Solar cycle explains <5% of recent climate variance

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

   CO₂ concentration (Mauna Loa, NOAA/GML Keeling Curve) is an I(1) non-stationary process with a near-monotonically accelerating secular trend, while global mean surface temperature (HadCRUT5, NASA GISTEMP v4) retains substantial low-frequency non-stationarity even after linear detrending. Regressing linearly detrended temperature on non-detrended CO₂ concentration constitutes a partially spurious regression: the residual I(1) variance not removed by linear detrending is mechanically absorbed by any co-trending regressor, inflating CO₂'s R² contribution independent of its causal role. Sunspot number (SIDC International Sunspot Number v2), being a genuinely stationary oscillatory process, will always yield low R² against a partially integrated dependent variable—so the <5% solar fraction may reflect stationarity mismatch rather than causal irrelevance, while the >85% CO₂ fraction may reflect shared stochastic trend rather than radiative forcing.

   Resolved

   Apply Augmented Dickey-Fuller (ADF) and KPSS unit root tests to the linearly detrended temperature residuals; if non-stationarity is not rejected at α = 0.05 in both tests, difference all series to I(0) before running the multiple regression. Alternatively, estimate a Johansen cointegration system (trace statistic, lag order selected by BIC) between temperature, CO₂ radiative forcing from the NOAA Annual Greenhouse Gas Index (AGGI, which converts mixing ratios to W m⁻² and avoids spurious collinearity with raw concentration), TSI from the PMOD composite (VIRGO/PMO6), and ENSO (MEI v2) and volcanic AOD (GISS stratospheric aerosol optical depth) as controls; the valid variance attribution comes from the error-correction model's I(0) short-run dynamics, not the levels regression. If CO₂ partial R² (LMG decomposition, R package `relaimpo`) remains >0.85 in the cointegrated or first-differenced specification, the original result stands; if it drops below 0.50, the claimed fraction is a stochastic-trend artifact, not a causal attribution.

2. Fact-Checker #02
   Raised

   SORCE/TIM carries a 1-sigma absolute calibration uncertainty of ±0.035% (~0.48 W/m²), nearly matching the full 11-year solar cycle peak-to-trough amplitude of ~1.3 W/m²; meanwhile the three standard TSI composites (PMOD, ACRIM, IRMB/ROB) disagree on the long-term trend by ~0.3–0.5 W/m², representing 25–40% of that cycle amplitude and producing materially different regression coefficients. The 2015 ISN v2.0 recalibration revised historical sunspot counts upward by ~20%, directly rescaling the solar proxy and shifting the derived variance fraction regardless of which TSI composite is selected. Across plausible composite and proxy combinations, attributed solar variance fraction spans roughly 2–10%, placing the 5% SUPPORTS threshold squarely inside the instrument-uncertainty band where it cannot be robustly distinguished from an indeterminate outcome.

   Resolved

   The regression must be run separately against each of the three TSI composites (PMOD v42_65_1709, ACRIM3, IRMB/ROB) using ISN v2.0-flagged sunspot data, with the solar variance fraction reported as a composite-spread 95% bootstrap confidence interval; the SUPPORTS threshold should be revised from <5% to <2%, below the lower bound of that uncertainty range and consistent with Foster & Rahmstorf (2011, ERL) multi-signal attribution estimates, while the >20% FALSIFIES threshold lies outside the uncertainty band and may stand. The detrending protocol must be applied identically to both temperature and CO₂ series before regression—or neither should be detrended—because regressing residual interannual temperature variance against the near-monotonic, undetrended CO₂ concentration will artificially inflate the CO₂ R² contribution and render the >85% CO₂ threshold unfalsifiable by construction.

3. Researcher #03
   Raised

   The regression omits ENSO (El Niño–Southern Oscillation) as a covariate, which is the dominant driver of interannual variance in detrended global mean surface temperature over the 1979–present window, accounting for roughly 20–40% of sub-decadal variance in standard instrumental records. Because several peer-reviewed studies document a weak but empirically non-trivial solar modulation of ENSO phase distribution through the stratospheric "top-down" pathway (UV heating of the tropical stratosphere altering the Hadley circulation), the sunspot predictor will spuriously absorb a portion of ENSO-driven temperature variance rather than only direct TSI radiative forcing. This upward bias on the estimated solar R² can push the apparent solar variance fraction above the 5% support threshold—and potentially toward the 20% falsification boundary—through a mechanism entirely unrelated to the direct irradiance channel the hypothesis is designed to test.

   Resolved

   Add the NOAA Multivariate ENSO Index v2 (MEI.v2, available monthly 1979–present at psl.noaa.gov/enso/mei/) and the NASA GISS stratospheric aerosol optical depth series (Sato et al., updated at data.giss.nasa.gov/modelforce/strataer/) as explicit regressors, replicating the Foster & Rahmstorf (2011, *J. Geophys. Res.*, doi:10.1029/2011JD016263) multiple-regression framework; the solar variance fraction must then be reported as the partial R² on the sunspot coefficient after projecting out ENSO and volcanic contributions, isolating the direct irradiance-forcing channel from confounding internal variability modes.

