Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Distinguishing Scams and Fraud with Ensemble Learning

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that an ensemble of three LLM prompts—Gemini on the trust-based definition and GPT-4 on two complementary questions—can label CFPB consumer complaint narratives as scam versus non-scam fraud with precision 0.95 to 0.97…

desk verdict Useful first organic scam/fraud corpus and an honest error analysis, but the headline performance numbers are in-sample and should not be read as validated. read the letter →

arxiv 2412.08680 v1 pith:UNQURZS7 submitted 2024-12-11 cs.CR cs.AIcs.HCcs.LG

Isha Chadalavada , Tianhui Huang , Jessica Staddon This is my paper · ORCID
classification cs.CRcs.AIcs.HCcs.LG
keywords LLMensemblescamdetectionfraudcomplaintsCFPBdatabasepromptengineeringconsumernarrativeserroranalysischain-of-thoughtprompting
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper sets out to separate two kinds of consumer harm in the CFPB complaints database: scams, where the victim is tricked into authorizing a harmful transaction, and non-scam fraud, where the transaction is unauthorized. The authors build a manually labeled set of 300 complaints, iterate on prompts for Gemini and GPT-4, and combine the two models with an 'all must say scam' rule. On that labeled set the ensemble reaches precision 0.95 and recall 0.84, and a manual check of a random 10% of its positive predictions (133 narratives) finds precision 0.97. The paper also reports error patterns: both models lean on secondary details such as customer-service disputes and claim denials, treat company names as signs of reputation, and perform worse on very long or heavily redacted narratives. If correct, these findings give researchers a high-precision corpus of organic scam narratives and concrete guidance for how consumers and developers should use LLMs for scam defense.

What carries the argument

The load-bearing mechanism is the conjunction ensemble: scam is predicted only if all three prompts—Gemini with the trust-based definition, GPT-4 with the 'money stolen vs. tricked vs. neither' question, and GPT-4 with the 'potential scam vs. not scam' reputation question—independently classify the narrative as a scam. This AND rule is what buys precision, at the price of recall. The prompts themselves rely on chain-of-thought reasoning, asking each model to explain its label, and on a small set of in-prompt examples. The ensemble is built for the specific 'fraud or scam' issue tag in the CFPB database and is not a general scam detector: on complaints without that tag, precision and recall drop to 0.25 and 0.33.

What would settle it

Manually label a random sample of 'fraud or scam' narratives using a codebook that requires objective evidence of deception (e.g., trust-building followed by a voluntary transfer to an unknown party), then compare the ensemble's precision on narratives that contain the word 'scam' versus those that never use the word; a large drop on the word-free subset would show the model detects the label rather than the scam.

Watch

Extended reading notes

Core claim

The central claim is that a simple ensemble of three LLM prompts, evaluated with a conjunction rule, distinguishes scam from non-scam fraud in CFPB complaint narratives at practically usable precision. The definition used is that a scam involves the complainant being tricked into voluntarily taking a self-harming financial action, whereas non-scam fraud is an unauthorized action; the prompts encode this distinction and require each model to explain its answer. The final ensemble, denoted F, predicts 'scam' only when Gemini (Prompt A), GPT-4 with the two-option question (Prompt B), and GPT-4 with the reputation-based question (Prompt C) all agree. On the 300 manually labeled narratives L, F achieves precision 0.95 and recall 0.84; applied to 2,569 recent 'fraud or scam' narratives, it labels 1,333 as scams, and a manually reviewed random 10% sample of those positive predictions shows precision 0.97. The authors state that this is the first corpus of organic consumer scam narratives, as opposed to the synthetic scam narratives used in earlier LLM evaluations.

Load-bearing premise

The labeling rule that 'the customer is always right' treats complainants' self-reports of being scammed as ground truth, so if people call ordinary fraud or poor service a 'scam' without objective evidence of trickery, the model's high precision could reflect keyword matching rather than genuine scam detection.

