This manuscript deals with stringent problem in conflict prone Central Africa: Data scarceness, data gaps and sometimes errenous recordings.

While the technical description of the recovery period and the trend anaylis provide an excellent basis for future research, I miss a statement on possible reasons for differences in trends throughout the paper. At least one paragraph in the discussion (e.g at the end of section 4.6) exploring possible drivers of the differences in trends would round up the manuscript. Is it just differences in manifestation of global circulation changes due to climate change? Or it could be due to differences in Land Use Change around the site or further away? May it be attributable to logging or agriculture, to mining or construction, etc. ...  . Emphasising such issues would greatly increase the possible audience and the impact of this important recovery mission. 

The article documents, recovers and analyses historical weather records of Congo basin in present-day DRC. The weather records considered in this analysis are primarily from 1960s to 1990s, some records extending into early 2000s. The study undertakes digitalisation (imaging) of 9,885 sheets of paper from 37 weather stations in DRC. Using previously developed ML-based data transcription python-module ‘Meteosaver’ to recover more than one million observations of temperature and precipitation. The resulting dataset consists of daily maximum, minimum, and average temperatures (dry bulb and wet bulb) measured three times a day, and daily precipitation.

The analysis was shown to have marked warming in daily maximum, minimum, and average temperatures, corresponding to approximately 0.7°C, 0.3°C, and 0.5°C, respectively, of warming during the 1961–1990 period. It was also reported that increasing frequency of hot days and fewer cold days as compared to earlier decades in the dataset. Due to limited rainfall data, only two stations were found to experience increased frequency of heavy rainfall. Though the changes inferred from this study can not be applied to whole of DRC due to limited spatial coverage, but this study enhances our understanding of changes in climate in DRC, one of the most data sparse region in the world.

Specific comments:

The article is generally well-written and undertakes substantial data-rescue work to produce climate dataset and analyse past climate of Congo basin. The authors describe the current state of data (un)availability in Central Africa for understanding climate change in the region. They use numerous paper records archived in DRC and Belgium to fill in the data gap, and the sources and paper format is described in sufficient detail to understand background of the work being undertaken. The dataset produced has potential to answer many questions regarding scale and impact of climate change in the equatorial Africa.

This paper presents historical climate dataset for DRC, using ML-based transcription process and using statistical QC/QA steps to ensure data accuracy. While ML-based transcriptions have been researched in number of fields, using it to produce a published climate dataset of this size is probably first documented case. This attempt opens a realistic possibility of using ML-based transcriptions to extract historical observations from elsewhere. 

The region’s climate has been quantitatively assessed by analyzing long-term in-situ weather observations for the first time. The analysis performed within the limited spatiotemporal dataset shows that the climate has gotten warmer and maybe wetter as well. The results based on limited data support conclusions drawn, subject to further investigations.

Though the paper transcribes and analyses historical weather observations, discussion on observation techniques or instrument metadata has been largely absent. As the biases and offsets are often strongly linked to weather measurement techniques, e.g. how temperature scale was read, was a correction scale was provided/used, bucket size of rain gauge etc. The instrument make and model can also be helpful in assessing systematic biases and can be used to perform QA/QC checks and corrections.

Some discussion on why we expect to find these temperature or rainfall trends in Central Africa will be helpful. Has any earlier studies found similar trends either in observational data or model data?

The manuscript titled “From Paper to Proof: Revealing Congo Basin Warming Through Rescued Climate Archives” by Derrick Muheki et al. presents a valuable community resource by rescuing and digitizing historical meteorological observations from the Democratic Republic of the Congo (DRC). The development of a QA/QC workflow and the creation of a long-term observational dataset with more than 1 million observations from 37 stations represent an important contribution to climate research in a data-sparse region.

The manuscript is well written, and the workflow and results are clearly presented. The comments below are primarily intended to improve methodological clarity and reproducibility.

1. The dataset combines observations collected over several decades across multiple weather stations. During this period, instrumentation, measurement precision, calibration procedures, and observational practices may have changed. It would be helpful if the authors could comment on whether such temporal and interstation inconsistencies were assessed or accounted for (e. g., through homogenization or other quality-control procedures), and discuss their potential influence on the derived long-term climate trends.
2. The QA/QC procedure uses a fixed uncertainty range of 0.2°C for temperature and 0.2% for relative humidity. Could the authors comment on how these thresholds were selected? In particular, were these based on empirical validation, rounding, or sensitivity analysis. Also, is it uncertainty in the range of the reported variable or an acceptance criterion?
3. In section 2.3.4, the authors mention that when U, e, and Δe are internally consistent but the reconstructed dry-bulb temperature, T, differs from the transcribed value, the algorithm replaces the transcribed T using equations (5) and (6). This appears to assume that the discrepancy is mainly attributed to T rather than to uncertainties in e and Δe. Since small uncertainty in e and Δe can also propagate into the reconstructed T, could the authors comment on why an asymmetric criterion is considered.
4. In section 2.4, the authors mention that the software obtained transcription for 37 stations was validated against manually transcribed records from five stations. It would be helpful to discuss how representative these stations are of the broader archive in terms of climatic regions, handwriting styles, and temporal coverage.
5. In section 3.1, the authors mention that 1, 011, 297 records achieved the “highest internal quality flag.” Could the authors briefly describe the flagging scheme used here. Also, the confirmed fraction for precipitation is substantially lower (30%). It would be useful to clarify how this lower confirmation rate affects the reliability and representativeness of the precipitation analyses.
6. In the methodology section, the authors describe the construction of kernel density estimation (KDEs) for individual stations, but it is not clear how the aggregated temperature distributions (Fig. 10) are obtained from these station-level KDEs. In particular, whether the aggregated distributions are generated by pooling all daily observations or by combining station-level distributions, and whether any weighting is applied to account for differences in data availability among stations.

Overall, I appreciate the authors for their substantial effort involved in rescuing, digitizing, quality-controlling, and analyzing the historical meteorological dataset for the DRC. The dataset fills an important observational gap and will provide a useful resource for future climate and environmental studies. I believe the manuscript is suitable for publication after the authors have addressed the above mentioned minor points.