In this study, the authors attribute uncertainty in the radiative forcing of aerosol-radiation interactions (RFari) by proposing a “constrained parameter” (CP) method, based on Monte-Carlo calculations that calculates sensitivity of RFari to uncertain input parameters. They find that refractive index and size distribution are associated with the largest sensitivities. CP-derived sensitivities differ compared to “one at a time” (OAT) methods for some parameters, although the impact of covariances seems small overall.

The study is interesting because it attributes uncertainty more finely (i.e. to more fundamental parameters) than previous studies. It would however require major revisions before publication because the description of the state of the art is outdated, the discussion of the strength of the CP method needs clarifying, and the description of the authors’ calculations is not sufficiently detailed.

Main comments:

• The authors criticise previous RFari uncertainty attribution work for their lack of account for covariance between uncertain factors and their inability to quantify contributions to total uncertainty. But their literature review stops in 2018, and studies published since then have addressed many of the authors' concerns, including using methods very similar to their proposed CP method. Thorsen et al. (2020, 2021) https://doi.org/10.1175/Jcli-D-19-0669.1 https://doi.org/10.1175/Jcli-D-19-1009.1 and Elsey et al. (2024) https://doi.org/10.5194/acp-24-4065-2024 (and references therein) are particularly relevant. The authors need to place the motivation and results of their study in the current state of the art.
• The list of challenges in section 2.2 is interesting, but it is often unclear why they really challenge uncertainty attribution. Challenges 1 and 5 are real issues but would pose a problem for any uncertainty attribution methodology, including CP. Challenges 2 and 3 are real, but temporal variations do not necessarily translate to variations in uncertainty contributions. Challenges 4 on covariances is real, but crucially depends on which factors are considered and the size of the covariances. So the relevance of each challenge to uncertainty attribution needs to be clarified by extending the discussion.
• Section 3.2 is incomplete and would not allow to reproduce the authors’ calculations. First, it is unclear what the central values given in Table 1 are supposed to represent, and therefore which aerosols the uncertainty analysis is relevant to. A typical industrial aerosol? A typical Chinese aerosol? In addition, the uncertainty ranges (the sigmas) are missing for all parameters, but are obviously crucial to apply both the OAT and CP methods. They must be added to Table 1.

Other comments:

• Abstract, lines 11-12: Need to make the description of the CP method clearer. The current description could apply equally to the OAT method.
• Line 25: RFaci is missing from the list of strong radiative forcing. It is probably much stronger than RHari
• Line 27: Do you mean “has not undergone substantial revisions”? Otherwise, the “Despite significant advances” does not work with the sentence.
• Lines 28-31: Are those model-observation differences still actual? I think the latest IPCC report concluded those disagreements were mostly resolved now.
• Line 25: The list of previous studies is missing recent work that already addresses the shortcomings of older work.
• Section 2.1 is difficult to understand because many variables are not defined: the exponent T, function g, A, the arrow operator, …
• Line 118: Why “linear trend” specifically?
• Caption of Table 1: I suggest deleting “Observation” in the caption, since many data sources are either not from observations or not directly relevant to the observation site.
• Lines 251-253: It would be interesting to also look at the sensitivity of RFari to AOD, SSA and g, to compare with similar studies.
• Section 3.3.2 would work more logically before section 3.3.1. Describe the physical meaning of the results before the OAT/CP differences. Overall the ranking of sensitivity follow expectations, given that Mie theory is used, but I am surprised that coating thickness is important to sensitivity, but kappa is not despite directly affecting aerosol size. Why is that?