Editorial extensions

If this is right

  • A high-precision subset of the CFPB database can now be treated as a corpus of organic scam narratives, supporting future measurement of LLM scam-defense behavior.
  • The error-pattern analysis implies that consumers should keep narratives focused on interactions with the suspected scammer and should simulate or generalize personal information rather than redact it, since redaction hurts performance and long off-topic text increases errors.
  • Developers fine-tuning LLMs for scam defense should treat company names and secondary events like claim denials as unreliable signals, because both models leaned on them in error explanations.
  • The ensemble's success on tagged 'fraud or scam' complaints does not transfer to untagged complaints, so the CFPB's issue tag or an equivalent filter is a necessary precondition for the method.
  • The observed length effect (performance degrades beyond roughly 3,000 characters) gives a concrete rule of thumb for prompt design and user guidance.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The 'customer is always right' labeling rule likely makes 'scam' labels correlate with the presence of the word itself; a held-out test using narratives that describe deception without ever saying 'scam' would clarify whether the model detects trickery or a keyword.
  • The reliance on secondary evidence suggests a latent bias: complaints that mention poor customer service or denied claims may be systematically over-labeled as scams, which could matter if such models ever inform consumer advice.
  • The conjunction design implies a precision/recall trade-off curve; varying the number of agreeing models or the strictness of individual prompts would map that curve and may show that two models already suffice under some operating points.
  • Because the model only works on narratives already tagged 'fraud or scam', a practical deployment would still need a first-stage classifier or user self-selection; extending the ensemble to untagged narratives is an open problem the paper does not solve.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper develops an LLM ensemble prompt (Gemini with Prompt A, GPT-4 with Prompts B and C, combined by conjunction) to distinguish scam from non-scam fraud complaints in the CFPB database. The authors manually labeled 300 'fraud or scam' narratives (set L) through an iterative process, report that the ensemble F achieves precision .95 and recall .84 on L, apply F to 2,569 narratives from 2024 to identify 1,333 scams, and manually check a 10% sample of predicted positives (n=133) with precision .97. They also analyze error patterns related to secondary information, company reputation, narrative length, and redaction, and they note that F performs poorly (precision .25, recall .33) on complaints not labeled 'fraud or scam'.

Significance. If the performance claims are trustworthy, the contribution is valuable: an organic corpus of consumer scam narratives, an ensemble prompting technique, and an error analysis that identifies concrete weaknesses of LLMs in scam detection. The paper is transparent about some limitations, such as the model's specialization to the 'fraud or scam' subset. The main strengths are the manual labeling effort, the explicit prompting details, and the initial error-pattern findings. However, the central quantitative claims are undermined by the evaluation protocol, which is in-sample for the headline precision/recall and measures only precision externally. The contribution is therefore promising but not yet fully supported.

