Explore markets ODDYSSEY / PUBLIC-DATA FRAMEWORK
A clear record of where the evidence comes from, what we extract, which metros enter the comparison, and how the screening score is built.
Public-data market screening and hypothetical fleet operations. Not a safety assessment or deployment approval.
SOURCE REGISTER
The verified.v2 public-data release links every available measurement to its saved evidence record. It records the source, period, geography, transformation, assumptions, retrieval time, and raw-file hash so a number can be traced back to its input.
2024-acs5__2024-tiger__NOAA-1991-2020__TIGER-edges-2024__AFDC-current__commute-B08013-B08303-r2e1e3650e5a6a13ca8a488533f5fec66352797963cd41a23430642bac6ad0ff0a50569449d013a9445df6Population, households by vehicle availability, commute mode, and commute duration. The corrected commute measure uses aggregate travel minutes divided by workers who did not work from home; B08301 remains the transit-share universe. Both commute inputs and their margins of error are recorded for all 35 candidate and reference CBSAs. Input MOEs are not confidence intervals for the derived ratio or score.
Census ACS dataCBSA and county polygons define the comparison area. Whole member-county land area uses Census ALAND; counties are assigned by representative point. TIGER network edges supply public road centerline length, road classes, and junction topology. Walkways, trails, private resource roads, and parking-lot roads are excluded. City-proper boundaries are not used for ranking.
2024 Census CBSA geographyPrecipitation and snowfall use the nearest reporting station within 100 km. Hot-day counts average up to three qualifying stations in that radius. The pipeline requests standard units and quality attributes, records station IDs and distances, and leaves unavailable snowfall missing.
NOAA climate normalsFilter electric-fuel (ELEC), public-access stations with operational status E; count reported DC fast ports, deduplicate station IDs, and join coordinates to frozen geography. Missing DC counts invalidate the feature. Public charging does not measure private depot access or secured fleet capacity.
AFDC data downloadsDated public operator evidence defines the reference-market cohort and its recorded category. The verified.v2 release contains 15 enabled commercial references. This public snapshot is not evidence of private fleet availability, internal strategy, or service coverage beyond what the source states.
Waymo public ride marketsThis bundled map layer provides visual context. It is not the scoring geography and is not used for CBSA or county joins. Upstream attribution is preserved.
PublicaMundi source fileChoose a candidate or public reference market to inspect the release’s stored measurements and complete source records. This page does not recalculate rankings.
Exact feature values, missing reasons, and full provenance records appear here after you choose a metro.
EXTRACTION REGISTRY
The active ranking.v2 model has 15 required variables, grouped into three pillars. We show the source and extraction rule beside the exact registry key, transformation, and within-pillar weight.
Public climate, commute, and road-network proxies. Familiarity compares a candidate with a whole reference environment.
| Variable | Unit | Extraction | Source | Weight | Transform |
|---|---|---|---|---|---|
Annual precipitationannual_precipitation_mm | mm/year | Nearest qualifying NOAA 1991–2020 station within 100 km; convert inches to millimeters. | NOAA NCEI | 10% | log1p |
Annual snowfallannual_snowfall_mm | mm/year | Nearest NOAA station reporting snowfall within 100 km; convert inches to millimeters. Missing is not zero. | NOAA NCEI | 10% | log1p |
Days at or above 90°Fhot_days_32c | days/year | Mean annual days with maximum temperature ≥ 90°F (32.22°C), averaged across up to three qualifying stations within 100 km. | NOAA NCEI | 10% | linear |
Mean commutemean_commute_minutes | minutes | B08013_E001 aggregate travel minutes ÷ B08303_E001 workers not working from home. | 2024 ACS | 20% | linear |
Road densityroad_density_km_per_km2 | km/km² | Included TIGER road centerline kilometers ÷ CBSA land area in km². | 2024 TIGER/Line | 10% | log1p |
Intersection densityintersection_density_per_km2 | intersections/km² | Consolidated degree-three-or-higher road junctions ÷ CBSA land area in km². Nearby divided-road junctions are clustered within 20 m. | 2024 TIGER/Line | 10% | log1p |
Freeway sharefreeway_share | fraction | Length of Census road classes S1100 and S1630 ÷ included road length. | 2024 TIGER/Line | 10% | linear |
Arterial sharearterial_share | fraction | Length of Census road class S1200 ÷ included road length. | 2024 TIGER/Line | 10% | linear |
Local-road sharelocal_road_share | fraction | Length of classes S1400, S1640, and S1730 ÷ included road length. A Census road-class grouping, not an FHWA functional class. | 2024 TIGER/Line | 10% | linear |
Public charging quantity and coarse county coverage. Neither measure establishes a private depot or technical readiness.
