data standard for exchanging mortgage loan information between systems. Used for AUS submission to DU/LPA, secondary market delivery (ULDD), and lender-to-investor handoffs. Mid-sized lenders evaluate LOS platforms in part on MISMO 3.4 export quality; defects in ULDD validation cause secondary market delivery delays of 3–10 business days per loan.

Real-time, two-way synchronization of loan data between an external system and ICE Mortgage Technology Encompass via the Encompass API. Bidirectional means changes made in either system propagate to the other automatically. Common use: AI document processing on a third-party platform writing extracted fields back into Encompass while loan officers continue working in Encompass as the system of record. Confer maintains 180+ bidirectional Encompass field mappings covering all 15 E-Folder containers (U1–U15).

Continuous capture of the 110+ Home Mortgage Disclosure Act Loan Application Register fields from origination data as a loan progresses through its lifecycle, rather than as a year-end batch process. Eliminates 80–120 FTE-hours of annual HMDA filing prep and dramatically reduces inaccurate-HMDA exposure (CFPB enforcement penalties: $50K–$2M+). Pairs with FFIEC edit-check validation that runs continuously rather than during the March filing crunch.

Real-time conversational AI that handles inbound borrower status calls, walks through outstanding conditions, prompts document uploads, and transfers to a loan officer when needed. Distinct from IVR-with-text-to-speech: a real voice AI pulls live loan data and answers borrower questions on the call. Replaces 30–90 minutes of LO/processor time per loan over the lifecycle. Confer ships Voice AI (Kylie) running on VAPI infrastructure.

Append-only ledger capturing every agent action, data mutation, decision, and state transition on a mortgage loan with operator, timestamp, before/after state, and reasoning where applicable. Append-only means existing entries cannot be edited or deleted. Enables regulator and internal QC to reconstruct exactly what happened on a loan without inference. Required architecture for SOC 2, FFIEC, and CFPB exam defensibility. Confer captures every agent action and data write into this ledger by default with row-level security ensuring tenant isolation.

Quality control checks that run on every loan before close, rather than sampling 10% of files post-fund. Validates income reproducibility, AUS findings vs. final terms, TRID timing, HMDA completeness, and ULDD/MISMO export against investor specs. Targets sub-0.5% critical defect rate vs. the ACES Q4 2024 industry average of 1.79%. Reduces investor repurchase exposure ($15K–$50K per loan) by catching defects while they can still be cured.