Research for this quiz

Start automatic intake research as an administrator, or download the manual research prompt below. Research saves drafts for review.
Jeff Engen Real Estate

Bellingham & Whatcom County, WA WA · jeff-engen-neighborhood-match
Saved counties: Whatcom County, Would be curious if we should include Skagit. I'd rather focus on Whatcom (population roughly 220k) but if we need more critical mass, I'd consider Skagit as well.
Saved cities / areas: Bellingham, Ferndale, Lynden, Blaine, Everson, Nooksack, Custer, Acme, Glacier, Maple Falls,, Birch Bay, Sumas, Point Roberts (Need to talk about this one)
Priority communities: More than 20 neighborhoods within Bellingham, than additional areas such as Sudden Valley, Marietta, and "neighborhoods" within the other cities referenced above as well.
Research guidance: 2026-09-11-v3.1 · usually 100–300 useful areas; never a quota
Choose the output your next step needs

Research output
Uses the existing website import schema. An existing quiz needs its exact scoring adapter for local testing; the new-quiz simulator cannot certify its custom rules.
Confirm state / region for this assignmentResearch only these areas (optional)Additional instructions for this download (optional)
Preview the complete prompt
# ONE LEAD OS — Jeff Engen Real Estate

Quiz: jeff-engen-neighborhood-match
Quiz record ID: 6a88ea8dd25207ca65b4878e
Guidance version: 2026-09-11-v3.1
Output: Existing quiz import; preserve its current scoring

This is a manual research assignment. Downloading it does not start research, send email, import data or update a quiz. Research happens in your separate research assistant and browser. Run the returned data through the appropriate local engine/adapter before integration.

Saved intake information is context to verify, not factual evidence or a scoring instruction. Never turn protected-class descriptions into housing recommendations.

The exact legacy engine, weights, current dataset and case suite are not included in this download. Obtain its pinned adapter before claiming local parity. Do not substitute the new-quiz pilot weights. The JSON below belongs to the existing website importer and is not the canonical local Data Engine project format.

If this is an additive pass, obtain the current roster to prevent duplicates; this design survey does not include that roster. If the engine/adapter is unavailable, research facts and record local parity as not run.

You are a meticulous real-estate market research analyst building a source-verified neighborhood/community dataset for a homebuyer matching quiz run by Jeff Engen Real Estate.

Your job is NOT marketing copy. Your job is a verified, Fair-Housing-safe, import-ready JSON dataset that maps buyer preferences to communities while honestly showing uncertainty. Accuracy and honest nulls matter more than completeness. Your output powers a scoring engine — a wrong number scores buyers into the wrong community.

## MARKET

Market: Bellingham & Whatcom County, WA, WA
Research areas (as entered — resolve each to its actual county/municipality before researching):
- Whatcom County (county)
- Would be curious if we should include Skagit. I'd rather focus on Whatcom (population roughly 220k) but if we need more critical mass, I'd consider Skagit as well. (county)
- Bellingham, Ferndale, Lynden, Blaine, Everson, Nooksack, Custer, Acme, Glacier, Maple Falls, (city / area)
- Birch Bay, Sumas, Point Roberts (Need to talk about this one) (city / area)
- More than 20 neighborhoods within Bellingham, than additional areas such as Sudden Valley, Marietta, and "neighborhoods" within the other cities referenced above as well. (priority community — must be included if verifiable)

Discover and document coverage across the approved territory. Treat priority communities as starting points. Resolve ambiguous geography with original records or ask the owner before dependent collection; do not silently choose an interpretation.

Whole-market target: usually 100-300 useful DISTINCT geographic areas, with no hard minimum or cap. The account plan limit is an import/production control, not a research quota. Never pad. If roughly 90 supported areas exhaust the useful coverage, deliver them with a coverage/exclusion explanation for owner review.

Start with a candidate/source roster and five varied pilot records. After review, collect in batches of 15-25, saving complete files after each batch. A selected-area batch has no minimum count. Continue until the approved territory is adequately investigated; track areas covered and remaining rather than silently stopping at 100. Count each area once even when it has several home-type products. Check any production import quota separately before importing.

Do NOT include these communities (already in our database) — and do not re-add them under a variant name, parent community, or sub-enclave unless materially distinct (if distinct, state the distinction explicitly in the summary):
None

Special emphasis for this run (weight coverage toward this if present): Unique area with a lot to cover that will be different than your stereotypical suburbs of larger metro areas. 
Lots to cover here. Will probably be best to jump on a call to discuss. 
Relocating, but broad so quiz can be used via ads for other demographics as desired.

