Prepared 2026-07-22. Figures verified as of this date unless marked otherwise.
Source base: research_notes/DECISION_BRIEF.md (binding), research_notes/*.md. All Ready-specific product claims are [UNVERIFIED — confirm with manager] unless noted.
Every model is scored 1–10 on all 14 criteria from the master prompt. 10 is always the favorable end. For the two "bad-thing" criteria, the scale is inverted so the table sums cleanly: Dependence = 10 means low dependence on another company; Distraction = 10 means low risk of distraction from the core mortgage business. Scores reflect Mitchell specifically — a new, analytically strong, less-extroverted MLO at a small Florida-based lender with a fresh Illinois license — not a generic loan officer.
| Criterion | A: Mtg-first relationship | B: Dual-track | C: Zillow Flex | D: Scenario Desk | E: First-Home Lab | F: SE/Investor spec. | G: Hybrid (D+E+F-lite) |
|---|---|---|---|---|---|---|---|
| Speed to first application | 5 | 4 | 5 | 7 | 4 | 4 | 7 |
| Speed to first closing | 5 | 5 | 7 | 5 | 3 | 4 | 5 |
| Trust | 6 | 4 | 4 | 9 | 7 | 6 | 9 |
| Differentiation | 3 | 4 | 2 | 8 | 9 | 8 | 9 |
| Product fit (Ready) | 7 | 5 | 2 | 8 | 8 | 4 | 8 |
| Local demand | 7 | 6 | 7 | 8 | 8 | 6 | 8 |
| Compliance simplicity | 7 | 3 | 2 | 8 | 6 | 4 | 7 |
| Economics | 6 | 5 | 2 | 7 | 6 | 7 | 7 |
| Skill-building | 7 | 5 | 6 | 9 | 7 | 6 | 9 |
| Sustainability (less-extroverted) | 5 | 3 | 3 | 7 | 9 | 7 | 8 |
| AI leverage | 5 | 4 | 4 | 9 | 10 | 7 | 10 |
| Dependence on another company (10 = low) | 7 | 6 | 1 | 6 | 7 | 4 | 6 |
| Long-term defensibility | 4 | 4 | 2 | 7 | 9 | 7 | 9 |
| Risk of distraction (10 = low) | 7 | 2 | 2 | 9 | 6 | 5 | 7 |
| Total (max 140) | 81 | 60 | 45 | 107 | 99 | 79 | 109 |
(Sustainability rescored after red-team review, M3: D was 9, now 7 — the desk's steady state is genuinely low-social, but its ramp months 1–2 are Model-A-style outreach with better scripts, and scoring the steady state while hiding the ramp misled the reader the score exists for. G inherits the same ramp reality: 9 → 8, cushioned by warm-first entry and the half-dose week-2 ramp added in file 12. Totals updated; the ranking is unchanged.)
The totals are a sanity check, not the decision by themselves. Read the per-model paragraphs — several models win individual criteria that the winner does not.
The safe default, and the baseline every other model must beat. It scores decently almost everywhere because it does nothing wrong: mortgage-only focus keeps compliance simple (7) and distraction low (7), and Ready's core agency products [UNVERIFIED — confirm with manager] fit the $250–430k Tinley/Oak Forest/Orland/Lockport market well (product fit 7, local demand 7 — Tinley median $364k, FHA limit $541,287 clears the whole first-timer band; products_ihda.md §4).
Its weaknesses are exactly the master prompt's fear: it is textbook. "Build relationships, join the chamber, call agents" is what every new MLO in the Harlem Ave corridor is told, so differentiation scores 3 and long-term defensibility 4 — there is no asset being built, only goodwill that resets if Mitchell changes companies. It also under-serves his personality: undirected relationship-building is the highest-social-load, lowest-structure version of the job (sustainability 5), and AI adds only marginal leverage to generic networking (5). Speed scores are middling (5/5): without a specific reason for an agent to send a file, the first application waits on luck.