major comments (3)
  1. [Section 3, 'Prompt Design and Iteration' and 'Final Prompt and Performance'] The reported precision of .95 and recall of .84 on set L are in-sample estimates. The text states that prompts were iteratively revised after evaluating against L and manually reviewing errors on L, so the final ensemble F and its conjunction rule were selected using L. Reporting performance on the same data used for model selection is fitting-then-evaluating; these numbers are optimistically biased and cannot be interpreted as expected performance on new complaints. The paper should either hold out a portion of L for final evaluation or clearly reframe the .95/.84 as training-fit diagnostics and base the capability claim on an out-of-sample evaluation.
  2. [Section 3, 'Final Prompt and Performance'] The external evaluation—manual review of a randomly selected 10% sample (n=133) of the 1,333 predicted positives—measures only precision on predicted scams. It does not measure recall because false negatives are never sampled. Since the paper's headline capability claim includes recall of .84, and since the in-sample recall is not a valid estimate, the paper currently provides no evidence about out-of-sample recall. The authors should either annotate a random sample of all 2,569 narratives (or a random sample of narratives not predicted as scam) to estimate recall externally, or explicitly state that out-of-sample recall is unknown.
  3. [Section 3, 'Training data'] The 'customer is always right' labeling rule makes the gold-standard labels partly a function of complainant assertions. If a complainant writes 'I was scammed' with no further detail, the label is 'scam' by construction. This creates a potential confound: F may achieve high agreement with L by detecting explicit scam language or complaint structure rather than by identifying deceptive schemes. The paper should test this, for example by reporting performance on the subset of L where the narrative does not contain the word 'scam' or explicit scam assertions, or by discussing why such a test is not feasible. Without this, the high scores on L could reflect superficial cue detection.
minor comments (5)
  1. [Section 3, 'Training data'] The phrase 'high inter-related reliability' appears to be a typo for 'inter-rater reliability'; please correct.
  2. [Section 3, 'Training data'] There is a typo in 'complainaint' in the sentence 'the complainaint's statements would be assumed correct'; please correct to 'complainant'.
  3. [Section 3, 'Prompt Design and Iteration'] The sentence 'Several prompts included example complaints and labels since examples have been associated with improved performance in other contexts (e.g., [21])' cites a survey of LLM annotation; a more specific reference on few-shot prompting would strengthen the claim.
  4. [Section 4, 'Performance and Narrative Length'] Figure 1 is described in text but the caption only says 'Precision and recall of the final model, F, as a function of complaint narrative length in characters.' It would help to state the binning or smoothing method used to produce the curves, since the underlying data are discrete narrative lengths.
  5. [Section 5] The sentence 'We evaluated the prompt on a random subset of CFPB complaints that are not labeled with the 'fraud or scam' issue or sub-issue and found lower precision and recall' is vague about the sample size and selection; please add details so readers can gauge the reliability of the .25/.33 numbers.

Circularity Check

1 steps flagged · score 6.0 of 10

Reported precision/recall on L are in-sample fits from iterative prompt development; only precision has a partial external check, so the central capability claim is partly self-confirming.

  1. fitted input called prediction [Section 3 'Prompt Design and Iteration' and 'Final Prompt and Performance']
    "We developed a prompt with high precision and recall through an iterative process of evaluating prompts against training data and improving the prompt according to observed error patterns. ... On the set L, F achieves precision of .95 and recall of .84."

    The final ensemble F is the product of iterating on the labeled set L: prompts were revised whenever they made errors on L, so the reported precision/recall on L are values of the fitted objective, not independent estimates. The only held-out check in the paper, the manual review of 133 predicted scams, measures precision on predicted positives only and does not estimate recall, so the headline recall of .84 remains an in-sample fit. The central capability claim is therefore partly forced by the fitting process rather than demonstrated out of sample.

full rationale

The paper's central quantitative claim is that the ensemble F achieves precision .95 and recall .84 on the labeled set L. Section 3 states explicitly that the prompt was developed by iteratively evaluating prompts against this same training data and revising the prompt according to observed error patterns. Reporting precision and recall on L is therefore reporting the objective of the fitting process, not an unbiased estimate of performance on new complaints. This is the main circular step: the high in-sample numbers arise partly by construction. The paper does provide one independent check—manually evaluating a random 10% sample of the 1,333 narratives F predicted as scam, yielding precision .97—which supports the precision component on predicted positives. It does not, however, measure out-of-sample recall or the false-negative side, so the recall claim remains a fitted value. The paper also honestly acknowledges in Section 5 that F has much lower precision (.25) and recall (.33) on a random subset of CFPB complaints without the 'fraud or scam' label, which limits the scope but does not remove the in-sample circularity for the reported headline numbers. The 'customer is always right' labeling rule is an additional validity concern, but it is not itself a derivational circularity under the paper's stated codebook; it would require showing the model's predictions and labels agree solely because both key on the word 'scam'. Overall, the central recall claim is partly self-confirming due to training-set evaluation, warranting a 6 rather than a higher score because the external precision check and the explicit scope caveat provide some independent content.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claims rely on a small self-labeled training set, prompt choices fitted to that set, and a labeling rule that accepts complainant self-reports at face value. There are no external labels, no held-out test set for recall, and no baseline comparison, so the ledger of assumptions is heavy relative to the strength of the reported numbers.

