Tropical cyclones intensify faster than 20th-century models predict

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 apparent super-Emanuel intensification signal is likely an artifact of correlated observational technology upgrades rather than a physical thermodynamic effect. IBTrACS best-track wind speeds in the Western Pacific and Indian Ocean basins were derived from subjective Dvorak technique estimates before roughly 2002, systematically underreporting peak intensities, while post-2003 records incorporate the Advanced Dvorak Technique (ADT), SSMIS passive microwave imagery, and GPS dropsonde validation—all of which recover higher peak winds. Simultaneously, pre-Argo (pre-2003) ocean heat content in the same basins relied on sparse XBT casts with known depth-bias corrections, causing OHC to be underestimated in the same early period. Because both variables share a measurement discontinuity concentrated near 2003–2005, OLS will fit a spuriously steep slope that mimics a super-Emanuel scaling without any change in actual intensification physics.

   Resolved

   Apply a Chow test for structural break at 2003 in the OLS regression of ln(peak wind) on OHC, then restrict the primary inference to the North Atlantic alone using HURDAT2, where NOAA Hurricane Hunter aircraft reconnaissance provides internally consistent wind speed measurement back to 1945 and eliminates the satellite-era discontinuity. In that subsample, include calendar year as a continuous covariate and re-estimate the partial slope on OHC; if the coefficient remains above 1.07 after absorbing the secular trend and passes the Chow test for stability across the pre/post-Argo split, the thermodynamic hypothesis survives observational confounding. If the Atlantic-only, year-controlled slope falls to ≤1.05 or exhibits a significant break at 2003, the result is driven by correlated measurement upgrades rather than intensification physics, and the insurance pricing uplift conclusion is unsupported.

2. Fact-Checker #02
   Raised

   Best-track tropical cyclone intensity estimates (HURDAT2, IBTrACS) carry a 1σ uncertainty of roughly ±10 knots (~5–8% in wind speed) for Atlantic storms and ±15 knots for Pacific/Indian Ocean storms where Dvorak satellite technique dominates, propagating to ±0.05–0.08 in ln(max wind); Argo-derived upper-ocean heat content (0–700 m) carries a per-basin-area uncertainty of approximately ±0.3–0.5 × 10⁸ J/m² (1σ), larger still for pre-2005 XBT-era records. Combining these error sources, the standard error on a OLS regression slope fitted to ~30–50 annual basin-level data points is conservatively ±0.10–0.20 at 95% confidence, which is five to ten times larger than the 0.02-unit gap separating the SUPPORTS threshold (>1.07) from the FALSIFIES threshold (≤1.05). Additionally, the metric axis is stated as "OHC in 10⁸ J/m²" while both thresholds are expressed "per °C," introducing a unit mismatch that renders the stated slope values dimensionally uninterpretable without an explicit OHC-to-temperature conversion factor.

   Resolved

   Adopt an errors-in-variables (total least-squares) regression using IBTrACS v04 with quality flag = "official agency" only, and Argo/EN4 OHC with the Cheng et al. (2022) bias-correction applied to XBT data pre-2005; propagate wind-speed and OHC uncertainties as heteroscedastic weights. Widen the decision band so SUPPORTS requires a slope exceeding the Emanuel 5%/°C benchmark by at least one full pooled-uncertainty standard deviation (~0.15 on the ln-scale), and FALSIFIES requires the 95% bootstrap CI upper bound to lie entirely below that benchmark; specify the OHC-to-temperature conversion explicitly (e.g., using NCEI's 0–700 m global mean heat capacity factor) so the slope and thresholds share consistent units before any comparison is made.

3. Researcher #03
   Raised

   The primary uncontrolled confounder is basin-level vertical wind shear (200–850 hPa), which is a first-order determinant of tropical cyclone intensification and covaries endogenously with ocean heat content through climate modes, particularly ENSO. During La Niña episodes, Atlantic upper-ocean heat content rises while upper-level westerlies weaken, simultaneously suppressing shear and permitting deeper intensification; the OLS coefficient on OHC therefore absorbs the shear channel, inflating the apparent thermal sensitivity. This single omission is sufficient to generate spurious departure from Emanuel scaling even if the true OHC-intensity relationship lies within the predicted 5%/°C envelope.

   Resolved

   Basin-mean 200–850 hPa vector wind shear should be computed from ERA5 reanalysis (Copernicus C3S; variables u, v at pressure levels, 0.25° × 6-hourly) and entered as a time-varying covariate in the regression, partitioning the atmospheric dynamic pathway from the thermodynamic one. The Niño 3.4 index (NOAA PSL monthly) can instrument for the joint OHC–shear covariation, enabling 2SLS estimation that isolates the direct thermal forcing effect. Adding basin × calendar-quarter fixed effects using the IBTrACS v04 best-track panel would additionally absorb seasonally persistent atmospheric states correlated with both regressors, yielding an identified slope attributable specifically to sub-surface thermal energy rather than confounded climate-mode variance.