| Variable | Unit | Extraction | Source | Weight | Transform |
|---|---|---|---|---|---|
Public DC ports per 100,000 peoplepublic_dc_ports_per_100k | ports/100,000 people | Deduplicated operational public DC fast ports ÷ CBSA population × 100,000. | AFDC / NLR + ACS | 70% | log1p |
Population in counties with public DCpopulation_share_in_counties_with_dc | fraction of CBSA population | Population in member counties with at least one operational public DC port ÷ CBSA population. | AFDC / NLR + ACS | 30% | linear |
Scale and transportation context. These variables are not observed ride-hailing demand.
| Variable | Unit | Extraction | Source | Weight | Transform |
|---|---|---|---|---|---|
Resident populationpopulation | persons | 2024 ACS B01003_E001 published CBSA estimate. | 2024 ACS | 30% | log1p |
Population densitypopulation_density_per_km2 | persons/km² | Resident population ÷ CBSA land area in km²; water area excluded. | ACS + 2024 TIGER/Line | 25% | log1p |
Zero-vehicle household sharezero_vehicle_household_share | fraction of households | 2024 ACS B08201_E002 households with no vehicle ÷ B08201_E001 total households. | 2024 ACS | 30% | linear |
Transit commute sharetransit_commute_share | fraction of workers 16+ | 2024 ACS B08301_E010 public-transit commuters ÷ B08301_E001 workers in the commute-mode universe. | 2024 ACS | 15% | linear |
COMPARISON COHORT
Each named market resolves to an official Census CBSA, keeping the whole regional geography consistent across Census, charging, and road inputs.
The 20 candidate metros are the project’s configured target expansion cohort. The repository does not record a city-by-city inclusion rationale or a quantitative nationwide selection cutoff, so this set should not be read as a representative U.S. sample or as the highest-scoring metros nationwide. CBSA-level comparison was selected because metro geographies capture regional travel and align with the source data better than city-proper boundaries.
Configured candidates evaluated by the screening model.
Enabled commercial reference metros used for normalization and familiarity comparisons.
The 15 reference metros belong to the normalization cohort; they are not candidates in the expansion ranking. Reference status reflects dated public evidence and does not establish service boundaries or fleet availability.
SCORING METHOD
The engine is deterministic and release-based. It uses only the configured public measurements and explicit model assumptions.
Each result is bound to one immutable data release and one model version. The verified.v2 normalization cohort includes all 20 configured candidates and 15 enabled, complete reference metros. No city outside that set contributes to the bounds.
Apply the registry transform first: linear values stay unchanged; log1p requires a finite, nonnegative value. Normalize against the release’s frozen minimum and maximum, then clip to [0, 1]. A constant feature is removed globally for that release. Bounds never refit when filters, reference choices, or weights change.
Familiarity measures weighted Euclidean distance across the nine normalized ODD features to each eligible complete reference vector; the nearest whole reference is selected, with similarity = 100 × (1 − distance). Readiness and Opportunity are weighted combinations of their normalized features.
Combine Familiarity at 40%, Readiness at 20%, and Opportunity at 40%. Sort by full-precision Expansion Score, then canonical city ID. Scores are relative indices for their release, not probabilities, demand forecasts, or comparable values across releases.
All 15 required ranking measurements must be present for a candidate to be ranked. No imputation or city-specific weight redistribution occurs. Optional road fields do not affect rankability. Legal evidence stays outside the numeric score.