Illinois genuinely permits holding both licenses (IDFPR FAQ, verbatim "no prohibition"; illinois_licensing.md §1), which is why this scores above C. Everything else argues against doing both actively in year one. Compliance simplicity is 3: same-transaction dual roles hit a wall of unresolved rules — USDA prohibits it outright (HB-1-3555 §4.2), FHA removed its ban but lender overlays often keep it, VA is unverified, and the Reg Z 1026.36(d) compensation question is unresolved — which is why the DECISION_BRIEF's year-one rule is never both roles on one transaction. Advertising gets harder too: 38 Ill. Adm. Code 1050.940 bars putting his NMLS ID on non-mortgage listing ads without equal-prominence mortgage wording.
The killer is risk of distraction: 2. Real estate sales is a second full-time apprenticeship with its own pipeline, and every hour spent on it comes out of the mortgage learning curve. Trust also suffers (4): a brand-new person doing two jobs at once reads as uncommitted to both. The license itself is still worth holding for literacy and optionality — that is Model A/D/G behavior, not Model B.
The research delivered a clear verdict. Economics score 2: on a $350k Flex sale, GCI of ~$8,750 loses ~35% to Zillow's success fee, then 50% to the Crosstown team split, netting ~$2,594 gross per closing before costs — roughly $18–64/hr effective across scenarios once dead-lead nurture hours are counted (brokerage_zillow.md §6, splits VERIFIED, hours ESTIMATE). A first-year Flex agent closing 8–12 Zillow deals nets maybe $14k–31k pre-tax for 400–550 hours of evenings-and-weekends speed-to-lead availability.
Compliance/conflict score 2: Flex teams are measured on a "Zillow Home Loans contact rate" KPI and Flex steering is the subject of a Nov-2025 class action — structurally incompatible with being an active Ready MLO whose job is to win those same borrowers' loans. Dependence scores 1, the worst number on the board: Zillow owns the leads, the CRM data, the fee, and the performance standard. To be fair where the model earns it: speed to first closing is 7, the best in the table — lead-fed buyer conversations produce a transaction faster than anything Mitchell can build organically, and the sales-skill reps are real (skill 6). That is the honest tradeoff: C buys speed by selling independence, economics, and alignment. The DECISION_BRIEF's answer — decline Flex; if the broker license should stay active, park it at Crosstown Hybrid 75/25 (no Flex leads) or Fathom/eXp, contingent on Ready's written outside-business-activity approval — captures the only part of C worth keeping.
The strongest single model, and the primary engine. It converts Mitchell's actual advantages — finance degree, analytical patience, AI-assisted research, willingness to be measured on response time — into the thing agents visibly lack from lenders: fast, structured, honest answers on hard files. Trust scores 9 because the offer is a trust mechanism: "give me one difficult scenario or one buyer who needs a second look" asks for nothing exclusive and proves competence before asking for business. Sustainability scores 7, not 9 (rescored after red-team M3): the steady state is an inbox-and-analysis job punctuated by short calls, run in 60–90 minute sprints — a genuine personality fit — but months 1–2 are a high-social outreach ramp (sequences, dials, in-person) that no amount of Scenario Desk branding removes. The mitigations are structural (warm-first pilot entry, half-dose week 2, the energy KPI L14), not cosmetic, which is why the score is 7 and not lower. AI leverage 9 and skill-building 9: every scenario brief is simultaneously a deliverable, a product-knowledge rep, and a reusable content seed; ten briefs teach more underwriting than a month of networking. Local demand is 8 — the Harlem Ave corridor packs RE/MAX 10 (~300 agents), KW Preferred, CB Southwest, and RE/MAX Synergy into ~2 miles, so a 30-agent pilot list needs no driving.
Honest weaknesses: speed to first closing is 5 — slower than C. Scenario requests precede applications, which precede closings; the desk realistically produces its first funded loan later than a lead-fed model would. Economics is 7, not 10: revenue per hour is excellent once trust converts, but the first ~60 days are investment. Dependence is 6, not high: the desk's answers are only as good as Ready's actual IL product menu, which is [UNVERIFIED — confirm with manager], and the IHDA gap (Ready is NOT on the participating-lender list, eff. 2025-12-18) caps what he can promise DPA-dependent buyers.
