Assimilation of atmospheric composition retrievals presents computational challenges due to their high data volume and often sparse information density. Assimilation of compact phase space retrievals (CPSRs) meets those challenges and offers a promising alternative to assimilation of raw retrievals at reduced computational cost (Mizzi et al., 2016). This paper compares analysis and forecast results from assimilation of Terra/Measurement of Pollution in the Troposphere (MOPITT) carbon monoxide (CO) CPSRs with independent observations. We use MetOp-A/Infrared Atmospheric Sounding Interferometer (IASI) CO retrievals and Measurement of OZone, water vapor, carbon monoxide, and nitrogen oxides by in-service AIrbus airCraft (MOZAIC) in situ CO profiles for our independent observation comparisons. Generally, the results confirm that assimilation of MOPITT CPSRs improves the Weather Research and Forecasting model with chemistry coupled to the ensemble Kalman filter data assimilation from the Data Assimilation Research Testbed (WRF-Chem/DART) analysis fit and forecast skill at a reduced computational cost compared to assimilation of raw retrievals. Comparison with the independent observations shows that assimilation of MOPITT CO generally improved the analysis fit and forecast skill in the lower troposphere but degraded it in the upper troposphere. We attribute that degradation to assimilation of MOPITT CO retrievals with a possible bias of ∼ 14 % above 300 hPa. To discard the biased retrievals, in this paper, we also extend CPSRs to assimilation of truncated retrieval profiles (as opposed to assimilation of full retrieval profiles). Those results show that not assimilating the biased retrievals (i) resolves the upper tropospheric analysis fit degradation issue and (ii) reduces the impact of assimilating the remaining unbiased retrievals because the total information content and vertical sensitivities are changed.