Renewable energy share predicts measurable PM2.5 decline

Five independent claude-sonnet-4-6 calls, one per persona — Skeptic, Fact-Checker, Researcher, Compliance-Guard, Falsification-Auditor. Each writes its hardest objection from its own seat, paired with the methodological resolution it would accept. Run on the static catalogue spec Jun 3, 2026; a live council for any topic is at /try.

1. Skeptic #01
   Raised

   The observed PM2.5 decline may be driven primarily by simultaneous coal-fired power plant retirements rather than renewable energy addition per se, since renewable share increases and coal capacity retirements are mechanically collinear in nearly every grid transition. EIA Form 860 data show that in U.S. regions where renewable penetration rose fastest between 2015–2023 (e.g., MISO, PJM sub-regions), coal retirements averaged 4–7 GW per year concurrently, and coal combustion is the dominant source of SO₂ and NOₓ that form secondary PM2.5. The regression's renewable-share coefficient would therefore absorb the causal effect of coal retirement, inflating the apparent magnitude of β and producing the ≤ -0.8% threshold even if renewables themselves have no independent air-quality mechanism.

   Resolved

   Include retired coal capacity (MW, sourced from EIA Form 860 annual generator retirement schedules, matched to regional grid balancing authorities) and plant-level SO₂/NOₓ emission reductions (from EPA CEMS, EPA Clean Air Markets Division datasets) as explicit covariates in the panel regression alongside renewable share. Run the model both with and without the coal-retirement variable and apply a Frisch–Waugh–Lovell partial regression to isolate the renewable-share coefficient net of coal exit; if β_renewable attenuates from below -0.8 to within the falsification band (|β| < 0.2) after adding retired-coal-MW as a control, the alternative mechanism is confirmed. A Hausman-style comparison of the two coefficient vectors, with the decision rule that p < 0.05 on the difference constitutes confounding, would settle whether the renewable-share effect is independent or merely a proxy for fossil fuel retirement.

2. Fact-Checker #02
   Raised

   The /api/openaq and /api/aqi endpoints aggregate EPA Federal Reference Method monitors (24-hour PM2.5 precision ±2 μg/m³, systematic bias ±0.5 μg/m³ per 40 CFR Part 50 Appendix L) alongside low-cost optical particle counters (e.g., PurpleAir nephelometers with ±20–50% relative uncertainty prior to correction), yielding a blended annual-average uncertainty of roughly ±1–3 μg/m³ or 5–15% relative at a typical regional baseline of 8–15 μg/m³. The predicted signal — an 8–15% PM2.5 decline per 10-point renewable-share gain — sits at or below this noise floor, and the decision band separating FALSIFIES (|β| < 0.2) from SUPPORTS (β < −0.8) spans only 0.6 percentage points of slope; classical OLS attenuation bias from measurement error in both the outcome and the renewable-share predictor (which carries 6–18 month reporting lags at inconsistent spatial granularity across jurisdictions) will routinely produce 95% confidence intervals that straddle this gap without resolution.

   Resolved

   Restrict the PM2.5 data pipeline exclusively to OpenAQ records flagged `"sourceName": "reference grade"` with ≥75% hourly completeness (EPA Method 2.12 threshold), eliminating low-cost-sensor noise, then apply an errors-in-variables or two-stage least-squares estimator that instruments renewable-share with EIA Form 860 grid-level capacity-addition records to correct attenuation bias in β. Revise the SUPPORTS threshold to β < −1.5% and the FALSIFIES boundary to |β| < 0.5% — bands wide enough to exceed the ±0.5 μg/m³ systematic bias floor — and verify testability by confirming that bootstrapped panel-regression 95% CIs are narrower than 0.5 percentage points before declaring the hypothesis falsifiable under the stated 24-month window.

3. Researcher #03
   Raised

   The regression does not control for the concurrent displacement of coal-fired generation by natural gas, which typically co-occurs with renewable buildout under the same policy environment that drives both investments. Coal combustion produces PM2.5 directly and generates SO₂ that forms secondary sulfate aerosol; when a region retires coal capacity simultaneously while adding renewables, the renewable-share coefficient absorbs the pollution-reduction effect of the coal-to-gas substitution, biasing β away from zero and overstating the causal contribution of renewables per se. GDP and population controls do not absorb this channel because coal-to-gas transitions can occur independently of economic output or demographic change.

   Resolved

   Include monthly generation-weighted coal share and natural gas share as additional regressors in the panel, sourced from EIA Form EIA-923 (U.S. balancing-authority level) or the ENTSO-E Transparency Platform's "Actual Generation per Production Type" dataset (EU bidding zones), both of which provide the unit-level fuel-specific generation needed to construct these shares at the regional resolution of the PM2.5 outcome. For sharper identification, instrument for renewable share using the interaction of pre-sample renewable resource endowment — NREL Annual Technology Baseline capacity-factor grids for wind or NASA MERRA-2 surface solar irradiance — with staggered state/national renewable portfolio standard mandate schedules, exploiting plausibly exogenous geographic variation in buildout timing while the coal-share covariate absorbs the parallel fossil-fuel substitution pathway.