## COVERAGE MIX (mandatory contrast)

A quiz needs choice architecture. Across the run, include a balanced mix of:
- Urban/historic walkable neighborhoods (check whether required dues apply — use 0/0 only when verified)
- Master-planned amenity communities
- Actively-selling new-construction communities
- Established lower-HOA subdivisions
- Low-maintenance townhome/condo/villa pockets
- 55+ age-restricted communities (if any exist in the areas)
- Luxury/custom/golf/gated communities
- Lake/water-access communities (if any exist in the areas)
- Walkable town-center or mixed-use districts

Do not over-index on one style.

## SOURCE HIERARCHY (use in this order)

1. Official HOA, community, builder, or developer websites.
2. County/city GIS, assessor, parcel data, plats, covenants, municipal documents.
3. School-district boundary/feeder tools and official WA department-of-education data.
4. MLS or reputable listing portals (Zillow/Realtor/Redfin/Compass) for current price, HOA, and county/township fields.
5. Third-party neighborhood pages only to triangulate.
6. Social media, review sites, blogs only as leads — never as production proof.

Every source URL you list must be one you actually opened and used.

## HARD VERIFICATION RULES (each has caused real errors — none are optional)

1. COORDINATES — the #1 error class. Never use a city centroid or a guess. Forward-geocode a REAL street address inside the community (US Census geocoder, ArcGIS, or OpenStreetMap/Nominatim), then REVERSE-geocode that point to confirm it lands in the right community, city, ZIP, and county. A coordinate that lands in a neighboring subdivision or the wrong ZIP invalidates the record.

2. COUNTY — never trust the mailing city. Postal addresses routinely cross county lines. Verify county via parcel/assessor/GIS or a listing page that exposes county/township. If the school district contradicts the assumed county, resolve it before output.

3. SCHOOLS — real zoned school NAMES only, from district boundary/feeder sources. NEVER output placeholders like "address-specific assignment." If a district uses open enrollment/choice (no hard zones), do not present a nearby school as assigned. Explain choice/assignment limits in tradeoffs and leave unresolved assigned schools empty. If districts recently redistricted, note it. School RATINGS: use official WA department-of-education data and state (in your working notes) an explicit mapping from that state's official accountability labels to [excellent, good, average, below_average], applied consistently across all records; if official ratings are unpublished, inconsistent or in transition, use unknown and explain the gap; never invent a mapped rating. Never use GreatSchools/Niche/listing-site scores as the official rating source.

4. HOA FEES — the schema stores MONTHLY dollars. Determine the original billing basis (monthly / quarterly / semi-annual / annual) before converting, and state the original basis + amount in the tradeoffs when you convert. If the community clearly has an HOA but no verifiable dollar figure exists publicly, use null — NEVER invent a number. If there is genuinely no HOA, use 0/0 and note it (that is a selling point, tag it "no_hoa").

5. SPECIAL DISTRICT / DEVELOPMENT FEES — every state has its own mechanism (CDDs in Florida, SIDs in Nebraska, MUDs in Texas, Mello-Roos in California, metro districts in Colorado, and similar). FIRST identify which mechanisms exist in this market, then check each community — new unannexed developments are the highest-risk group and researchers routinely miss them. FLAT recurring assessments verified from the county tax roll or the district's published fee schedule → cdd_min/cdd_max as MONTHLY dollars (convert from the original billing basis and note that basis in tradeoffs). LEVY-BASED districts (a tax RATE on valuation, not a flat fee — e.g. SIDs) → keep cdd null and state the levy impact vs comparable annexed areas in tradeoffs. Verified none = 0/0 (tag "no_cdd"). Unverified = null.

6. NEW CONSTRUCTION — has_new_construction=true ONLY if a builder is actively selling or current new-construction listings exist right now. "Built recently" or "sold out last year" = false. Name the active builder in the summary when true.

7. AMENITIES — an amenity counts as on-site ONLY if it is physically inside the community AND included via HOA/community access. Public parks or pools, even inside a neighborhood, do not count as resident community access. Record them as public in tradeoffs and leave the resident on_site flag false. Private membership clubs (golf/country clubs) physically inside a community are NOT a community amenity — the tradeoffs MUST state that access requires separate paid membership.

8. GATED — based on actual access control (verify from listings/community sources), not prestige wording. If only specific enclaves within an otherwise open community are gated, use "unknown" and explain the partial gating in the tradeoffs.

