Climate-attributable migration is accelerating ahead of IPCC AR6 projections

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 dominant alternative explanation is that armed conflict and governance collapse — which are spatially and temporally correlated with drought-affected regions across the Sahel, Horn of Africa, and Central America's Dry Corridor — are the actual drivers of observed UNHCR outflow acceleration, generating a spurious climate-migration slope when conflict intensity is omitted from the regression. ACLED and UCDP/PRIO-GED data confirm that battle-event frequency and political violence spiked in the same grid cells and years (2015–2023) that the drought and temperature anomaly endpoints flag as high-stress, meaning OLS on the climate composite absorbs conflict-driven displacement and inflates the apparent climate coefficient well above any AR6 benchmark. Additionally, non-stationary changes in destination-country asylum registration capacity (EU policy post-2015, U.S. Title 42 enforcement fluctuations) introduce administrative artifacts into the UNHCR dependent variable that can independently generate apparent acceleration trends entirely decoupled from climate forcing.

   Resolved

   Augment the panel regression with UCDP GED battle-deaths-per-100k and ACLED political-violence-event counts as time-varying covariates and include destination-country × year fixed effects to absorb registration-policy variation; then instrument the World Bank cereal-yield-per-hectare agricultural productivity proxy with the FAO Food Price Index — a global commodity shock plausibly exogenous to local conflict dynamics — using 2SLS, reporting both OLS and IV coefficients with a Hausman endogeneity test. The decision rule is explicit: if the IV-estimated climate-productivity coefficient remains above the AR6 WGII Chapter 7 SSP2-4.5 central migration-response estimate (approximately 0.8 additional cross-border migrants per 100k per standardized climate-stress unit) across ≥3 regions after conflict controls, the hypothesis survives; if the Hausman test confirms endogeneity and the IV coefficient collapses within the AR6 ensemble range, the observed acceleration is attributable to correlated conflict rather than independent climate forcing and the hypothesis is falsified.

2. Fact-Checker #02
   Raised

   The compound metric multiplies at least four data streams with independent error budgets: ERA5 regional temperature anomalies carry ±0.3–0.5°C 1-sigma uncertainty in data-sparse regions (IPCC WGI AR6 §1.5.2); CHIRPS-derived drought-week-area indices carry ±15–25% bias in sub-Saharan Africa and South Asia where signal is largest; World Bank agricultural value-added proxies lag 2–3 years and carry ~15% methodological uncertainty in low-income countries; and UNHCR cross-border outflow counts undercount undocumented climate-related movement by an estimated 40–60% (Abel et al. 2019, Boas et al. 2019). Propagated in quadrature, the combined 1-sigma uncertainty on the composite slope easily spans a factor of 2–3×, meaning the "greater than 2× AR6 central estimate" threshold sits entirely inside the measurement noise. Furthermore, AR6 Cross-Chapter Box MIGRATE does not report migration projections as a regression slope in these units (per-degree × drought-week-area × productivity), so no single "AR6 SSP2-4.5 central estimate" in the required dimensional form actually exists against which to anchor the falsification threshold.

   Resolved

   First, re-express the AR6 benchmark by translating the Rigaud et al. (2021) internal-migrant range (median ~216 M by 2050 under SSP2-4.5) into an annualized, per-region slope with explicit units, then construct a 90% bootstrap confidence interval on your observed regression slope using 10,000 resamples stratified by region and year. Apply UNHCR data-quality tier flags (Tier 1: government statistics; Tier 2: UNHCR estimates) and restrict primary inference to Tier-1 country-years; run Tier-2 as a sensitivity arm. Correct temperature inputs using ERA5-Land rather than ERA5 (reduces regional 1-sigma to ~0.15°C), use the multi-index drought ensemble mean of SPI-3, SPI-6, and PDSI to reduce single-index bias, and declare the hypothesis testable only if the lower bound of your 90% CI exceeds the upper bound of the AR6 ensemble range after full uncertainty propagation.

3. Researcher #03
   Raised

   The dominant uncontrolled confounder is concurrent armed conflict and political instability, which operates through the following channel: the regions exhibiting the largest drought-week-area × temperature anomaly values (Sahel, Horn of Africa, northern Central America) are simultaneously the regions with the highest conflict intensity, and conflict independently causes the cross-border displacement captured in UNHCR outflow statistics. Because climate stress and conflict co-occur non-randomly — resource scarcity amplifies state fragility and inter-group violence — the OLS slope on the climate composite will absorb conflict-driven displacement, producing an upwardly biased coefficient that mechanically inflates the apparent ratio relative to AR6 central estimates. Without separating these pathways, the hypothesis cannot distinguish "climate-attributable migration accelerating" from "conflict-attributable migration being mislabeled as climate-attributable."

   Resolved

   The analysis should incorporate the UCDP Georeferenced Event Dataset (UCDP GED, available at ucdp.uu.se, country-year and subnational resolution) as a time-varying conflict-intensity covariate — specifically battle-related fatalities per 100k — entered additively in the regional panel regression alongside the climate composite. An instrumental-variables approach strengthens identification further: rainfall deviation in upstream riparian states (derived from ERA5 monthly precipitation, Copernicus CDS variable "total_precipitation") serves as a plausibly exogenous instrument for local conflict onset, allowing a two-stage specification that isolates the residual climate-to-migration channel net of conflict-mediated pathways; the first-stage F-statistic and overidentification tests should be reported. Restricting the falsification sample to country-years with UCDP conflict fatalities below a low threshold (e.g., <25 battle deaths) provides an additional robustness check in which the confounding pathway is structurally minimized.

