Where should an incentive
budget go?

Q = B / s   ·   N = Q · a · (1 − d)

B is an upfront incentive budget, s the incentive paid per offered job, a the share whose site choice changes, and d the share of induced site jobs relocated from other cells in the study area. Q is the number of supported offers; N is the conditional additional job stock. All supported offers incur the incentive cost, including those without additionality.

Cost / job-year = B / [N · Σₜ₌₁ᵀ (1 + r)⁻ᵗ]

Additional jobs are assumed to persist at a constant level for T years, with end-of-year discounting at r. At zero net job growth the cost ratio is undefined, even if relocation improves access. Fractional jobs are expected quantities in a continuous scenario. There is no wage, tax revenue, multiplier, construction-cost or benefit-cost model.

Six candidate cells are selected in descending order of existing primary employment, retaining centres separated by at least 2 km on the UTM grid. Let xₖ be the budget share in candidate k. Induced site employment Qaxₖ is added there. Relocated jobs Qad are removed from all study cells in proportion to baseline employment, including candidate cells. This donor pattern is an explicit simplifying assumption.

ΔJⱼ = Q a Σₖ xₖ 1[j=k] − Q a d · Jⱼ / Σₘ Jₘ

Optimization maximizes a weighted mean of all-worker and lower-earnings access, subject to xₖ ≥ 0, Σₖxₖ = 1 and xₖ ≤ the concentration cap. With constant subsidy and response assumptions across sites, the objective is linear: allocate in descending score order until the cap binds. The fixed donor term does not affect the site ranking. Cluster cards report gross induced jobs and the net cell change after donor subtraction. Equal allocation is a comparison policy. No site-specific construction capacity or cost is inferred.

Aᵢ = Σⱼ Jⱼ · exp[−ln(2) · tᵢⱼ / h]

Directed shortest-path travel time tᵢⱼ weights jobs by half-life h. Unreachable destinations contribute zero. Residential primary-job counts weight group means. ΔAᵢ uses ΔJⱼ in the same expression. This is potential access, with no skill matching, commuting-flow prediction, job competition or hiring effect.

2020 Census-block centroid records are summed into 1 km UTM zone 17N cells. Each cell centre snaps to the largest weakly connected road component within 750 m. Access connectors use 15 mph and within-cell travel uses two minutes. Mapped numeric speeds and one-way restrictions are retained. Speed fallbacks are 55 mph motorway, 45 trunk, 35 primary, 30 secondary, 25 tertiary/unclassified, 20 residential/link and 10 living street.

Roads extend beyond the study window to permit detours. Job destinations and residential weights are limited to the window. Turn restrictions, intersection delay, traffic, service/private roads and transit are omitted. The road snapshot and 2023 employment have different vintages. Grid aggregation, centroid snapping and missing external opportunities limit interpretation.

Employment columns use 0.2 display metres per primary job, with a 20 m minimum marker. Investment columns use 250 display metres per $1 million allocated, with a 20 m minimum for positive allocations; zero allocations are hidden. Access-change columns show the magnitude of ΔA, scaled within the current scenario; teal denotes a gain and rust a loss. The access layer shows cells with resident workers; all modeled cells remain in exports. Heights are statistical encodings, not buildings or flood depths. Terrain is sampled at column centres; Cesium attribution remains visible.

Default policy parameters are illustrative, not calibrated to Asheville, a specific tax program, or the research literature. The Bartik research review on incentive additionality motivates examining the counterfactual, but supplies no estimated coefficient to this prototype. No current incentive eligibility, available land, employer commitments or project costs have been verified.

U.S. Census Bureau LODES 8.4: 2023 North Carolina WAC and RAC, S000, JT01 primary jobs, plus the 2020-block crosswalk. Public LODES uses disclosure protection; small-area counts are not exact employer records. Technical documentation.

Roads and rivers © OpenStreetMap contributors, ODbL. OSM-derived graph and geometry in the download retain that license. Census counts are U.S. public data. The source manifest records query, retrieval date, snapshot and SHA-256 hashes. Scripts and numerical checks live with the website source.

Scenario exports include inputs, accounting, allocations, cell results, source provenance and coverage. No client data or private credentials are included.