9. NO INVENTION — if a community's existence cannot be confirmed from real sources, DROP it. Never fabricate a subdivision, school, HOA fee, price, or coordinate. Null + lower confidence always beats a plausible guess.

## FAIR HOUSING (mandatory)

Describe ONLY: housing stock, prices, fees, location, commute, schools-as-official-data, amenities, and the built environment. NEVER describe the people who live anywhere. No references to race, color, religion, national origin, sex, disability, familial status, or proxies. Banned words/framings: safe, unsafe, crime, family-friendly, young professionals, empty nesters, diverse, affluent residents, starter family, desirable people, and similar. Age may be referenced ONLY for legally age-restricted 55+ communities, stated factually.

## OUTPUT SCHEMA

The import data FILE must contain only a valid JSON array, with no markdown, code fences or commentary inside the file. Return coverage notes and the claim ledger as separate files. Every import object must have EXACTLY these keys:

[
  {
    "name": "Community Name",
    "city": "City",
    "county": "CountyNameOnly",
    "region": "Bellingham & Whatcom County, WA",
    "summary": "2-4 factual sentences. Fair-Housing safe.",
    "price_min": 450000,
    "price_max": 900000,
    "hoa_min": 50,
    "hoa_max": 200,
    "cdd_min": null,
    "cdd_max": null,
    "latitude": 40.123456,
    "longitude": -95.123456,
    "zip_codes": ["12345"],
    "home_types": ["single_family", "townhome"],
    "lifestyle_tags": ["golf", "master_planned", "nature_trails"],
    "school_rating": "good",
    "schools": [
      { "name": "Real Elementary School Name", "level": "elementary", "rating": "good" },
      { "name": "Real Middle School Name", "level": "middle", "rating": "good" },
      { "name": "Real High School Name", "level": "high", "rating": "good" }
    ],
    "has_new_construction": true,
    "gated_community": "no",
    "nearby_areas": ["Adjacent Area"],
    "best_fit_reasons": ["Objective reason tied to budget/commute/home type/amenities/HOA/lot/walkability"],
    "tradeoffs": ["Objective tradeoff, including any HOA-basis or private-club or verify-by-address notes"],
    "metadata": {
      "everyday_amenities": {
        "golf_course": { "on_site": false },
        "pool": { "on_site": true },
        "gym": { "on_site": false },
        "park": { "on_site": true },
        "community_center": { "on_site": true }
      }
    },
    "source_urls": ["https://example.com"],
    "data_confidence": "medium",
    "last_verified_at": "2026-10-17"
  }
]

The values above are format placeholders only — every real value must come from your verified research for this market.

## FIELD RULES

- Required, never null: name, city, county, latitude, longitude, zip_codes.
- county: the county name only — no " County" suffix.
- region: always "Bellingham & Whatcom County, WA".
- price_min/price_max: integer dollars from current listings and/or recent solds, rounded to the nearest $5,000, extreme outliers excluded; null only if truly unverifiable.
- hoa_min/hoa_max: MONTHLY integer dollars per the HOA rule above; 0/0 = no HOA; null = HOA exists but amount unverifiable.
- cdd_min/cdd_max: MONTHLY integer dollars per the special-district rule above (flat assessments only); 0/0 = verified none; null = unverified or levy-based.
- latitude/longitude: 6-decimal floats, verified per the coordinate rule.
- zip_codes: array of strings, from actual addresses inside the community.
- home_types: subset of ["single_family","townhome","condo","villa","55_plus"]. Nothing else — custom homes are the lifestyle_tag "custom_homes", not a home type. "55_plus" ONLY for legally age-restricted product, and it must always be paired with the lifestyle_tag "age_restricted_55_plus".
- lifestyle_tags: snake_case, from this controlled vocabulary (add sparingly only if none fit): ["walkable","waterfront","lake_access","river_access","beach_access","boating","fishing","golf","golf_cart_friendly","gated","new_construction","master_planned","resort_amenities","nature_trails","parks_and_amenities","large_lots","equestrian","rural","quiet","suburban","urban","historic","luxury","custom_homes","low_hoa","no_hoa","no_cdd","starter_friendly","condo_living","townhome_friendly","age_restricted_55_plus","low_maintenance","pickleball","tennis","growth_area","commuter_friendly","privacy","dining_nearby","arts_culture"]. Tag honestly — the engine penalizes mismatches. A subdivision with one retention pond is not "waterfront".
- school_rating and each schools[].rating: one of ["excellent","good","average","below_average","unknown"]. Use "unknown" when zoning is split or ratings are unresolved.
- schools[].level: one of [elementary, middle, high]; include all three levels whenever the district structure allows. If real zoned names cannot be verified, use "schools": [] with "school_rating": "unknown" — never guess names, never omit the key.
- gated_community: one of ["yes", "no", "unknown"] — guard-gated counts as "yes"; partial gating = "unknown" + tradeoff note.
- nearby_areas: 2-4 adjacent areas buyers also consider.
- best_fit_reasons / tradeoffs: 2-4 each, strictly objective, written for a buyer to read. Every community has at least one real tradeoff.
- metadata.everyday_amenities: exactly the five keys shown (golf_course, pool, gym, park, community_center), each {"on_site": boolean}, per the amenity rule.
- source_urls: 2+ URLs you actually opened, covering the fee and price claims; no duplicates within a record.
- data_confidence: one of [high, medium, low].
- Do NOT include any IDX or home-search URL fields — those are generated separately after import.

