TTAS — Data Lineage

Each of the 26 columns in the TTAS manifold is classified into one of five data natures. This page documents where every value comes from, how it was transformed, and the confidence level of its provenance.

Data Sources

FRED (Federal Reserve Economic Data)

Endpoint: api.stlouisfed.org/fred/series/observations

Pulled: 2026-05-14

Nature: Observed

Confidence: High (0.90–0.95)

953 rows, 1947–2026. Mortgage rate series is direct; CPI and unemployment are national-level context variables.

Fields: mortgage_rate_30y, cpi_all_urban, unemployment_rate, median_sales_price_us

Census ACS 5-Year Estimates

Endpoint: api.census.gov/data/2023/acs/acs5

Pulled: 2026-05-14

Nature: Observed

Confidence: Medium-High (0.85)

24 Tulsa ZIPs returned with null metric values. ACS 5-year estimates have 1-year lag; 2023 estimates reflect 2019–2023 data.

Fields: median_household_income, owner_occupied_share, median_home_value, rent_burden_count

Realtor.com Research Data

Endpoint: realtor.com/research/data (CSV download, not API)

Pulled: 2026-05-14

Nature: Observed (aggregates) / Calibrated (per-property)

Confidence: Medium-High (0.70–0.85)

Tulsa metro data, monthly. Listing prices and rents are observed; per-property values are calibrated synthetic. Data extends 2016–2026; pre-2016 values in timeseries are interpolated.

Fields: median_listing_price, median_rent, active_listing_count, median_days_on_market

OSMnx Street Network + POI

Endpoint: OpenStreetMap via osmnx Python library (on-demand)

Pulled: n/a (opt-in)

Nature: Observed (when enabled) / Synthetic (default)

Confidence: Medium (0.55) when enabled; Low (0.40) when synthetic

When enabled, replaces synthetic centrality and amenity with real OSM data. Not enabled by default due to network dependency.

Fields: street_centrality, amenity_density

Tulsa County Assessor / City of Tulsa Open Data

Endpoint: planned — not yet integrated

Pulled: n/a

Nature: Planned (Observed)

Confidence: n/a

Would provide parcel-level ground truth for property values and transaction dates. Currently not integrated.

Fields: parcel_values, property_class, sale_dates

Per-Column Provenance

Column	Nature	Primary Source	Confidence	Transformations
median_listing_price	Calibrated	Realtor.com Research Data	70%	synthetic_generation → real_aggregate_calibration → monthly_median_scaling_to_realtor_com
inventory_velocity	Calibrated	Realtor.com Research Data (days on market)	70%	synthetic_generation → dom_to_velocity_conversion → real_aggregate_calibration
monthly_rent_estimate	Calibrated	Realtor.com Research Data (rent) + synthetic rent-to-price model	65%	synthetic_rent_to_price_ratio → real_aggregate_calibration → monthly_median_scaling
annual_income_estimate	Calibrated	Census ACS (ZIP median income) + synthetic per-property noise	65%	synthetic_generation → zip_income_calibration → property_noise
ownership_cost_monthly	Derived	Computed from median_listing_price, mortgage_rate_30y, property_tax_rate	80%	mortgage_annuity_formula → tax_and_insurance_estimate
rent_margin_monthly	Derived	Computed from annual_income_estimate, dti_max, monthly_rent_estimate	75%	budget_less_rent
buy_margin_monthly	Derived	Computed from annual_income_estimate, dti_max, ownership_cost_monthly	75%	budget_less_ownership_cost
affordability_index	Derived	Computed in preprocess.py — ownership_cost / max_affordable_payment	75%	affordability_gap_normalization
rent_to_price_ratio	Derived	Computed from monthly_rent_estimate / median_listing_price	70%	annualized_rent_to_price_ratio
rent_vs_buy	Modeled	Decision rule from buy_margin, rent_margin, opportunity score	60%	threshold_rule → buy_margin_gt_-0.10*rent → opportunity_gt_0.42
regime_hint	Modeled	Deterministic regime classifier based on month number	55%	month_index_threshold: <24 Stable, <57 Overheated, <72 Rate Shock, >=72 Opportunity
mortgage_rate_30y	Observed	FRED MORTGAGE30US	95%	api_fetch → monthly_resample → interpolate_gaps
cpi_all_urban	Observed	FRED CPIAUCSL	95%	api_fetch → monthly_resample
unemployment_rate	Observed	FRED UNRATE	95%	api_fetch → monthly_resample
median_sales_price_us	Observed	FRED MSPUS	90%	api_fetch → quarterly_to_monthly_interpolation
median_household_income	Observed	Census ACS 5-Year Estimates (Table S1901)	85%	api_fetch → zip_code_join
owner_occupied_share	Observed	Census ACS 5-Year Estimates (Table S2502)	85%	api_fetch → zip_code_join
median_home_value	Observed	Census ACS 5-Year Estimates (Table S2502)	85%	api_fetch → zip_code_join
school_rating	Synthetic	Tulsa ZIP profiles (hardcoded anchors in fetch_data.py)	40%	profile_lookup → micro_location_noise → clip_0_1
street_centrality	Synthetic	Tulsa ZIP profiles (hardcoded anchors)	40%	profile_lookup → noise → clip_0_1
amenity_density	Synthetic	Tulsa ZIP profiles (hardcoded anchors)	40%	profile_lookup → seasonal_adjustment → noise → clip_0_1
crime_index	Synthetic	Tulsa ZIP profiles (hardcoded anchors)	35%	profile_lookup → trend_adjustment → noise → clip_0_1
flood_risk_score	Synthetic	Tulsa ZIP profiles (hardcoded anchors)	35%	profile_lookup → seasonal_cos → noise → clip_0_1
walk_transit_score	Synthetic	Computed from synthetic walk and transit components	35%	weighted_combination → clip_0_1
economic_mobility_index	Synthetic	Tulsa ZIP profiles (hardcoded anchors)	35%	profile_lookup → trend_adjustment → noise → clip_0_1
dti_max	Synthetic	Beta distribution draw (a=2.2, b=3.0), scaled to [0.28, 0.47]	30%	beta_sample → scale_to_range → clip