free parameters (3)
  • Ensemble prompts and conjunction rule = Prompt A (Gemini), Prompts B and C (GPT-4), plus AND combination
    Chosen by iterative evaluation on the 300-narrative training set L; the headline precision/recall is reported on the same L.
  • Narrative length thresholds for grouping (875, 1602 chars) = 875 and 1602
    Used to divide the 300 narratives into short/medium/long groups for the redaction analysis; thresholds are chosen by hand.
  • Performance decline threshold (~3000 chars) = Approximately 3000
    The observation that LLM performance declines beyond ~3000 characters in Figure 1 is based on visual inspection and is not derived from a model.
assumptions (4)
  • domain assumption Scam definition: a scam is a deceptive act where the user is tricked into authorizing a financially self-harming action; non-scam fraud is an unauthorized financial action.
    Used as the codebook for manual labeling (Section 2).
  • ad hoc to paper 'Customer is always right' labeling rule: complainant statements are treated as correct unless contradicted.
    Section 3. This rule can make labels reflect narrative self-reports rather than objective deception, creating a potential shortcut for the model.
  • domain assumption The 300 CFPB complaints labeled 'fraud or scam' are representative of the broader scam/fraud complaint population.
    The authors generalize from L to the 2,569 unlabeled narratives; no sampling strata or diversity checks are reported.
  • domain assumption Manual labels by the three authors are reliable despite no quantitative inter-rater agreement statistic.
    The paper states disagreements were resolved but does not report Cohen's kappa or similar; the stability of the label set is asserted, not measured.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Distinguishing Scams and Fraud with Ensemble Learning." pith.science (2026). https://pith.science/paper/UNQURZS7

@misc{pith2026241208680,
  author       = {Pith},
  title        = {Pith review of: Distinguishing Scams and Fraud with Ensemble Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UNQURZS7}},
  note         = {Machine review of arXiv:2412.08680}
}
read the original abstract

Users increasingly query LLM-enabled web chatbots for help with scam defense. The Consumer Financial Protection Bureau's complaints database is a rich data source for evaluating LLM performance on user scam queries, but currently the corpus does not distinguish between scam and non-scam fraud. We developed an LLM ensemble approach to distinguishing scam and fraud CFPB complaints and describe initial findings regarding the strengths and weaknesses of LLMs in the scam defense context.

Figures

Figures reproduced from arXiv: 2412.08680 by the authors.

Figure 1
Figure 1. Precision and recall of the final model, [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Each figure shows the accuracy of the final model, [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. OpenAlex reports about 2 citations worldwide. Full citation record

  1. "It Warned Me Just at the Right Moment": Exploring LLM-based Real-time Detection of Phone Scams

    cs.HC 2025-02 conditional novelty 4.0 of 10

    LLM-based turn-by-turn analysis of phone conversations can flag scams in real time with high recall, and an optional 'UNCERTAIN' label trades recall for precision.

Reference graph

Works this paper leans on

27 extracted references · 23 canonical work pages · cited by 1 Pith paper

  1. [1]

    Student loans and financial distress: A qualitative analysis of the most common student loan complaints

    Matthew Adam Bruckner and Christopher J Ryan Jr. Student loans and financial distress: A qualitative analysis of the most common student loan complaints. Loy. Consumer L. Rev., 35:203, 2023

    work page 2023

  2. [2]