4. Compliance-Guard #04
   Raised

   The PREDICTS field directly projects this pre-threshold regression result into U.S. federal climate-damages litigation and EU Climate Law (2021/1119) Article 19 enforcement proceedings; citing an unvalidated variance-decomposition as proof that solar-forcing defenses are scientifically meritless would violate Federal Rule of Evidence 702 under the Daubert standard (*Daubert v. Merrell Dow Pharmaceuticals*, 509 U.S. 579 (1993)), which requires peer-reviewed, independently replicated methodology with disclosed error rates before a federal court may treat a finding as established science. Any corporate climate disclosure—under SEC Release 33-11275, IFRS S2, or EU CSRD (Directive 2022/2464)—that elevates this pre-threshold result to a settled material-risk determination risks Rule 10b-5 liability or CSRD non-compliance for presenting unverified statistical output as established fact. Carbon-tax exemptions or OECD policy instruments modified on the premise of a solar-offset claim derived from this analysis before threshold crossing face parallel administrative-law challenge on the grounds of insufficient evidentiary basis.

   Resolved

   Reliance in any litigation filing, regulatory submission, mandatory disclosure, or policy instrument is prohibited until the SUPPORTS threshold is formally crossed—Solar variance fraction < 0.05 AND CO₂ variance fraction > 0.85—confirmed through peer-reviewed publication with independent cross-validation against at least two temperature datasets (e.g., HadCRUT5 and RSS TLT) and two TSI reconstructions (e.g., PMOD and ACRIM composites) to neutralize data-endpoint dependency risk from the duplicated `/api/solarcycle` endpoint listed in the experiment design. Any expert declaration entered in U.S. federal proceedings must additionally disclose API data provenance, regression specification, and full confidence intervals and survive a Daubert hearing before being treated as admissible established science rather than a falsifiable hypothesis. EU Climate Law Article 19 enforcement submissions citing this variance attribution must await endorsement by an authoritative synthesis body—at minimum IPCC Working Group I—before the European Commission or CJEU may treat the solar-versus-CO₂ ratio as a settled evidentiary baseline.

5. Falsification-Auditor #05
   Raised

   With only approximately four complete 11-year cycles in the 1979–present record, the bootstrap standard error on partial R² from a solar-cycle regression is roughly ±8–12 percentage points, meaning the 20% FALSIFIES threshold sits within one to two standard errors of plausible confound arising from random ENSO phase alignment with solar—reachable under the null by chance even if the true solar fraction is near zero. Compounding this, the absence of a physically motivated solar-to-temperature lag term (literature supports 1–3 years) and the unspecified detrending method (linear vs. quadratic vs. HP-filter choices shift residual variance attribution by 5–10 percentage points) create a structural suppression bias that could prevent the solar variance fraction from crossing 20% even when the true fraction is 15–18%, making the threshold both spuriously reachable by noise and simultaneously blocked by design.

   Resolved

   Construct an empirical null distribution by block-bootstrapping the solar index with block length ≥11 years across 10,000 random phase shifts relative to the temperature record, then read off the 95th-percentile solar partial R² under the null; if that percentile already approaches 20%, the FALSIFIES threshold must be raised (e.g., to 0.30–0.35) to lie outside the null's natural reach. Simultaneously run a full sensitivity sweep over three detrending methods, four lag offsets (0–3 years), two solar proxies (sunspot number vs. SORCE/TIM TSI), and four reanalysis products (HadCRUT5, GISTEMP, Berkeley Earth, ERA5), reporting the complete envelope of solar partial R²—if no specification combination crosses 20%, the FALSIFIES band is confirmed unreachable and must be renegotiated downward (e.g., to 0.10) to make the hypothesis genuinely two-sided.

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 aligns with the curated catalogue status (supported).

1. supported

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

• ExxonMobil, Chevron, and Shell face dismissal of solar-forcing affirmative defenses in ≥45 of the 50+ US federal climate-tort actions filed 2024–2026, as courts apply the <5% solar attribution finding to negate causation rebuttals, leaving an estimated ≥$120 billion in aggregate claimed damages cleared for trial by end of 2027.
• The European Commission initiates ≥3 Article 19 enforcement referrals to the Court of Justice of the European Union against member states that invoked solar-variability justifications for delayed compliance with Regulation 2021/1119 net-zero milestones, with first-instance rulings expected by 2028.
• Finance ministries in ≥5 OECD jurisdictions — including Canada, Norway, and Australia — eliminate carbon-pricing exemptions or sectoral rebates premised on natural solar-cycle offsets, recovering an estimated $3–8 billion in aggregate annual carbon-tax revenue by end of 2027.
• Munich Re NatCatSERVICE and Swiss Re Institute remove solar-cycle uncertainty loadings from boreal and subarctic catastrophe-frequency models, compressing the natural-variability risk premium on approximately $45 billion of reinsured property exposure by 30–60 bps at the January 2027 treaty renewal.
• Verra's Verified Carbon Standard secretariat mandates re-verification of ≥150 MtCO₂e of issued credits whose additionality arguments cited solar-cycle interference as a confound, placing ≥20 MtCO₂e at risk of invalidation and driving an 8–15% spot-price discount on affected credit vintages by 2027.