The best asset-building model on the board. Differentiation 9 and defensibility 9 rest on verified white space: local mortgage SERPs are branch pages and directories; the only individual-MLO content presence found is 2017-era (Gene Mundt, New Lenox); and nobody — literally no result found — is publishing (1) town-specific IHDA/DPA-stack guides, (2) the Cook-vs-Will property-tax payment math ahead of the 2026 south-triad reassessment, or (3) condo-financing triage ahead of the 2026-08-03 Limited Review sunset that hits ~1 in 5 units of Tinley/Orland attached stock (local_market.md §5, §10). AI leverage is 10 — content research, drafting, and repurposing is the single highest-ROI AI workflow available to him. Sustainability 9: writing and small workshops beat cold-room networking for his temperament.
The tradeoffs are equally real. Speed to first application 4, first closing 3 — the worst revenue-speed profile except B's distraction case; SEO and audience compounding are quarters-long games, which is why E cannot be primary. Compliance is 6, not simple: everything published is an advertisement (NMLS IDs + nmlsconsumeraccess.org on every asset; no self-produced rate/payment content without approved assumptions; fair-lending review of geographic targeting), and the brand name itself needs Ready's approval. And the honest-positioning constraint bites: until the IHDA question is resolved, Lab content must teach the DPA stack without implying Mitchell can originate IHDA loans.
A genuinely good idea resting on an unverified foundation, which is what holds it to 79. The demand signal is real: a trades-heavy corridor (Tinley chamber 400+ members, multi-chamber network, contractor-dense Will County growth corridor), an underserved "banks say no to the self-employed" referral story, and 2–4-unit investor stock in the near-south suburbs (local_market.md §4). Differentiation 8 and defensibility 7 follow — non-QM fluency is scarce among branch-bank competitors.
But product fit scores 4: bank-statement, DSCR, and asset-qualifier products are not currently on Ready's site despite the family's impression — the IL product menu is [UNVERIFIED — confirm with manager] (ready_mortgage.md §4). Building a specialist identity on products Mitchell may not be approved to originate is the one way this model fails completely, and dependence scores 4 for the same reason. Compliance is 4 (business-purpose/occupancy-honesty exposure on DSCR; ATR care on bank-statement). Speed is slow (4/4): specialist referral networks take months, and non-QM files are harder first files than agency loans. Correct dose: a content franchise and 2–3 scenario drills a week — the F-lite experiment — not a full pivot.
G is not a separate strategy; it is the disciplined combination the master prompt asks for, and it wins because D and E cover each other's weakest criteria. D supplies what E lacks — speed to first application (7) and near-term revenue — while E supplies what D lacks — a compounding, ownable asset (defensibility 9) that survives agent churn and even a company change. The engines share one supply chain: every anonymized Scenario Desk brief becomes Lab content; every Lab reader with an agent becomes a Scenario Desk introduction. F-lite rides along at experiment dosage with a hard day-60 evaluation, capturing the trades-corridor upside without betting on an unverified product menu.
Where G honestly loses points: speed to first closing stays 5 — the hybrid inherits D's patience requirement, and anyone who wants a closing in the first 45 days would pick C and pay C's price for it. Distraction is 7, not 9 — running two engines plus an experiment is more surface area than pure D, and the plan mitigates this with fixed weekly time caps rather than pretending the risk away (see 09_primary_business_plan.md §7). Compliance is 7 (E's advertising burden dilutes D's 8). Dependence stays 6: the whole plan still runs on Ready's stale-page IL license (MB.6850275 ACTIVE per IDFPR, but apply.readyml.com still says Florida-only) and unresolved IHDA status — which is why the first five manager questions exist.
Plain-English reasoning. Mitchell's scarce resources are hours, trust, and focus. Model C spends all three buying someone else's leads at ~$2,600 a closing inside a steering conflict. Model B splits his apprenticeship in half. Models A and F are respectively too generic and too unverified to lead. Model D turns his real strengths — analysis, responsiveness, honesty about being new — into a service agents can test at zero risk, and it produces revenue-relevant activity fastest. Model E turns the byproduct of that service into an asset nobody in the footprint is building, on topics with expiration-date urgency (reassessment year, condo-rule sunset, a DPA window that reopened two days ago). Run D for this quarter's business, E for next year's, F-lite as a cheap option on the trades corridor — and defer everything else on purpose.