Red Team — Fragility Analysis

Adversarial review of failure modes, fragile assumptions, and near-limit validations. Use this page to probe the model's weakest points.

Operational Warning: This page surfaces the most fragile elements of the CE model stack. Items listed here represent genuine epistemic uncertainty, contested empirical territory, or calibration choices that are known to be sensitive. These are not bugs — they are disclosed risks.

Fragility Items (14)

assumption A-CLM-001
Alternative approaches
  • Use ECS with explicit ocean heat uptake parameterisation
  • Use TCRE (transient climate response to cumulative emissions) directly
assumption A-CLM-002
Alternative approaches
  • Use GCP 2024 estimate of 54–56 GtCO₂ (CO₂-only); note non-CO₂ gases add ~10 GtCO₂e
assumption A-DAM-001
Alternative approaches
  • Burke et al. (2015) non-linear damage function — gives ~23% global GDP loss at 3°C
  • Weitzman (2010) tipping-point adjusted damages
  • META model (Dietz et al. 2021) — explicit tipping cascade
assumption A-DAM-002
Alternative approaches
  • Fully dynamic hazard modelling using CMIP6 GCM outputs at peril-level
assumption A-ECO-003
Alternative approaches
  • Partial efficiency with explicit coverage rate and leakage parameters
assumption A-TECH-001
Alternative approaches
  • Conservative scenario: only TRL≥7 technologies achieve significant deployment by 2040
assumption A-TECH-002
Alternative approaches
  • No-BECCS scenario (Dooley et al. 2018)
  • High-BECCS scenario (IEA NZE low-land-use variant)
assumption A-UNC-001
Alternative approaches
  • Correlated sampling using Cholesky decomposition; pairwise correlations from IPCC scenario ensemble
assumption A-GOV-001
Alternative approaches
  • Realistic implementation scenario with 60–70% NDC delivery rate
parameter P-DAM-001
parameter P-ECO-002
parameter P-UNC-002
parameter P-FIN-001
validation_warning VAL-ECO-002

Sensitive Equations

IDNameCategorySensitivityUncertaintyAssumptions
EQ-CLM-001 Transient Climate Response (TCR) warming estimate Climate CRITICAL TCR uncertainty (±0.6°C at 1σ) dominates temperature pathway uncertainty up to 2050 TCR is treated as a single representative scalar (best estimate 1.8°C, likely 1.2–2.4°C per AR6); Instantaneous equilibration of forcing to temperature (ignores ocean heat uptake lag); Log-linear forcing from CO₂ only; non-CO₂ forcings captured in scenario multipliers
EQ-CLM-002 Carbon budget remaining (integrated emissions constraint) Climate CRITICAL Budget uncertainty ±220 GtCO₂ at 1σ (IPCC AR6); dominated by TCRE distribution Linear interpolation of annual emissions between scenario waypoints; Non-CO₂ forcing converted to CO₂-equivalent using GWP100 from AR6; Permafrost carbon feedbacks partially included per AR6 Table SPM.2 footnotes
EQ-CLM-003 Emissions pathway abatement requirement Climate HIGH Reference pathway uncertainty ±3 GtCO₂e/yr; target pathway dependent on budget choice Reference pathway is 'current policies' from UNEP EGR 2024 (57.4 GtCO₂e in 2025); Linear ramp assumption for abatement deployment between waypoints; LULUCF net emissions treated separately from energy-system abatement
EQ-DAM-001 DICE-style damage function (quadratic) Physical Damage CRITICAL Damage function specification is the single largest uncertainty in long-run economic cost; estimates vary by 10× across literature Damages are symmetric around global mean temperature (ignores distributional heterogeneity); Quadratic form implies accelerating but bounded damage — challenged by tipping-point literature; α = 0.00267 from DICE-2023; alternative calibrations range 0.001–0.01; Does not capture catastrophic or non-linear tipping point damage
EQ-DAM-002 Annualised average loss (catastrophe actuarial) Physical Damage HIGH Model-to-model AAL variation ±30–50% across major cat model vendors for any given region/peril Poisson process for event occurrence (memoryless, independent events); Stationarity of hazard distribution (violated under climate change — CE applies non-stationarity adjustment); Vulnerability functions are sector-constant within broad industry categories
EQ-ECO-001 Marginal abatement cost curve (MAC) — carbon price equilibrium Economics CRITICAL Carbon price range for 1.5°C: $50–250/t by 2030 (IPCC AR6 Table 13.SM.2) Smooth, convex MAC curve — in practice, MAC curves have discontinuities at technology step changes; Perfect competition in carbon markets (no market power, no transaction costs); MAC curve calibrated to IEA NZE scenario technology cost projections
EQ-ECO-002 Structural growth with climate adjustment (augmented Solow) Economics HIGH Ω specification drives long-run GDP uncertainty; α uncertainty ±0.05 has minor near-term impact Constant returns to scale in K and L; Capital income share α = 0.35 (OECD average; varies significantly across developing economies); Climate damage enters multiplicatively (not additively) — implies no threshold catastrophe
EQ-ECO-003 Social Cost of Carbon (Ramsey discounting) Economics CRITICAL SCC values range $15–$400 in mainstream literature; discount rate choice is fundamentally an ethical/political decision Ramsey utility discounting with CRRA utility function; SCC range driven almost entirely by choice of ρ: Nordhaus ≈$20, Stern ≈$200, EPA 2023 ≈$190; Damage function specification second-largest driver (see EQ-DAM-001)
EQ-FIN-001 Weighted average cost of capital (WACC) — climate-adjusted Finance HIGH Climate risk premium estimates vary 0–300bps across methodologies; sector-specific calibration adds ±100bps Climate risk premium is additive to baseline WACC; Physical risk premium calibrated to cat bond spreads and TCFD disclosures; Transition risk premium calibrated to fossil fuel stranded asset write-down scenarios
EQ-UNC-002 Portfolio abatement de-duplication factor (overlap discount) Uncertainty HIGH δ range 0.10–0.35 translates to ±5–7 GtCO₂e/yr uncertainty in net abatement at 30 Gt gross δ = 0.22 is a central estimate; empirically poorly constrained; Covariance structure: ρ_elec = 0.21 for energy technologies; ρ_CDR = 0.25 for carbon removal; Linear scaling of overlap with portfolio size — likely underestimates overlap at high ambition levels

Model Decision-Grade Gaps

IDNameClassificationValidationKey Limitation
MDL-006 CatastropheModel Scenario, Reference Benchmarked to historical insured loss records 1980–2023 (Swiss Re sigma) Vendor AAL spread ±40% is not propagated to UI — single central estimate shown
MDL-007 DSGEModel Scenario, Experimental Not validated for climate scenarios — validated only for standard macro shocks (RMSE <1.2% GDP for 2010–2023 historical period) Experimental classification — DSGE parameters are not empirically identified for climate shocks