Editing Openai/6938382f-f9e4-8003-a158-6d812c1bce84 (section)

===== 5.1 Misuse of error propagation / confidence intervals =====
* Random-walk propagation of a structural bias: Systematic LWCF bias is treated as a per-year independent uncertainty. That’s akin to saying: “My thermometer is off by ±1°C, therefore after 100 measurements my uncertainty in the final average is ±10°C,” which is not how systematic error works.
* Ignoring constraints from model behavior: If ±15°C uncertainty were real, ensembles under fixed forcing would quickly diverge to wildly different climates. They don’t.

Even Roy Spencer (who is generally skeptical of mainstream climate sensitivity estimates) published a detailed critique arguing Frank’s treatment of cloud-error propagation is not valid and leads to absurd consequences. drroyspencer.com<ref>{{cite web|title=drroyspencer.com|url=https://www.drroyspencer.com/2019/09/additional-comments-on-the-frank-2019-propagation-of-error-paper/|publisher=drroyspencer.com|access-date=2025-12-10}}</ref>