[FAIR v2.2.2] CO₂ Emission Scenarios Produce Identical Results Across Configurations

[cccccwendy]

I am currently working with FAIR version 2.2.2, downloaded from GitHub, to simulate temperature trajectories under two CO₂ emission pathways: SSP1-1.9 and SSP2-2.6, each with and without added military emissions. However, I am encountering an issue where all scenarios produce virtually identical outputs, despite significant differences in the underlying emissions input data.

My setup includes:
Emissions for CO2 FFI provided manually via fair.emissions.loc[...], converted to GtC.

CO2 AFOLU emissions set to zero for simplicity.

Configs defined as a single model: "ACCESS-ESM1-5_r1i1p1f1", applied to all scenarios.

Species and properties loaded via read_properties() from the default species_configs_properties.csv.

Climate configs filled using 4xCO2_cummins_ebm3.csv with fill() as in the official notebooks.

The .run() command is used once after all inputs are loaded.

No errors or NaNs are reported in any emission array or climate config.

The unexpected behavior:
Both SSP1-1.9 and SSP2-2.6 temperature trajectories are virtually identical .

Mean emissions for each scenario differ (I verified this), but fair.temperature output is flat and overlapping.

This occurs regardless of whether I use .mean().item() or plot the entire time series.

My key questions:
Could this be caused by using the same config for all scenarios? I originally intended to isolate policy effects.

Is it mandatory that each scenario be paired with a unique config, even when only emissions vary?

Can fair.run() differentiate between scenarios if configs are identical?

Could any implicit overwrite be happening due to xarray broadcasting behavior?

Any insights you could provide would be greatly appreciated. I would also be happy to share my code or Jupyter notebook if that would help.

Thank you for maintaining such a valuable open-source model.