The build kit

Architecture, integrations & the seam

What talks to what, what CLEAR must supply, and the boundary between the two volumes.

The integration map

                          ┌──────────────────┐
   POS ──MISMO 3.4──▶     │                  │  ◀──events──  POS
                          │    NachoLOS      │
   Pricing ◀──plan code──▶│                  │──MISMO──▶ Investors
                          │   one loan       │──UCD────▶ GSE
   Credit / flood ───────▶│    object        │──MIN────▶ MERS
   VOE / VOA / 4506 ─────▶│                  │──case#──▶ FHA Connection
   Appraisal (XML) ──────▶│                  │──GUS────▶ USDA
   Title / MI / fraud ───▶│                  │──COE────▶ VA
   AUS (DU / LPA) ◀──────▶│                  │
   Doc generation ◀──────▶└──────────────────┘

Every integration is two-way and tracked. An order placed from the LOS records the request, the response, the timestamps and the document, in the file. That is the standard the current vendor-ordering screen already meets — the one thing it does well — and it applies to all of them.

Vendor by vendor — the actual stack

Sharon's answers, which supersede any assumption about what a lender "typically" uses.

Category	Today	State	What v1 needs
Pricing	Polly	in use	Plan-code mapping both directions
Credit	Certified Credit	works	Tri-merge; representative-score rule applied; supplements
Flood	ServiceLink	works	Order + certificate stored
VOE / WVOE	Certified → Work Number	"doesn't always work"	Reliable in-system ordering
Appraisal / AMC	Appraisal Firewall	standalone, not integrated	Order + XML auto-populate; ROV process
Doc generation	DocMagic	"SUPER clunky"	⭐ True field & program sync — see below
Compliance	Doc-vendor + manual tool	manual, huge output	Replaced by the gate engine
MI	Certs pullable; quotes external	partial	Quotes + certs in-system; non-delegated upload path
Title	Email from the LOS	not recorded anywhere	Request through the LOS, action recorded
Fraud	First American	in use	⭐ wants DataVerify
AUS	DU + LPA	in use	In-system, findings stored, re-run diff
MERS	Manual on the MERS site	not integrated	MERS OnLine/API register + transfer
FHA Connection	Manual on the HUD site	not integrated	Case # + CAIVRS + insuring
GSE / UCD	Manual	not integrated	Round-trip in-system

⚠️ Two vendor cautions worth writing down. (1) The credit vendor's "MERS" product is a verification report, not registration or transfer. Buying it expecting MERS lifecycle capability leaves the actual gap unfilled. (2) Integration availability is not integration quality. Several capabilities are nominally available today and are routed around in practice because they do not work reliably. The acceptance criterion is that staff stop going to the vendor's website — not that an endpoint responds.

⭐ The document-generation sync is the highest-value integration. Sharon's #3 v1 pain. Today the closer must leave the LOS and complete fields on the vendor site for anything that is not a straight agency, VA or FHA loan — Non-QM with prepayment penalties, DPAs, DSCR, and even USDA. Every trip is a re-key and a chance to diverge. Acceptance test is blunt: pick the five hardest programs and produce closing documents without leaving the system.

⭐ The credential checklist — what CLEAR must supply

This is the schedule risk. These are sourced by the lender, not by developers, and on programs like this they are consistently what delays a real end-to-end test. Each needs an owner and a date now, not in month four.

Credential	Needed for	Lead time	Owner
DU (via DO) + LPA (S2S)	Running AUS in-system	long	CLEAR
MERS OnLine / API + org ID	Register and transfer	long	CLEAR
FHA Connection access + roles	Case #, CAIVRS, insuring	medium	CLEAR
VA portal / COE access	Entitlement, funding fee, LGC	medium	CLEAR
USDA GUS	Guarantee, eligibility	medium	CLEAR
Credit reseller production credentials	Credit, flood, VOE, 4506	medium	CLEAR
Document vendor production + test	Disclosures, docs, UCD	medium	CLEAR
Pricing engine API	Plan codes, locks	short	CLEAR
Appraisal / AMC API	Orders + XML	medium	CLEAR
Investor delivery endpoints + profiles	Investor Connect	per investor	CLEAR
Warehouse line MERS org numbers	Interim funder registration	medium	CLEAR
GSE UCD collection access	UCD round-trip	medium	CLEAR

Recommended v1 order: credit → AUS → document-generation sync → MERS → appraisal. Credit and AUS first because nothing downstream is real without them; doc sync third because it is the deepest pain; MERS fourth because it is fully manual today.

⚠️ The dependency nobody has flagged

NachoPE — the pricing engine the LOS feeds — is not solid yet. An independent review in July found all 58 contracted features present but zero production-ready, with critical security and financial-integrity defects. The LOS is downstream of plan codes and pricing. A fee sheet that is exactly right, fed by a price that is wrong, produces a wrong disclosure. This belongs on the risk register, not in a separate document nobody reads next to this one.

The technical floor — Phase 0

A source-level review of the existing NachoLOS codebase found no global authentication middleware — authentication is applied per-endpoint, which means any endpoint missing the dependency is open. That is a Phase-0 finding, not an architecture question, and it should be closed before anything in this Atlas is layered on top. Alongside it: confirm the tenancy model (multi-tenant, row-level isolation, day one — retrofitting tenancy is reliably the most expensive mistake available in this category), migrations, CI, and the four environments including a sandbox-backed integration environment where a full application can flow POS → LOS → documents → delivery without touching a real borrower.

⭐ Migration — the problem nobody has scoped

A cutover date arrives with a live pipeline in every stage — some disclosed, some approved, some with documents out.

Approach	Cost	Risk
Drain — new loans in the new system, existing pipeline finishes in the old	Two systems for ~90 days; staff work both	Lowest risk, highest fatigue — recommended
Full migration — move everything	Data mapping for every in-flight state	High — disclosure history and TRID timing are the hard part
Hybrid — migrate pre-disclosure files only	Moderate	Moderate — a reasonable compromise

What must migrate regardless: closed-loan history for QC, HMDA, exam and repurchase response — read-access at minimum. A cutover that abandons the historical record is not a cutover. Not urgent today; unforgivable to discover in month eleven.