Complete strategy.md - fix truncation, add renter transparency layer and regional intelligence sections

rental-platform/strategy.md

@@ -18,4 +18,114 @@ The model is trained on existing landlord-reported failure outcomes working back
 - Academic housing research: pre-cleaned and labeled datasets
 
 **What the model predicts:**
-Not just "will this person pay rent" — it identifies which specific risk factors contributed to failure in similar profiles. The output is actionable reasoning
+Not just "will this person pay rent" — it identifies which specific risk factors contributed to failure in similar profiles. The output is actionable reasoning, not just a score. Example: "This applicant has three signals associated with month 2-3 payment gaps in similar profiles."
+
+**Signals to correlate:**
+- Income instability patterns
+- Application behavior (how a form is completed is itself a signal)
+- Debt-to-income trajectory, not just snapshot
+- Employment sector risk (gig vs salaried vs seasonal)
+- Geographic cost-of-living mismatch
+
+**Routing logic:**
+The model routes, it does not simply reject. High confidence → straight through. Medium confidence → additional verification steps. Low confidence → flagged before wasting anyone's time.
+
+**Sequencing:**
+1. Rules-based vetting engine first
+2. Accumulate outcome data as renters move through the system
+3. Train prediction model on real outcomes combined with reverse-trained historical data
+4. Prediction layer gates entry before vetting starts
+
+**Legal consideration (critical):**
+The Fair Housing Act applies. The model cannot use race, national origin, familial status, religion, sex, or disability — even indirectly through proxy variables such as neighborhood or zip code. The feature set must be built around financial behavior and verifiable circumstances only. A housing attorney must review the feature set before training begins.
+
+---
+
+## Regional Intelligence Layer
+
+The model segments by state and region. Rental market dynamics vary dramatically by geography and a model trained on undifferentiated national data introduces noise that makes predictions less useful and less trustworthy.
+
+**Why regional segmentation matters:**
+A missed payment in Indianapolis signals something different than one in Miami. Cost of living, average rent as a percentage of local income, local employment volatility, eviction law timelines, and seasonal employment patterns all vary by region. A gig worker in Austin faces different income stability risk than one in Cleveland. Eviction law alone is a major variable — Texas landlords can move through eviction in roughly 21 days, New York can take 6 months or more. The risk calculus for a landlord is completely different depending on where the property sits.
+
+**What regional layering enables:**
+Landlords log in and their properties are mapped to specific markets. Every renter profile they review is scored against the context of the property being applied for, not a generic national benchmark. A renter who looks borderline nationally may look strong in a low-cost Midwest market or genuinely risky in a high-cost coastal one.
+
+**Regional public data sources:**
+- HUD fair market rent data by metro area (updated annually)
+- Census Bureau ACS — income, employment, and housing cost burden by region
+- State court records — eviction filing rates and outcomes by county
+- BLS regional employment volatility by sector
+- CFPB regional debt and delinquency patterns
+
+**The geographic network effect:**
+As more landlords use the platform across different regions, every outcome — paid on time, missed payment, early exit — feeds back into the regional model and tightens predictions over time. The platform gets smarter in each market the more it operates there. Geographic expansion is both a growth strategy and a model improvement strategy.
+
+**The product framing this enables:**
+This positions the platform less as a rental matchmaking tool and more as a regional risk intelligence layer for the residential rental market — a meaningfully stronger and more defensible long-term product description.
+
+---
+
+## Renter-Facing Transparency Layer
+
+The model and vetting process should not be a black box to the renter. The renter is a paying member, not just a subject being evaluated. Transparency into what the process looks at — and what they can do to strengthen their profile — is a core part of the value proposition that separates this from a standard background check.
+
+This also solves a real problem young renters face: they don't know why they get rejected, and they don't know what to fix. A score with no explanation is useless. A score with actionable context is something they can act on.
+
+### Before Vetting: Readiness Assessment
+Before a renter begins the formal vetting process, they are shown a pre-check that explains:
+- What documents and information will be required
+- What each dimension of the vetting process looks at and why
+- What "ready" looks like for their specific situation and target market
+- What common gaps look like for renters their age and income level, and how to address them ahead of time
+
+The goal is to reduce abandonment during vetting. If someone hits a step they're not ready for with no context, they drop off. If they understand what's being asked and why, they either complete it or return once they've addressed the gap.
+
+### During Vetting: Live Progress Visibility
+As the renter moves through the vetting pipeline they see:
+- Which dimensions have been verified
+- Which are still pending and what's needed to complete them
+- Where they stand relative to the threshold for their target market
+- Specific guidance if a step requires additional documentation or follow-up
+
+### After Vetting: Profile Breakdown
+The renter receives more than a badge. They receive a readable breakdown of their profile:
+- Which factors contributed positively to their approval
+- Which factors were borderline and what they mean in context
+- What their profile looks like to a landlord in their target region specifically
+- What, if anything, they could do to strengthen their standing over time
+
+This is not a rejection notice with no path forward. Even renters who do not pass initial vetting should leave with a clear understanding of what changed and what to do about it.
+
+### Ongoing: Profile Monitoring and Proactive Guidance
+After approval, the platform monitors for material changes to the renter's profile:
+- Income changes (job loss, new employment, sector shift)
+- Debt or financial behavior changes
+- Lease history as it accumulates
+- Market changes in their target region
+
+If something shifts that affects their standing, they are notified proactively — not surprised at the next renewal or application. They are also shown what the change means in plain terms and whether any action is needed.
+
+### Why This Strengthens the Business Model
+- Reduces dropout during vetting → more completed members → larger pool for landlords
+- Builds genuine trust with renters → better retention, word of mouth
+- Creates a feedback loop where renters improve their profiles over time → pool quality improves naturally
+- Differentiates from every competitor, all of whom treat the renter as a passive subject rather than an active participant
+- Supports the mission framing: this platform is for renters, not just a landlord tool that renters fund
+
+---
+
+## Open Questions
+
+- [ ] Name / domain availability
+- [ ] Which city to pilot first
+- [ ] Co-founder or solo?
+- [ ] Vetting criteria definition — what exactly gets checked, in what order
+- [ ] Landlord acquisition strategy for cold start (need landlords before renters have somewhere to go)
+- [ ] Legal structure for nonprofit arm
+- [ ] Membership fee pricing model
+- [ ] Pursue hybrid tech infrastructure angle or keep pure platform play?
+- [ ] BiggerPockets as distribution/partnership target — cold outreach timing?
+- [ ] ML model feature set legal review — when to engage housing attorney?
+- [ ] Reverse training data collection strategy — scraping vs API access?
+- [ ] Renter readiness assessment format — checklist, guided flow, or conversational?