    Consumer response annual report, january 1 — december 31, 2023

    Consumer Financial Protection Bureau. Consumer response annual report, january 1 — december 31, 2023. https:// files.consumerfinance.gov/ f/ documents/ cfpb_cr-annual-report_2023-03.pdf , 2024

    work page 2023

  3. [3]

    Extracting training data from large language models

    Nicholas Carlini, Florian Tramer, Eric Wallace, Matthew Jagielski, Ariel Herbert- Voss, Katherine Lee, Adam Roberts, Tom Brown, Dawn Song, Ulfar Erlingsson, and Alina Oprea. Extracting training data from large language models. In 30th USENIX Security Symposium (USENIX Security 21) , pages 2633–2650, 2021

    work page 2021

  4. [4]

    Consumer complaints database

    The Consumer Financial Protection Bureau (CFPB). Consumer complaints database. https:// www.consumerfinance.gov/ data-research/ consumer-complaints/

    work page

  5. [5]

    The science of persuasion

    Robert B Cialdini. The science of persuasion. Scientific American, 284(2):76–81, 2001

    work page 2001

  6. [6]

    As nationwide fraud losses top $10 billion in 2023, ftc steps up efforts to protect the public

    Federal Trade Commission. As nationwide fraud losses top $10 billion in 2023, ftc steps up efforts to protect the public. FTC Press Releases, 2024

    work page 2023

  7. [7]

    Financial exploitation vulnerability and social isolation in older adults: Results from a longitudinal survey

    Gilbert Gimm and Scott Beach. Financial exploitation vulnerability and social isolation in older adults: Results from a longitudinal survey. Innovation in Aging, 4(Suppl 1):29, 2020

    work page 2020

  8. [8]

    AnnoLLM: Making large language models to be better crowdsourced annotators

    Xingwei He, Zhenghao Lin, Yeyun Gong, A-Long Jin, Hang Zhang, Chen Lin, Jian Jiao, Siu Ming Yiu, Nan Duan, and Weizhu Chen. AnnoLLM: Making large language models to be better crowdsourced annotators. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 6: Ind...

    work page 2024

Show all 27 references
  1. [9]

    John, Erin Picone-Decaro, Richard A Jenkins, and James W Carey

    Daniel J Hruschka, Deborah Schwartz, Daphne Cobb St. John, Erin Picone-Decaro, Richard A Jenkins, and James W Carey. Reliability in coding open-ended data: Lessons learned from hiv behavioral research. Field methods, 16(3):307–331, 2004

    2004

  2. [10]

    $1.8 billion going to victims of credit repair scam

    Truman Lewis. $1.8 billion going to victims of credit repair scam. ConsumerAf- fairs.com, 2024

    2024

  3. [11]

    Scam compliance and the psychology of persuasion

    David Modic and Stephen EG Lea. Scam compliance and the psychology of persuasion. A vailable at SSRN 2364464, 2013

    2013

  4. [12]

    GPT-4 Model Card

    OpenAI. GPT-4 Model Card. https://cdn.openai.com/papers/gpt-4-system- card.pdf, 2024. Accessed: 2024-05-12

    2024

  5. [13]

    Mortgage financing frauds in the us: An analysis of CFPB complaint database

    Ali Polat, Muhammad Mobeen Ajmal, and Abdul Rafay. Mortgage financing frauds in the us: An analysis of CFPB complaint database. In Concepts, Cases, and Regulations in Financial Fraud and Corruption , pages 1–27. IGI Global, 2023

    2023

  6. [14]

    Redmiles, Amelia R

    Elissa M. Redmiles, Amelia R. Malone, and Michelle L. Mazurek. I think they’re trying to tell me something: Advice sources and selection for digital security. In 2016 IEEE Symposium on Security and Privacy (SP) , pages 272–288, 2016

    2016

  7. [15]

    How to level up your fraud defense with chatgpt’s own ai sugges- tions

    Seth Ruden. How to level up your fraud defense with chatgpt’s own ai sugges- tions. BioCatch Blog Channel, January 27, 2023

