Software and data artifacts for pets26_coinjoin_analysis (cutoff time 2025-11-10, blockheight 923063)

Code and data artifacts for PETS'26 paper 'CoinJoin ecosystem insights for Wasabi 1.x, Wasabi 2.x and Whirlpool coordinator-based privacy mixers' (FIXME link)

☝️	IMPORTANT: This repository does NOT contain the up-to-date version of code, but only version of the code at the time of paper publication. Visit upstream repository for up-to-date version!

Repository directories structure

• coinjoin-analysis (commit 369fe7c) - Python processing of Dumplings project results. Up-to-date version: https://github.com/crocs-muni/coinjoin-analysis
• dumplings (commit edce638) - .NET project for scanning Bitcoin blockchain for coinjoin transactions. Up-to-date version: https://github.com/crocs-muni/Dumplings/
• API_monitoring - Tool for monitoring Wasabi 2 coordinators based on LiquiSabi with python script for collecting transaction IDs.

Datasets

The following datasets are used as a part of processing.

• Crawled coordinators + manual additions: /coinjoin-analysis/data/wasabi2/txid_coord.json
• Manual false positives: /coinjoin-analysis/data/wasabi1/false_cjtxs.json, /coinjoin-analysis/data/wasabi2/false_cjtxs.json, /coinjoin-analysis/data/whirlpool/false_cjtxs.json
• Processed Dumplings dataset (cutoff time 2025-11-10, blockheight 923063): FIXME zenodo

Replication

Main Results and Claims

Main Result 1: Detection of non-public coordinators

• Means of detection of non-attributed transactions and their prevalence (Figure 6), API monitoring + Dumplings on-chain extraction
• Algorithm for attribution to known cooridnators, connected clusters of unattributed (=> unknown coord)
• Selection of sensible threshold for attribution algorithm (Figure 13, 14)
• Provide expected list of transactions txid_to_coord_discovered_renamed.json + /Scanner/crawl_datasets.png
• Reasoning why it is separate cluster (coordinator) - almost no remix to other coords (Figure 18, Figure 19)

Main Result 2: Prediction of number of participants

• Extraction of prediction factor and error bounds from client-side experiments (linear factor + error bounds + selection of outputs due to better stability)
• Application to coinjoin in time (Figure 10)

Main Result 3: Liquidity statistics for separate coordinator

• Provide expected liquidity_xxx.json files
• Extraction, false positives, reordering, filtering
• Liquidity visualizations

Experiments

Experiment 1: run them all

1. Obtain Linux machine with at least 64G RAM (ideally 128GB). Around 98GB is required at peak (only for very short time, compensation of missing space via swap file is ok)
2. Prepare a base directorie where all the results will be created: e.g. mkdir work_dir
3. Set variable $BASE_PATH to the working directory -- e.g. export BASE_PATH=/home/user/work_dir
4. Clone this repository into $BASE_PATH/:

cd $BASE_PATH
git clone https://github.com/crocs-muni/pets26_coinjoin_analysis.git

