Solvent Streets

Methodology

This section describes the data sources, models, and assumptions behind the analysis presented in each dashboard.

Data sources

• Road, parking, and sidewalk geometry. OpenStreetMap via the Overpass API, under the Open Database License (ODbL). When a city has no ArcGIS or local-layer source configured, OSM is the sole source and coverage inherits OSM's gaps — most notably uneven sidewalk tagging and inconsistent classification of alleys and unpaved service roads.
• Jurisdictional layers. Optional ArcGIS FeatureService endpoints configured per city in pvmt.toml.
• City boundaries. Nominatim lookups or user-supplied GeoJSON, configurable per city.
• Initial PCI. A stand-in value supplied through config or the interactive slider. PVMT does not ingest field-measured PCI and makes no claim that the starting condition matches reality — changing this input is the right way to explore sensitivity to unknown starting condition.

The exact sources and endpoints used for a given example are listed in that example's Config tab.

Decay model

Each road classification decays independently via

PCI(t) = PCI₀ · exp(−k · t)

where k is an annual decay constant that depends on the road classification. Higher-class roads (motorway, trunk, primary) decay more slowly than lower-class roads (residential, service) because they are built to thicker, more rigorous design standards and typically receive more frequent maintenance. Default values are derived from LTPP data reported in FHWA-RD-01-156, Long-Term Pavement Performance and ship as part of the forecast package; they are continental-US averages and do not account for local climate, traffic, or construction quality. A config may set a per-city decay_rate to tune for local conditions (e.g. freeze/thaw or road salt); that override is applied as the rate for a typical road and scales every road class proportionally, so the per-class ordering (higher classes decay slower) is preserved rather than flattened. Sidewalks decay on a separate, slower track and are not treated as a highway class.

Cost model

Treatment costs are banded by PCI: each band has a representative $/sq m value, and costs between bands are linearly interpolated at the tier midpoints, so the cost-versus-PCI curve is smooth rather than step-shaped. Above the highest anchor (the midpoint of the preventive tier) and below the lowest anchor (the midpoint of the reconstruction tier), the cost is clamped to that anchor's value rather than extrapolated. Default cost tiers are expressed in $/sq m and sourced from FHWA treatment-selection guidance; they are calibration inputs, not measurements, and local bid prices will differ. Roads and sidewalks use independent cost tiers because the treatment economics differ substantially.

Condition spread

A real network is a distribution of conditions — some segments excellent, some failed — not a single average. Because the cost-versus-PCI curve is non-linear, pricing everything at the network-average PCI under-states the true program cost (the failed/poor tail is disproportionately expensive). To correct this without requiring per-segment condition data the model does not have, PVMT spreads the single configured average PCI into a distribution around that average (a Beta distribution on the 0–100 scale, with the average preserved exactly) and prices each slice separately. The result is a more realistic — and modestly higher — cost than pricing the average alone. This is a deliberately conservative approximation of the real spread; it is applied automatically and moves the solvency dollars in the direction validated against published city data. When field-measured per-segment condition becomes available, it will replace this assumed spread.

Treatment cycle

Real pavement is maintained on a multi-year cycle: a city treats roughly one slice of its network each year, not the whole network annually. The model captures this with a treatment cycle of N years (default 12, the midpoint of a typical 10–14 year municipal cycle, configurable via treatment_cycle_years). Each forecast year only 1/N of the network is scheduled for treatment, so the annual treatment need is the full-network retreatment cost divided by N. This is what makes the break-even budget a realistic annual program cost rather than the one-off cost of rebuilding the entire network at once. A cycle of N = 1 reproduces the older behavior (the whole network priced every year), which overstated the hold-steady budget several-fold.

The cost level and the cycle length are not independent — the break-even budget scales as cost ÷ N, so a cheaper cost basis and a shorter cycle trade off exactly. We anchor the cycle to the physical rate at which a city actually repaves (lane-miles treated per year, not dollars): for Berkeley that hold-steady cadence is ≈ 12 years, which is the default. With the cycle fixed there, the default bare-construction cost tiers reproduce the city's cited real hold-steady spend (~$5.6 per m² per year), so the break-even dollars are calibrated to reality rather than chosen freely. See the validation report for the derivation and the limits of this anchor.

Scenario comparisons

PVMT ships with three comparison runs driven by annual funding level, all using the worst-first allocation strategy (budget is spent on the lowest-PCI segments first):

• 25% funding — annual spend = 25% of year-1 worst-first need.
• 50% funding — annual spend = 50% of year-1 worst-first need.
• Full funding — unconstrained budget; spend whatever it takes to treat every segment that falls below the worst-first trigger each year.

A do-nothing baseline (no spend, uncontrolled decay) is shown alongside the funded runs for comparison.