4. Compliance-Guard #04
   Raised

   Because the hypothesis explicitly predicts a 15–25% upward recalibration of coastal-exposure insurance pricing, premature citation of this regression as established exposes insurers and reinsurers to enforcement under state insurance-department rate-adequacy statutes (e.g., NAIC Model Rate Filing Law) and actuarial-standard violations under ASOP No. 56 (Modeling), which requires that models used for ratemaking be validated and that material assumptions be documented and defensible. Simultaneously, any publicly traded (re)insurer or cedant that cites this slope in SEC climate-risk disclosures—required under the SEC's climate-disclosure rules or voluntarily under IFRS S2—before the SUPPORTS threshold is formally crossed risks a Rule 10b-5 material-misstatement claim, because a regression slope sitting in the indeterminate corridor (1.05–1.07) cannot yet support the materiality assertion that legacy catastrophe-model wind-speed distributions are systematically deficient.

   Resolved

   The regression slope derived from the /api/cyclones, /api/ocean, and /api/marine endpoints must demonstrably exceed 1.07 per °C across all three basins with confidence intervals that exclude 1.05 before any actuarial, rate-filing, or investor-disclosure use is permitted; the OLS result must be cross-validated against at least one independent observational dataset not used in model training and subjected to peer review in a recognized climatological or actuarial journal. Until that validation is complete and the SUPPORTS threshold is formally crossed, all downstream documents—rate filings, IFRS S2 scenario analyses, SEC risk-factor disclosures, and catastrophe-model vendor certifications—must carry an explicit disclaimer stating that the intensification scaling hypothesis remains falsifiable and that no binding price adjustment has been actuarially certified under ASOP No. 38 or No. 56.

5. Falsification-Auditor #05
   Raised

   The gap between the FALSIFIES threshold (slope ≤ 1.05) and the SUPPORTS threshold (slope > 1.07) is only 0.02, yet OLS regression slopes for ln(peak wind) against ocean heat content in basin-level datasets carry a standard error typically in the range of 0.03–0.05, arising from annual cyclone-count volatility, ~5–10 kt wind-speed measurement uncertainty in best-track records (Dvorak technique bias is itself basin-dependent), and Argo-era OHC retrieval noise of roughly ±0.15 × 10⁸ J/m². This means the 95% confidence interval on any single estimated slope will span approximately ±0.06–0.10, far wider than the 0.02 falsification band; a true slope sitting exactly at the Emanuel null (≈1.05) would produce observed slopes above 1.07 roughly as often as below 1.05, so the FALSIFIES zone is not selectively entered under the null — it is entered with the same probability as the SUPPORTS zone, rendering the hypothesis empirically one-sided.

   Resolved

   Conduct a Monte Carlo calibration under the null by generating 10,000 synthetic datasets with a true slope of 1.05, drawing cyclone counts from a Poisson process matching each basin's observed climatology, adding Gaussian wind-speed noise at ±7 kt, and adding OHC perturbations at the NCEI ensemble spread (~0.15 × 10⁸ J/m²); record the distribution of estimated OLS slopes to establish what observed slope value corresponds to the 5th percentile under the null — that value, not 1.05, should become the FALSIFIES threshold. If this Monte Carlo lower tail falls at, say, 0.92, expand the FALSIFIES band accordingly, and add a direct-validation arm using ERA5 reanalysis ensemble members as a sensitivity sweep to confirm that instrument-choice variance does not itself shift the slope estimate by more than the proposed 0.02 decision boundary.

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 (monitoring).

1. monitoring

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

• Moody's RMS and Verisk AIR Worldwide — whose wind-field models underpin pricing for approximately $45 billion in outstanding catastrophe bond notional — must revise Atlantic and western Pacific peak wind-speed distributions upward by 8–12% in the ≥Category 4 return-period tail, triggering re-underwriting of affected tranches within 18 months of peer-reviewed confirmation of the >7%/°C scaling.
• Lloyd's of London Realistic Disaster Scenario US Wind benchmark, currently anchored to a 1-in-10-year Gulf Coast Category 4 landfall, requires a wind-speed floor revision of one Saffir-Simpson category in the 1-in-50-year tail, widening catastrophe excess-of-loss rate-on-line by 40–80 bps for Gulf and Southeast US coastal layers at the January 1, 2027 renewal.
• Swiss Re and Munich Re corporate retrocession desks would revise expected annual loss loadings for Atlantic Basin peak-zone quota-share treaties by 20–30%, forcing retrocession premium-rate increases of 60–100 bps on affected accounts within two consecutive June–July renewal cycles.
• Fannie Mae and Freddie Mac conforming mortgage guidelines for properties in FEMA-designated Special Flood Hazard Areas across Gulf and Southeast Atlantic coastal counties would incorporate a 5–10% loan-to-value haircut and mandatory windstorm-insurance rider, repricing an estimated $380 billion in outstanding GSE-backed coastal mortgage exposure by the 2028 annual seller/servicer guide update.
• ASCE 7-28 (the American Society of Civil Engineers minimum design load standard referenced by all U.S. model building codes) would incorporate revised basic wind speed contours reflecting the ≥7%/°C intensity uplift, mandating re-engineering review for coastal structures permitted under ASCE 7-22 thresholds and raising estimated construction-cost premiums for new coastal builds by 4–7% upon the 2030 International Building Code adoption cycle.