Citation Audit Report

Paper: Systemic Risk Channels in Digital Finance
Date: 2026-03-28
Scope: Full paper (18 .tex sections, 160 bib entries, ~485 citation commands)

1. Executive Summary

An ultra-hostile citation audit examined every \citep{} and \citet{} command in the paper against its surrounding sentence and the actual content of the cited work. Two passes were performed:

The root cause was a systematic substitution pattern during LLM-assisted drafting: five cite keys for topically unrelated papers were repeatedly swapped in where canonical systemic risk references belonged. The % EVIDENCE: comments above paragraphs correctly identified the intended sources in most cases, but the actual \citep{} / \citet{} commands used the wrong keys.

2. The Systematic Substitution Pattern

Five cite keys were systematically substituted for canonical systemic risk references. This pattern accounted for the majority of WRONG citations:

3. All Fixes by Section

Below are all 58 fixes organized by .tex file. Each fix shows the before/after and justification. Expand each section to see details.

4. Prevention Framework

A heuristic citation screening script was created to detect future citation problems before they reach the manuscript:

verify_citations.py

Location: scripts/verify_citations.py

Usage:

python scripts/verify_citations.py \
  --tex-dir output/paper/sections \
  --bib output/paper/references.bib

What it does:

1. Parses all .tex files for \citep{} and \citet{} commands
2. Looks up each cite key in the .bib file to get the reference title
3. Extracts the surrounding sentence for context
4. Applies five heuristic flag rules
5. Outputs JSON with all citations and flags + prints a summary

Flag Rules

Example Output

Loaded 160 bib entries from output/paper/references.bib
Found 485 citation commands across 18 .tex files

============================================================
SUMMARY: 53 / 485 citations flagged
============================================================

Flags by rule:
  SUSPICIOUS_YEAR              28
  TOPIC_MISMATCH               17
  AUTHORITY_CLAIM               7
  DUPLICATE_CITE                3

Top-10 most-flagged cite keys:
  Adamyk2025Risk                  6  Risk Management in DeFi: Analyses...
  fsb2023highlevel                5  High-Level Recommendations...
  qin2021empirical                5  Quantifying Blockchain Extractable...
  imf2022gfsr_defi                4  Cryptoasset Risks: Is a Regulatory...
  Cisar2025Designing              4  Designing the future of bond markets...
  Chaliasos2024Smart              4  Smart Contract and DeFi Security...
  gromb2010liquidity              3  Limits of Arbitrage
  Werner2022Sok                   3  SoK: Decentralized Finance (DeFi)
  bis2024tokenisation             3  Tokenisation in the Context of Money...
  Alamsyah2024Review              3  A Review on Decentralized Finance...

5. Remaining Items

14 DECORATIVE citations (not yet fixed)

The audit identified 14 citations in the non-methodology sections that are decorative -- they add no specific evidentiary value but are not factually wrong. These were classified as lower priority and left in place:

Remaining EVIDENCE comment issues

Some % EVIDENCE: comments above paragraphs may still reference the wrong keys (the keys used in the old, wrong citations). These comments are invisible in the compiled PDF but could cause confusion during future editing. A systematic cleanup pass would bring them into alignment with the corrected cite keys.

14 problematic cite keys still in text

The five substitution-pattern keys still appear 14 times in the tex files, but these are now in correct usage contexts -- sentences where the actual topic of the cited paper matches the discussion. For example, Wen2024Foray legitimately appears in the composability risk chapter where DeFi attack synthesis is the topic. The verification script flags 0 of these as TOPIC_MISMATCH.

Short paper sections

The short paper (output/paper_short/) methodology section received 6 fixes. The remaining short paper sections were not audited in this pass.

Script limitations

The verify_citations.py heuristic screener has inherent limitations:

• Cannot verify whether a cited paper's content actually supports the claim -- only whether title keywords overlap
• SUSPICIOUS_YEAR flags produce false positives when recent survey papers legitimately cite earlier findings
• AUTHORITY_CLAIM detection relies on verb patterns and may miss subtler attribution errors
• Uses only the .bib title field (no abstract available) for semantic comparison

Citation Audit Report -- Generated 2026-03-28 -- Systemic Risk Channels in Digital Finance