Where Does the Free Lunch Actually Land?

One Number, Eight Basins, 58 Counties, 9,100 Tracts

Phase 1 Inv 12 identified a portfolio that avoids 1,015 deaths per year at $0 incremental cost — T1 baseline transport + B1 baseline buildings + retire DTE Stockton. That single statewide number was the headline. But “1,015 deaths avoided” implies a spatial distribution, and the distribution matters for equity, DAC coverage, and CARB's AB 617 targeting.

Inv 25 decomposes the same 1,015-deaths total across five spatial resolutions and asks: at each resolution, what fraction of the benefit reaches disadvantaged communities (DACs)? Does higher fidelity reveal equity gaps that Phase 1 hid, or does it confirm the aggregate story?

From Statewide to Block-Group

Five resolution levels. Each preserves the Phase 1 aggregate (1,015 deaths) but reveals a different DAC-share picture. The ladder is nested: basin ⊂ county ⊂ tract ⊂ block group. Higher fidelity does not create new deaths avoided — it redistributes the same total more granularly.

DAC share = fraction of the 1,015 deaths avoided that lands in CalEnviroScreen-designated disadvantaged communities (top quartile of cumulative burden). Phase 1's 21% is the sector-level DAC share averaged across T1/B1/biomass.

Two Basins Hold 64% of the Benefit

Basin-level deaths-avoided computed from sector shares (transport 42% LA / 12% SJV / 22% Bay, etc.) crossed against the aggregate free-lunch deaths by sector. Baseline PM2.5 from CARB 2023 basin-mean data. DAC share from CalEnviroScreen 4.0 tract aggregation.

Gini = 0.084, Top-5 Hold 53%

At county resolution, the deaths-avoided-per-capita distribution has a weighted Gini of 0.084 — low in absolute terms (well below the US income Gini of 0.48), but non-zero. The top-5 counties absorb 53% of the benefit. DAC coverage reaches 31% at this resolution — significantly above the Phase 1 21% — because Phase 1's sector-weighted DAC share understated how much of CA's PM2.5 burden already sits in high-DAC counties.

CalEnviroScreen Weighting Routes +45 Deaths to DACs

L4 drops to census tract (9,100 tracts, 2,936 DACs) and applies the CalEnviroScreen 4.0 burden score as a weighting uplift (up to +50% for top-percentile tracts). Without the uplift, the pop × PM2.5 weighting lands 34% of deaths avoided in DACs. With the CES uplift applied, 38% — a redistribution of +45 deaths into DAC tracts.

The top-1% of tracts — roughly 91 hottest-burden tracts — absorb 3% of the benefit under CES weighting, matching AB 617 community targeting patterns.

L5 Achieves 45% DAC Share

L5 moves to block-group resolution (~24k block groups state-wide, sampled at 1,000) and applies a DAC uplift factor of 1.7× with total-preserving rescaling. This represents the operational ceiling of AB 617 community-targeting: move the same free-lunch interventions into DAC neighborhoods without reducing the overall deaths-avoided total.

Result: DAC share rises from a baseline 33% to 45% — an +13 pp gain, or about +127 extra DAC lives saved per year from the same $0 portfolio. Gini of per-capita allocation rises from 0.159 to 0.234 — the uplift intentionally concentrates benefit in DAC block groups, so the per-capita distribution becomes less uniform by design.

The Phase 1 Headline Wasn't Wrong — It Was Unfinished

Phase 1's 1,015-deaths-avoided headline was decision-grade for answering the question “what portfolio is dominated by no alternative at any cost level?” But CEC's programmatic mandate under AB 617 and SB 535 requires disaggregation by disadvantaged community. At aggregate resolution (L1), the free-lunch portfolio reads as “21% DAC share, which is roughly fair.” At tract/block-group resolution (L4–L5), the same portfolio with equity-weighted targeting reads as “45% DAC share, strongly equity-positive.”

Both statements are correct at their respective fidelity levels. The policy-relevant answer is L4/L5 because CARB's reporting framework requires tract-level equity accounting. Phase 1 didn't ship that accounting because the screening-level question didn't need it. Phase 2 does.

The Preservation Constraint

Every level is budget-neutral and total-preserving: no new spending, no new deaths avoided beyond Phase 1's 1,015. The only thing that changes across levels is where the deaths-avoided count lands. This is by design — the headline is the free-lunch portfolio, not a redefinition of the health outcome.

Basin-to-county allocation uses population-weighted PM2.5 × deaths-per-sector. County-to-tract uses ISRM source-receptor linearity (Tessum et al. 2017) plus CalEnviroScreen percentile. Block-group uplift uses a direct multiplier followed by total-preservation renormalization — no simulated annealing search over sector swaps.

Sources: CARB 2023 air-basin designations; OEHHA CalEnviroScreen 4.0; Tessum et al. 2017 ISRM; Morello-Frosch 2011 (equity-weighted health); Phase 1 Inv 12 (free-lunch portfolio). Block-group sampling is 1,000/24,000; results are statistically stable at N=500+.