4. Compliance-Guard #04
   Raised

   The downstream prediction that climate-adaptive sovereign bonds and migration-financing instruments require 5–10x scaling directly implicates SEC Rule 10b-5 and EU SFDR (Regulation (EU) 2019/2088) Articles 8 and 9 product-disclosure obligations, because citing this acceleration claim in a bond prospectus or sustainability-linked instrument term sheet before the slope exceeds 2× the AR6 SSP2-4.5 central estimate across three validated regions would constitute a material misstatement or misleading omission of fact actionable under both regimes. In parallel, IFRS S2 paragraph 22 requires that scenario analysis in climate-related financial disclosures rest on "reasonable and supportable information," so any corporate or sovereign issuer embedding the 30–50% UNHCR underestimation claim in a climate-risk section of audited financial statements—prior to threshold crossing—exposes auditors and preparers to enforcement under IOSCO principles and applicable national securities regulators. OECD DAC concessional-finance allocation norms carry a further channel: donor governments re-sizing ODA migration envelopes against an unverified projection risk misreporting aid effectiveness metrics under DAC Creditor Reporting System rules.

   Resolved

   The hypothesis must not be incorporated into any securities offering document, IFRS S2 scenario disclosure, SFDR product classification, UNHCR planning baseline, or ODA allocation framework until the pre-registered SUPPORTS threshold is formally crossed—slope > 2× AR6 central estimate sustained across ≥3 geographically independent regions using all four API endpoints—and the regression is peer-reviewed in a recognized climate-migration or econometric journal and cross-validated against an independent ground-truth dataset not used in model fitting, such as IOM Displacement Tracking Matrix records or independent national census migration microdata. All interim outputs must carry a mandatory forward-looking statement disclaimer reading: "This analysis has not crossed its pre-registered SUPPORTS threshold; it may not be cited as established in securities disclosures, sustainability-product classifications, intergovernmental planning baselines, or instrument-sizing rationales pending peer review, independent replication, and formal regulatory or standards-body adoption."

5. Falsification-Auditor #05
   Raised

   The composite metric multiplies four independently noisy variables (drought-week-area, temperature anomaly, labor-productivity decline, and migration outflow per 100k), which compounds their individual measurement uncertainties—UNHCR cross-border attribution carries ±40–60% uncertainty, agricultural-productivity proxies derived from NDVI or WorldBank series add ±20–30% interannual variance, and drought-index choice (SPI-3 vs. SPI-12 vs. PDSI) alone shifts drought-week-area estimates by 40–60% across regions. The resulting slope confidence interval on any single regional regression is likely to span ±100–200% of the estimated central value, meaning it simultaneously overlaps both the FALSIFIES band and the SUPPORTS threshold; under the null, the slope estimate will wander across the AR6 ensemble boundary by pure noise, making the FALSIFIES condition trivially enterable but also trivially exitable without discrimination power.

   Resolved

   Run a permutation-based Monte Carlo null test by randomly reshuffling annual migration-outflow vectors against the climate-predictor time series within each region (5,000 draws), constructing the null distribution of regression slopes, and checking whether the AR6 ensemble range represents a statistically distinct zone rather than just the interior of that null distribution—if the AR6 range falls entirely within the central 80% of permuted slopes, the threshold must be tightened or the metric variance reduced before the hypothesis can be called two-sided. Independently, conduct a sensitivity sweep across all major drought indices and two agricultural-productivity proxies (NDVI-based and WorldBank crop-yield series), reporting slope estimates as a 2D sensitivity surface; only slope estimates that remain above the AR6 2× threshold across ≥75% of that surface should count as SUPPORTS, while estimates that collapse inside the AR6 range across the same surface provide a genuine, noise-robust FALSIFIES signal.

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 (forming) — 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. forming

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

• UNHCR's annual Revised Budget Appeal, at a $10.7B 2024 baseline, requires an upward revision of 30–50% to $14–16B by the 2027–2031 Strategic Directions cycle if cross-border displacement tracks at 2× the AR6 SSP2-4.5 projection, invalidating the current planning-baseline methodology used for donor pledge negotiations.
• Moody's and S&P sovereign analysts must widen credit-spread assumptions by 80–150 bps for high-vulnerability corridor issuers—Bangladesh, Niger, Ethiopia, and Honduras—within the next annual rating review cycle, as compound drought-driven emigration erodes labor-force and remittance-income assumptions not captured in existing AR6-aligned fiscal stress models.
• Munich Re and Swiss Re must revise annual average loss (AAL) estimates for Sub-Saharan Africa and South Asia compound-drought peril baskets upward by 40–70%, cascading into ILS catastrophe-bond coupon spreads widening 120–200 bps at the next January 1 treaty-renewal window, as migration-correlated agricultural failure becomes a co-trigger in loss models.
• World Bank IDA21 replenishment negotiators face a $20–35B unmodeled financing gap in climate-migration adaptation if compound drought-displacement runs at double AR6 projections, requiring a dedicated migration-resilience sub-envelope to be tabled before the IDA21 final replenishment meeting in 2026.
• The UNFCCC Santiago Network and the COP28-established Loss and Damage Fund require a minimum 3× scale-up in annual disbursement capacity—from approximately $700M pledged at COP28 to $2.0–2.5B per year—within 18 months of a confirmed 2× AR6 cross-border displacement attribution finding, as slow-onset compound events qualifying for fund access will outnumber acute-disaster claims.