5. Download dumplings.zip into $BASE_PATH

TODO

6. Copy coinjoin-analysis into $BASE_PATH

cp -r $BASE_PATH/pets26_coinjoin_analysis/coinjoin-analysis $BASE_PATH

7. Create python virtual environment

python3 -m venv ./venv
source ./venv/bin/activate

8. Install python requirements for coinjoin-analysis project:

cd $BASE_PATH/coinjoin-analysis
pip install -r requirements.txt

9. Run processing scripts (~4 hours on Debian Linux, 13th Gen Intel(R) Core(TM) i7-13700KF)

cd $BASE_PATH/coinjoin-analysis
./scripts/run_pets.sh

Investigate results

Inspect results produced in /btc/dumplings_temp2/Scanner/

• Figure 1 and 2 are manually drawn illustration diagrams
• Figures 3-5 are composition of ww2_as25_38_privacy_progress.png, ww2_as25_38_anonscoregain.png, ww2_realexp_num_inputs.png, ww2_realexp_num_outputs.png, ww2_realexp_as25_inoutheatmap.png, ww2_realexp_as38_inoutheatmap.png (generated from non-public dataset not shared due to privacy reasons as explained in paper)
• Figure 6 is /Scanner/crawl_datasets.png
• Figure 7 is composition of two versions of the same file with SORT_COINJOINS_BY_RELATIVE_ORDER=False (upper graph) and SORT_COINJOINS_BY_RELATIVE_ORDER=True (lower graph) /wasabi2/2023-03-01 00-00-00--2023-04-01 00-00-00_unknown-static-100-1utxo/wasabi2_input_types_nums_notnorm.png
• Figure 8 from paper is composition of results /wasabi1_zksnacks/2019-12-01 000-00-00--2020-01-01 00-00-00_unknown-static-100-1utxo/wasabi1_zksnacks_input_types_values_notnorm.png and /wasabi1_zksnacks/2020-10-01 00-00-00--2020-11-01 00-00-00_unknown-static-100-1utxo/wasabi1_zksnacks_input_types_nums_norm.png
• Figure 9 is composition of /wasabi1/wasabi1_zksnacks_cummul_values_norm.png, /whirlpool/whirlpool_cummul_nums_norm_nolegend.png, /wasabi2/wasabi2_cummul_values_norm_nolegend.png
• Figure 10 is composition of /wasabi2_zksnacks/wasabi2_zksnacks_wallets_predictions_dynamics.png and /wasabi2_kruw/wasabi2_kruw_wallets_predictions_dynamics.png
• Figure 11 is small part of /wasabi2_opencoordinator/2025-03-01 00-00-00--2025-04-01 00-00-00_unknown-static-100-1utxo/wasabi2_opencoordinator_input_types_values_notnorm.png in days 9-11.3.
• Figure 12 is ww2_coord_flows_values.png
• Figure 13 is /Scanner/crawl_all_coordinators_in_out_boxplot.png
• Figure 14 is composition of crawl_in_out_ratio_over_time__kruw.png, crawl_in_out_ratio_over_time__gingerwallet.png, crawl_in_out_ratio_over_time__opencoordinator.png, crawl_in_out_ratio_over_time__wasabicoordinator.png from /Scanner/ folder
• Figure 15 /Scanner/all_coord_discovery_analysis___drop__randomsingle_aggregated.png
• Figure 16 /Scanner/all_coord_discovery_analysis___drop__randomany2_aggregated.png
• Figure 17 /Scanner/all_coord_discovery_analysis___drop__tail_aggregated.png
• Figure 18 is composition of /wasabi2_unknown_2024_e85631/wasabi2_unknown_2024_e85631_cummul_values_norm.png and /Scanner/discovered_in_out_ratio_over_time__unknown_2024_e85631.png
• Figure 19 is composition of /wasabi2_unknown_2024_28ce7b/wasabi2_unknown_2024_28ce7b_cummul_values_norm.png and /Scanner/discovered_in_out_ratio_over_time__unknown_2024_28ce7b.png
• Figure 20 is as25_as38_wallet_predict_confidence.png (generated from non-public dataset not shared due to privacy reasons as explained in paper)
• Figure 21 is composition of /wasabi2_zksnacks/wasabi2_zksnacks_wallets_predictions_drops.png and /wasabi2_kruw/wasabi2_kruw_wallets_predictions_drops.png
• Figure 22 is composition of ww2_realexp_as25_fees.png and ww2_realexp_as38_fees.png (generated from non-public dataset not shared due to privacy reasons as explained in paper)
• Table 2 is generated from non-public dataset not shared due to privacy reasons as explained in paper
• Table 3 is generated from /Scanner/liquidity_summary_xxx.json files where xxx is name of a pool and with usage of "latex_summary", "earliest_cjtx", "earliest_time", "latest_cjtx", "latest_time" and "total_coinjoins" keys.
• Table 4 is generated from /Scanner/liquidity_summary_xxx.json files where xxx is name of a pool and with usage of "earliest_cjtx" and "latest_cjtx" keys.

Limitations

The replication of the following results and datasets extraction is not We are not able to

• Client-side characteristics extraction due to non-public client-side dataset
• Dumplings extraction dataset (due to change in time after cutoff date)
• API dataset collection not repeatable - coordinators and collection services no longer operational