    2023

  8. [16]

    A systematic survey of prompt engineering in large language models: Techniques and applications

    Pranab Sahoo, Ayush Kumar Singh, Sriparna Saha, Vinija Jain, Samrat Mondal, and Aman Chadha. A systematic survey of prompt engineering in large language models: Techniques and applications. arXiv preprint arXiv:2402.07927, 2024

    2024 arXiv

  9. [17]

    Using chatgpt for real-time fraud detection in finance

    Sarah. Using chatgpt for real-time fraud detection in finance. Dataleon. Ai Blog, December 11, 2023

    2023

  10. [18]

    Can llms be scammed? a baseline measurement study

    Udari Madhushani Sehwag, Kelly Patel, Francesca Mosca, Vineeth Ravi, and Jessica Staddon. Can llms be scammed? a baseline measurement study. arXiv preprint arXiv:2410.13893, 2024

    2024 arXiv

  11. [19]

    Online-dating romance scam in malaysia: An analysis of online conversations between scammers and victims

    Azianura Hani Shaari, Mohammad Rahim Kamaluddin, Wan Fariza Paizi Fauzi, and Masnizah Mohd. Online-dating romance scam in malaysia: An analysis of online conversations between scammers and victims. GEMA Online Journal of Language Studies, 19(1), 2019

    2019

  12. [20]

    Combating phone scams with llm-based detection: Where do we stand? arXiv preprint arXiv:2409.11643, 2024

    Zitong Shen, Kangzhong Wang, Youqian Zhang, Grace Ngai, and Eugene Y Fu. Combating phone scams with llm-based detection: Where do we stand? arXiv preprint arXiv:2409.11643, 2024

    2024 arXiv

  13. [21]

    Large language models for data annotation: A survey

    Zhen Tan, Alimohammad Beigi, Song Wang, Ruocheng Guo, Amrita Bhattachar- jee, Bohan Jiang, Mansooreh Karami, Jundong Li, Lu Cheng, and Huan Liu. Large language models for data annotation: A survey. arXiv preprint arXiv:2402.13446, 2024

    2024 arXiv

  14. [22]

    OFFICE OF CONSUMER RE- SPONSE

    The Consumer Financial Protection Bureau (CFPB). OFFICE OF CONSUMER RE- SPONSE. Narrative scrubbing standard version 6.6. https:// files.consumerfinance. gov/ f/ documents/ cfpb_narrative-scrubbing-standard_2023-05.pdf

  15. [23]

    How experts detect phishing scam emails

    Rick Wash. How experts detect phishing scam emails. Proceedings of the ACM on Human-Computer Interaction, 4(CSCW2):1–28, 2020

    2020

  16. [24]

    Chain-of-thought prompting elicits reasoning in large language models

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc V Le, and Denny Zhou. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824–24837, 2022

    2022

  17. [25]

    How to use norton’s free ai-powered scam detector

    Lance Whitney. How to use norton’s free ai-powered scam detector. ZDNet, September 14, 2023

    2023

  18. [26]

    Vulnerability to fraud among chinese older adults: Do personality traits and loneliness matter? Journal of Elder Abuse & Neglect , 32(1):46–59, 2020

    Tong Xing, Fei Sun, Kaipeng Wang, Jiawei Zhao, Mengxuan Wu, and Jie Wu. Vulnerability to fraud among chinese older adults: Do personality traits and loneliness matter? Journal of Elder Abuse & Neglect , 32(1):46–59, 2020

    2020

  19. [27]

    it’s a fair game

    Zhiping Zhang, Michelle Jia, Hao-Ping Lee, Bingsheng Yao, Sauvik Das, Ada Lerner, Dakuo Wang, and Tianshi Li. “it’s a fair game”, or is it? examining how users navigate disclosure risks and benefits when using llm-based conversational agents. In Proceedings of the CHI Conferen...

    2024

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.