4. Compliance-Guard #04
   Raised

   Pricing Social Impact Bonds with an 8–12% yield premium anchored to this unvalidated PM2.5 coefficient would constitute a material misrepresentation under SEC Rule 10b-5 (17 CFR §240.10b-5) if the regression has not yet crossed the SUPPORTS threshold, because the yield spread is contractually tied to a quantified air-quality outcome that remains falsifiable under the experiment's own criteria. Concurrently, embedding the provisional slope into parametric health-insurance rate filings exposes underwriters to state insurance-commission enforcement under NAIC Actuarial Standard of Practice No. 25 (Credibility Procedures), which prohibits using statistically non-credible data as a primary rate basis. Finally, citing this hypothesis as settled science in State Implementation Plans under Clean Air Act §172 (40 CFR Part 51, Subpart G) or in EU national air-quality plans under Directive 2008/50/EC would misrepresent the evidentiary basis available to regulators, risking EPA plan-adequacy challenges and citizen-suit liability under CAA §304, as well as European Commission infringement proceedings if plan attainment demonstrations rely on unvalidated causal claims.

   Resolved

   No financial instrument, insurance product, or regulatory plan submission may cite this hypothesis as established until the SUPPORTS threshold is formally crossed — specifically, a replicated slope coefficient β < −0.8% (p < 0.05) confirmed across at least five independent regional datasets using leave-one-region-out cross-validation, with the full data-pipeline joins (/api/energy, /api/aqi, /api/openaq, /api/air) independently audited for measurement equivalence and residual confounding. Any SIB prospectus issued before that gate must carry explicit language stating that the PM2.5-yield linkage reflects a hypothesis under active investigation rather than a demonstrated causal effect; parametric insurance filings must classify the ±15% forecast accuracy as provisional and ineligible for primary rate-setting; and submissions to EPA or EU competent authorities must characterize renewable-energy share as a corroborative indicator pending peer-reviewed, journal-published confirmation rather than as a validated attainment mechanism.

5. Falsification-Auditor #05
   Raised

   The FALSIFIES threshold of |β| < 0.2 is practically unreachable under the null because renewable energy deployment is structurally correlated with simultaneous economic restructuring, industrial relocation, and co-enacted emission-control policies that independently suppress PM2.5—meaning a spurious β well above 0.2 will persist even when renewables exert no direct causal effect. Regional PM2.5 panels exhibit interannual meteorological variance of 15–30% (driven by precipitation, wind-field shifts, and transboundary transport), which propagates into a β estimation standard error of roughly ±0.3–0.5 per 1% renewable-share unit across typical 24-month windows—already larger than the 0.2-unit falsification threshold. Because the GDP and population controls cannot fully absorb these correlated co-trends, the hypothesis cannot realistically be entered into the FALSIFIES band under a well-calibrated null, rendering it effectively unfalsifiable as written.

   Resolved

   Run a Monte Carlo falsification sweep by permuting renewable-share time series across economic-tier-matched regions 10,000 times, preserving seasonal and economic autocorrelation structure while severing any causal link to PM2.5; if the 5th percentile of the resulting null β distribution exceeds 0.2, the FALSIFIES threshold must be raised to that percentile value. Supplement with a differences-in-differences arm that uses quasi-exogenous coal-plant retirement events (sudden regulatory closures) as instrumental shocks to renewable share, collapsing the smooth economic confound and materially tightening residual PM2.5 variance; redefine FALSIFIES as requiring both |β_IV| < the empirically calibrated null-distribution bound AND the estimate falling within the permutation null interval at p < 0.05, so that a falsifying result is distinguishable from a confounded non-result rather than merely a noisy one.

Unlike the static stress tests above (synthesised against the frozen catalogue spec), this is what a 3-voice council found in the most recent biweekly review. Refreshed on the 1st and 15th of each month at 09:00 UTC. Each voice runs one bounded web search via Anthropic's web_search_20260209 tool, cites what it finds, and recommends a verdict. The verdict diverges from the curated catalogue status (monitoring) — the synthesis below explains why.

Agent draft incorporating the 9 cited findings from the live council above. Not auto-merged — surfaces here for human review. To accept, open a PR editing site/src/_data/hypotheses.json with the revised fields below. To reject, ignore and the proposal will refresh on the next council run.

1. monitoring

   May 30, 2026 · added to catalogue at status "monitoring"

• IFC and EBRD project-finance desks price health-co-benefit tranches in social-impact bonds for renewable-transition deals at an 8–12% yield premium over vanilla green bonds, targeting ≥$3 billion in issuance across PM2.5-burdened emerging-market corridors by end-2027.
• The U.S. EPA Office of Air Quality Planning and Standards revises State Implementation Plan guidance under Clean Air Act §172 to credit renewable-energy share targets as a quantified PM2.5 control measure, reducing projected compliance costs by $2–4 billion annually across current non-attainment areas within 18 months of EPA internal confirmation of the dose-response coefficient.
• The U.S. Interagency Working Group on Social Cost of Carbon incorporates the PM2.5 co-benefit pathway, raising the health-damage component of the central SCC estimate by $12–20/tonne CO2e in its next scheduled revision cycle, expected by 2027, and propagating that uplift into all federal energy-rule cost-benefit analyses.
• Moody's Investors Service revises its North American electric-utility rating methodology to treat quantified PM2.5 liability reduction as a positive credit factor, tightening investment-grade spreads by 20–40 bps for utilities sustaining ≥60% renewable generation share within two annual rating-cycle updates.
• Swiss Re Corporate Solutions introduces DALY-linked parametric health-indemnity products tied to regional renewable-energy share milestones, calibrated to a 0.3–0.6 DALY-per-1,000-persons reduction per 1% renewable-share gain, targeting five pilot jurisdictions and ≥$500 million in gross written premium by 2028.