Skip to content

FORWARD_NOTE.md

Latest commit

 

History

History
76 lines (58 loc) · 5.06 KB

FORWARD_NOTE.md

File metadata and controls

76 lines (58 loc) · 5.06 KB
Raw

CBDC repo fix bundle — what to merge and why

To: Andrew Maksakov Repo: github.com/andrewmaksakov/CBDC Bundle assembled: 21 July 2026 (from cbdc-privacy/ local canon)

What's in this bundle

  1. references.bib — the corrected bibliography (replaces the repo's copy wholesale)
  2. detection/ — the Section 4.5 detection-simulation validation pipeline (code, README, seeded results)

Why the bib replacement (item 1)

The references.bib currently in the repo (HEAD 2b177db, February) carries five fabricated author sets that were corrected locally but never propagated to the public surface:

Key Repo (wrong) Corrected (verified against live source)
OfflineCBDC2024 "Seres, Istvan Andras and others" Michalopoulos et al.
ZKPSurvey2025 "R, Anitha and others" Shashidhara et al.
UFLaw2024 "Golumbia, David and others" Jiang
OxfordCBDCSurvey2025 "Chen, Karen and others" Plato-Shinar, Maman, Shema Zaltokrilov, Yaacobi
Koti2024Graphiti seven invented names ("Nishanth Koti, Parth Kukkala, Ajith Patra, ...") Nishat Koti, Varsha Bhat Kukkala, Arpita Patra, Bhavish Raj Gopal (IACR ePrint 2024/1756, fetched 21 Jul; also published at ACM CCS 2024)

Also in this copy (all verified against Crossref/IACR/publisher on 21 Jul):

  • Campanelli2017 renamed to Campanelli2019 (LegoSNARK is CCS 2019; cite sites in main.tex updated)
  • Wang2025UnbalancedPSI title aligned to the published PoPETs form, "Unbalanced PSI from Client-Independent Relaxed Oblivious PRF" (Crossref 10.56553/popets-2025-0109)
  • choi2025cbdcprivacy completed with IER 66(2), 823–847; its BIS working-paper twin BIS2024Privacy is no longer cited in the manuscript (same Choi/Kim/Kim/Kwon survey — was double-cited as two independent sources; entry retained in the bib with a note)
  • ECBConsultation2021 note corrected: the consultation had 8,221 responses and privacy was ranked the most important feature by 43% of respondents (the old "41% focused on privacy" claim doesn't match the ECB's own report)
  • Eight entries appended 21 Jul (positioning pass, all verified against Crossref/IACR/ACM/BIS the same day): CamenischMaurerStadler1996, Camenisch2005CompactEcash, Garman2016Accountable, Wuest2022Platypus, Kiayias2022PEReDi, Buterin2024PrivacyPools, BISAurora2023, BISHertha2025. These back the repositioned Sections 2.2--2.4 (the cryptographic-compliance lineage and the BIS AML programs the paper now engages; see CHANGES-FOR-ANDREW.md in the paper folder). The bundle bib includes them so it remains a wholesale replacement for the repo copy.

A bibliography with invented author names is the kind of thing that sinks a submission on integrity grounds regardless of content — this needs to reach the public repo before anyone follows the paper's pointer to it.

Why detection/ (item 2)

The paper's Data-Availability statement and Section 8.3 point readers to the repo for the Section 4.5 detection-simulation code. The repo currently has only pet_aml_sim.py (the Section 5.4 latency/throughput stack sim — no classifier, no AUC code), so the availability claim is false until detection/ lands. The manuscript has been reworded in the interim to say the detection code "is included in the submission supplement and will be merged into the public repository" — merging this directory is what makes that sentence permanently true.

The pipeline is self-contained: run_all.py (seeded, seed 20260707), test_pipeline.py (regression tests incl. a degeneracy audit that hard-fails on the old num_wallets artifact), and results/ with the committed CSVs/JSONs the README numbers trace to. Exclude nothing; __pycache__ is already stripped.

What NOT to merge yet

  • main.tex — the repo copy is the stale February version, but don't overwrite it with the current local one either: the Section 4.5 empirical story is being rewritten around the detection pipeline results (Murad is steering that). Repo main.tex gets replaced when the rewrite lands.

Heads-up on your sim (separate from this bundle)

pet_aml_sim.py has an ordering bug worth fixing before the next run: risk_tier is read inside the transaction loop (travel-rule check and escalation, ~lines 468/475) but only assigned by run_batch risk propagation after the loop completes — so tier-based escalation can never fire and "zero escalations" is guaranteed by construction, not by the architecture. The paper's §5.4.3 interpretation has been reworded to say exactly that. A fix + seeded rerun (and a look at the 36.8% rejection-rate calibration) would let Table 3 and §5.4.3 carry an honest escalation number. Your artifact, your sign-off — happy to talk through it.

Merge commands (from a clone of the repo)

cp /path/to/repo-fix-bundle/references.bib references.bib
rsync -a /path/to/repo-fix-bundle/detection/ detection/
git add references.bib detection
git commit -m "Fix fabricated author sets in references.bib; add Section 4.5 detection validation pipeline"
git push origin main