FLIP ENGINE

Surface Data Engine: ingestion + multi-API aggregation. Listing URL parsing (Zillow __NEXT_DATA__ JSON, MLS Bridge in Phase 3), property records (ATTOM-first, Estated absorbed into ATTOM so this is forward-compatible), comp pulls.

Diagnostic Engine: CV + NLP condition inference. Listing-photo analysis for build / design quality and interior / exterior style, listing-text NLP for condition signals, lead paint compliance flags for pre-1978 listings (new EPA standard effective Jan 12, 2026).

Deterministic Engine: cross-analysis synthesis + scoring. ARV (3-6 comps, 0.5mi, 90-180d, weighted top 30-40%, drop high / low for outliers, never adjust >25% of comp), rehab band (Light $15-25/sqft, Medium $50-75, Heavy $90-135+, 15% contingency standard), profit math, risk flag composition, headline verdict.

Final values are consistency-weighted composites of what most APIs agree on, not single-source trust. Each datapoint carries source reliability + recency + variance scores feeding a confidence model. When sources disagree, the disagreement itself is surfaced rather than buried in a single number.

Property-singularity guard: production-breaking bug if engine pulls data from multiple properties into one analysis. Cross-API address / coord matching is mandatory at the ingestion seam.

Low / medium / high applied uniformly across cost bands, build / design quality, and interior / exterior style. Vincent's construction background is the IP here: siding choices, interior finishes, build details map to income brackets and feed the CV inference.

Pinned to UAD 3.6 (Uniform Appraisal Dataset, industry mandate landing in 2026) standards: Fannie-Mae C1-C6 (condition) + Q1-Q6 (quality). The pinning is deliberate. Riding the industry credibility wave instead of inventing a parallel vocabulary that has to be re-translated for every conversation with an appraiser, lender, or seasoned flipper.

Phase 1 (current): paste-listing analyze flow, 1-2 free / low-cost APIs, basic ARV, basic rehab, basic flip score, coming-soon tabs for unbuilt layers. Stripe billing, Supabase auth and storage, Anthropic Claude API for vision and NLP (UAD 3.6 definitions encoded in prompt; output schema preserved so a Restb.ai swap-in stays trivial when revenue justifies).

Phase 2: full CV image analysis, refined confidence scoring, UX polish. Phase 3: multi-API aggregation engine fully online (paid ATTOM / RentCast tier), browser extension as the Zillow-bypass unlock, MLS Bridge integration once gating clears.

Pricing $49 (Solo) / $99 (Pro) sits cleanly under the entry tiers of incumbent bundled tools. The whitespace was confirmed in research: no incumbent leads with paste-listing instant-verdict UX.

Decision-clarity UX moat: headline verdict (Strong / Moderate / Avoid) on top, supporting numbers below. The user sees the answer before they see the math. The math is there for the investors who want to challenge the answer, not as the gating ritual before the answer is revealed.

Tone is plain, non-flowery. Neighborhood summaries state pros and cons clearly. Risk flags must be real, not theatrical. An engine that cries wolf eight times per listing trains the user to ignore it on the ninth, when it matters.