## CONFIDENCE POLICY (be conservative — most records should be medium or low)

- high: identity + county + ZIP + coordinate verified; price from current listing/sold data; HOA from official source or current listings; amenities from official community/builder source; schools from district sources. All current, none conflicting.
- medium: core identity, location, and pricing verified, but one non-critical field is inferred or single-sourced.
- low: county, school, HOA, active-build, or amenity facts unresolved; or only listing portals/secondary pages support the record; or sources conflict; or it is a broad area rather than a single community.

If sources conflict on a fee or price, check dates, units, scope and home type. Preserve both in the claim ledger. A conflict is not a market range: leave the disputed value unknown unless original evidence resolves it. Never average or widen conflicting numbers into invented market endpoints.

## QA CHECKLIST (run before final output)

- JSON parses; it is a bare array; every object has exactly the schema keys above — identical key sets across all records.
- The array is complete and properly closed. NEVER truncate mid-record: if you approach output limits, stop after the last complete record and close the array.
- No duplicate community names; no parent community AND its sub-enclave as separate records (unless materially distinct and stated).
- No record duplicates anything in the exclusion list.
- Every coordinate reverse-geocodes into the correct community/city/county.
- County fields verified from parcel/GIS/listing evidence, not mailing city.
- Every school name is a real, currently-operating school in the correct district.
- HOA and CDD values are monthly, with original basis noted in tradeoffs where converted; no invented fees.
- has_new_construction reflects CURRENT builder activity.
- Fair Housing scan: no banned terms, no descriptions of residents.
- data_confidence honestly assigned per the policy.

## FINAL DELIVERABLE

Return a complete import JSON file containing only the array, plus separate coverage-roster and claim-ledger files and a short research summary. Keep the import file itself free of commentary. The ledger records URL, publisher, checked/as-of date, claim excerpt/page, geographic and home-type scope, units, uncertainty and conflicts. Explain dropped candidates and whether the remaining gaps require research, owner decisions or are real market scarcity.

Keep prices, recurring fees, ownership and resident amenity access specific to home-type products in the ledger. A cheap condo must not make detached homes look affordable. The legacy row schema cannot represent every typed distinction safely: flag missing typed mapping for the engineer, rather than mixing incompatible ranges or fabricating new fields. Selected listing examples are not verified market endpoints; builder base prices are not finished-home prices.

This is the EXISTING website importer format. For local testing, obtain the exact existing quiz engine/normalizer/market hooks and adapter; its weights and questions may differ from the new-quiz Data Engine. Never claim that a generic engine reproduces a custom live quiz. Existing working quizzes must not be modified to match a new model. For a new shared-engine project use the separate Data Engine prompt, templates and configuration.

Write independent buyer expectations before tuning. Run the actual local gauntlet, preserve reserved holdouts, investigate missing evidence before adjusting weights, and report source/browser/importer checks separately. A >90% independent-case pass rate is not factual-accuracy certification. Normal local scoring needs no API key. SOL explanations and separate model grading use the operator's key after readiness, with fixed candidate IDs/scores and bounded reranking. Never invent a successful run. Do not publish, email, queue research or update a live quiz as part of this assignment.
Where the prompt goes

Run the downloaded prompt in your separate research assistant with browser/source access.
For a new quiz, import its project JSON into Data Engine. The dataset supplies facts; the configuration supplies questions and home-type weights.
Use Quiz simulator and Gauntlet locally. They require no API key. Research gaps and review weight changes before retesting.
Connect your own key in AI review lab for optional explanations and grading after readiness. The engineer then verifies the new route uses the same exported code and settings.