The forecast library also implements a preventive-first strategy (prioritize highest-PCI segments that are still in the preservation window), but the default UI comparisons do not exercise it. Preventive vs. worst-first allocation is governed by per-strategy efficiency multipliers; those multipliers are illustrative calibration constants chosen to reflect the direction and sign of the effect reported in FHWA-HIF-12-042, Pavement Preservation: Preserving our Investment — that $1 of preventive maintenance is reported to avoid $6–$10 of future reconstruction — not to reproduce that benefit-cost ratio as a single-year spending efficiency.

Area growth

Optional compound annual growth applies to pavement area each year:

Area(y) = Area₀ · (1 + g)^y

where g is configured per city (default zero). This lets an example model a city that is still expanding its street network; it does not model demolition or removal.

Solvency metrics (streets/roads only)

The dashboard's Financials headline and the cross-city leaderboard report three solvency figures. They are computed on the roads/streets cohort only — the aggregate scenarios blend roads, parking, and sidewalks but cost the blend at road tiers, which would mis-price sidewalks, so an absolute dollar claim must be roads-only. They are derived from a worst-first run at the city's configured annual budget.

• Insolvency year — the first forecast year in which the cumulative deferred backlog reaches one full treatment cycle of deferred work (N × the annual scheduled need — a whole network's worth of treatment). Because the deferred backlog is a monotonically non-decreasing accumulator (see below), once a city is an entire cycle behind it does not recover within the model, so this is the "unrecoverable" threshold. A do-nothing network defers a full slice every year and crosses around year N; a city funding a fraction f of each slice crosses around N / (1 − f), so well-funded cities never reach it and are reported as solvent through the horizon. The metric therefore discriminates among the genuinely underfunded; on the funded side, funding gap is the primary signal. Reported only when a current budget is configured.

• Hold-steady (break-even) budget — the smallest constant annual budget whose final deferred backlog is within a small relative tolerance (a fraction of the annual need) of zero: the budget that funds the network's annually-scheduled treatment slice and keeps pace with the cycle. Found by bisection over budget; the search's upper bound is the peak do-nothing annual need over the horizon, which is sufficient to fully fund every year.

• Funding gap — (break-even − current budget) / current budget, the primary cross-city ranking metric. Negative when a city already budgets at or above its hold-steady level. Reported only when a current budget is configured.

Three caveats apply to these figures specifically:

• The deferred backlog is cumulative unmet need, not a recoverable balance. It only ever grows; spending in a later year reduces new need but never pays down backlog already accrued. Read it as "total treatment value foregone to date," not as a debt that can be cleared.
• In the years before the insolvency crossing, the reported annual_spend series can exceed the configured budget. The allocator routes leftover budget on a fully-funded cohort into extra PCI recovery (a surplus branch), and that extra is counted as spend. So an annual_spend above current_budget in early years is expected, not an error.
• Break-even assumes the cost-versus-PCI relationship is monotone (true for the default cost tiers: worse pavement costs more to treat). A pathological custom cost_tiers curve that violates this could make the bisection overstate the break-even budget — a conservative direction (it never understates the gap).

Compare-tab scores

The multi-city Compare view presents two distinct per-city scores:

• Solvency Score (primary, the project's headline) — the road-budget coverage ratio, current_budget / break_even_budget, rendered on a 0–100 scale: 100 = funded to hold the network steady, 50 ≈ needs twice today's budget, 25 ≈ needs four times. Over-funded cities cap at 100 for display but keep their uncapped coverage as the ranking key, so the most-surplus city ranks first (a single gold medal); ties break on the later/absent insolvency year. Computed on roads/streets only and only for cities with a cited current_budget — cities without one cannot be scored and are shown N/A, unranked.
• Infrastructure Index (secondary) — a purely descriptive composite of asset scale, ownership share, walkability, unit cost and blended decay, min-max normalized across the compared cities. It is not a solvency judgment (its "bigger network / more city-owned = higher" weights are descriptive, not normative); it exists to summarize network character and is always computable, so every city places.

Assumptions and limitations

• Initial PCI is a user-chosen stand-in, not a field measurement.
• Decay-rate defaults are continental-US averages; local climate, traffic, and construction quality shift the true values.
• Hex aggregation loses sub-hex heterogeneity — a hex with one badly deteriorated street and three good ones looks like a moderately deteriorated hex.
• No network effects are modeled (detours, closures, access externalities, project bundling).
• All costs are in today's dollars. No inflation, discounting, or present-value adjustment is applied.
• Treatment effectiveness is modeled as an efficiency multiplier on spend, not as a deterministic post-treatment PCI bump.
• Output is a planning-grade estimate, not an engineering specification.

References

• FHWA-RD-01-156 — Long-Term Pavement Performance, Federal Highway Administration. Source for per-class decay-rate defaults.
• FHWA-HIF-12-042 — Pavement Preservation: Preserving our Investment, Federal Highway Administration. Motivates the direction of the preventive-first vs. worst-first efficiency adjustment.
• OpenStreetMap contributors — road, parking, and sidewalk geometry (ODbL).
• pvmt source — implementation reference. See docs/architecture.md for the ingest and compute pipeline.