The Atlas says what to build and why. This says what it looks like — 51 tappable screens, the journey on one page, the UI kit, the integration map, and all 195 features as a filterable backlog. Built so the team can start from a picture instead of a document.
51 tappable screens across the borrower journey, the AI clone, the LO desk, leadership, partners and compliance. Phone and desktop frames, with a working English/Spanish toggle.
Open the prototype → ⇢The journey mapEight stages by five actors on one page — and the proof that the borrower is never alone in a stage. Every card opens the real screen.
See the map → ◐The UI kitColor, type, components and the voice rules, as live HTML. Small on purpose: fewer choices means four developers ship one product.
Open the kit → ⚙Architecture & integrationsHow the POS, the vendors, the LOS and the CRM connect — plus the credential checklist CLEAR has to source, which is the item most likely to stall the build.
See the wiring → ☰The build backlogAll 195 features with acceptance tests, priorities and phases, generated straight from the Atlas. Filter by P0 or by phase and hand it to a sprint.
Open the backlog → 📕The complete document — hand this to the teamEverything in one file: cover, how-to-use routing table, clickable contents, the decision memo for leadership, and the full 13-part specification. Rev. 2 — Aug 10, 2026. One upload into your tracker or doc system.
Download the PDF →Six pages, for leadership. The ask, Option C, and the two-week sprint.
PDF → 📄Specification onlyThe 13-part Atlas without the memo. ~21,000 words.
PDF → { }Feature catalog — JSONAll 195 features as structured data: id, group, name, accept, pri, phase. Import straight into a tracker.
The entire specification as plain markdown, for ingestion, diffing or feeding to your own tooling.
ATLAS-FULL.md →Docked on every screen, in the LO’s own words, and structurally unable to quote a rate. Three layers: text at launch, voice as fast-follow, video opt-in only. You cannot buy this from a vendor.
Everyone builds a partner portal; nobody uses it. We push status to the agent where they already are, scoped to their buyers, financials never exposed.
The TRID clock starts by system, not by memory. Consent is append-only. The guardrail block is a visible feature, because the audit trail is the product.
Part 0.5 recommends building — but not on cost. On the fact that the clone and the realtor experience are not purchasable, and per-seat pricing taxes recruiting growth. It still needs real vendor quotes at 100 and 300 seats.
Every demo shows the happy path. The rescue screen, the stalled-loan radar and the throttled nudge are where a funded loan is actually won.
The full borrower journey and the clone in Spanish at launch. Staff side stays English; counsel decides which disclosures are translated.
Twenty minutes on the screens surfaces more disagreement than twenty pages. Argue about the pictures now, while changing them is free.
DU/LPA, credit reseller, Truv, DocMagic, MERS. They are long-lead, they gate real testing, and they are exactly what stalled the Nacho build.
The guardrail list is the one design decision that is hard to retrofit. Better to have it signed off before the clone is written than after.
81 features are P0. Filter the backlog to Phase 1 and commit to the spine: tenancy, entry, pre-qual, URLA, disclosures, Tier-1 clone, LendingPad.
TRID, RESPA §8, the AI charter, trigger leads, TCPA. Needed before the clone ships, not after.
Who at CLEAR owns DU/LPA, MERS, credit, Truv and DocMagic accounts — with a name and a date.
GoHighLevel near-term or Total Expert. It changes the integration surface (L8).
Full remote closings on day one, or hybrid first with RON as a fast-follow.
“CLEAR POS” is a placeholder. The borrower will see this name on the most stressful purchase of their life.
Design prototype — not working software. Internal review only.
Built by Willis for Christa Votaw · CLEAR Home Loans