Working notes behind email_tracking. Every query, every fold, every residue, every quote checked, every external source, and every judgement call. Corpus-level caveats — the venue scope, the selection funnel, the stability of each field — are on corpus and are not repeated here.

This is a working log, not prose. It is for someone checking a number.

No ~~DISCUSSION~~ block: this is the first page in the provenance: namespace to be written in this sitting, and the convention it follows is that comments belong on the content page. ¹⁾

The task brief asked for four slices to be inventoried before the page boundary was decided, and for the counts to be published here. They are:

Two slices the brief did not name came out of the inventory and are on the page:

One page with sections, not a parent with children. Reasons, in the order they mattered:

1. Slice (b) has zero papers. A privacy:email_unsubscription child would have been a page whose entire content is “the corpus is silent, and Karel is writing the paper”. That is one <WRAP todo> on a parent, not a page.
2. Slice (a) has 8 papers. Below the threshold anything else on this wiki has been given its own page (programming:crawler:openwpm is 60, security:phishing 139).
3. Slices © and (d) are the two that could carry pages at 15 and 23, but splitting them from (a) and (a′) would have separated papers that share the instrument. [1Kirchner, Robin; Koch, Simon; Kamangar, Noah; Klein, David; Johns, Martin (2024): "A Black-Box Privacy Analysis of Messaging Service Providers' Chat Message Processing", in: Proceedings on Privacy Enhancing Technologies. (DOI)] is the honey-message method applied to chat; it belongs next to [2Englehardt, Steven; Han, Jeffrey; Narayanan, Arvind (2018): "I never signed up for this! Privacy implications of email tracking", Proceedings on Privacy Enhancing Technologies 2018(1):109-126. (DOI)], not in a “messaging” child.
4. The precedent named in the brief, security, went parent-plus-children because it had five slices of 50–880 papers each. This topic has one slice of 23 and four of 8–15.

Recorded as a future split point: if the SMS/calls/messaging slice passes roughly 35 papers — it gained 6 in 2025–2026 alone — it should become privacy:message_spam and the parent should keep only (a), (a′) and (b). Nothing else on the page needs to move for that.

The line drawn between PHISH and this page: what a message discloses to a third party is here; whether a user falls for a message is there. Two papers sat exactly on it and went to PHISH — [3Chand, Anish; Nikiforakis, Nick; Vadrevu, Phani (2025): "Doubly Dangerous: Evading Phishing Reporting Systems by Leveraging Email Tracking Techniques", in: Proceedings of the USENIX Security Symposium. (Link)] was the borderline in the other direction and stayed here, because its object is what email-tracking techniques reveal, with the phishing-report pipeline as the setting.

There is no query for this page. No field in the extraction means “the measured object is a message”. The population is a published candidate pool plus a verdict per paper — but not a hand-written verdict for all 354 of them, and an earlier draft of this page and of the content page both said so. The real accounting, which the report now prints:

All 173 non-OFF papers are hand-mapped with a reason. The 125 defaulted papers were screened at title level and nothing further was recorded about them individually. Until 2026-09-02 the default reason string read “read and judged off-topic: the message vocabulary is incidental to the object measured” and was printed per paper in the published residue — a canned sentence asserting a per-paper reading that did not happen, and one that would have hidden a typo'd MAP key as a silent OFF. It now says what is true.

Three signals, unioned. A paper enters the pool if any fires.

• Signal 1 — the title. SLUG_RE over the slug with hyphens replaced by spaces: /\b(e?mails?|mailbox|inbox|newsletter|spam|spammer|spammers|spamming|sms|mms|smish|smishing|robocall|robocalls|unsubscribe|telemarketing|pixel|lead)\b/i
• Signal 2 — full-text density. Six counters over paper.cols.txt with whitespace collapsed, and a low threshold on each, because the pool is hand-read: email >= 40, spam >= 25, sms >= 25, pixel >= 6, unsub >= 5, lead >= 8. The regexes are in the script.
• Signal 3 — the messaging-app address space. /(phone )?number enumerat|contact discovery|contact upload|address book|honey (message|token)/gi at >= 3. Added after signals 1 and 2 were built, because they missed the messenger-enumeration papers entirely: no title says “email” or “SMS” and the body says “contact discovery”. That addition brought in [4Hagen, Christoph; Weinert, Christian; Sendner, Christoph; Dmitrienko, Alexandra; Schneider, Thomas (2021): "All the Numbers are US: Large-scale Abuse of Contact Discovery in Mobile Messengers", in: Proceedings of the Network and Distributed System Security Symposium. (Link)], [5Gegenhuber, Gabriel K.; Frenzel, Philipp E.; Günther, Maximilian; Ullrich, Johanna; Judmayer, Aljosha (2026): "Hey there! You are using WhatsApp: Enumerating Three Billion Accounts for Security and Privacy", in: Proceedings of the Network and Distributed System Security Symposium. (Link)], [6Kang, Junkyu; Lee, Soyoung; Kwon, Yonghwi; Son, Sooel (2026): "Connecting the Dots: An Investigative Study on Linking Private User Data Across Messaging Apps", in: Proceedings of the Network and Distributed System Security Symposium. (Link)] and [7Niksirat, Kavous Salehzadeh; Velykoivanenko, Lev; Mätzler, Samuel; Mulders, Stephan; Tamò-Larrieux, Aurelia; Boldi, Marc-Olivier; Humbert, Mathias; Huguenin, Kévin (2025): "Addressing the Address Books' (Interdependent) Privacy Issues", in: Proceedings of the USENIX Security Symposium. (Link)] — four of the twelve papers in the EID slice, including its newest.

Three papers were admitted by hand, each with the probe that found it recorded in HAND_ADDED rather than by loosening a threshold:

Four homographs dominate the vocabulary and no regex separates them from the topic. Each was measured:

tools[].name is folded into families tagged by the question the instrument answers, ordered, first match wins. Full definition in msg_fold.mjs; the table is on the content page. The two things worth recording here:

Exact-string undercounts the fold repairs, on this small a population:

The residue is 159 distinct strings and is printed in full in the report output in §8, not summarised. It is almost entirely one-off infrastructure. Two things in it are worth a reader's eye rather than a fold: Faker, Fake Name Generator, This person does not exist, 2Captcha and DeCaptcher appear because subscribing to twelve thousand newsletters means generating twelve thousand plausible identities and solving twelve thousand CAPTCHAs; and Email privacy tester, EmailHarvester, Honey Messages Framework and css-inline are the only four mail-specific research tools named by more than nobody. Neither observation is a number, which is why they are here and not on the page.

A pool built from titles and full-text density can miss a paper whose topic only shows up in a schema field. Four schema-side probes were run against the finished pool. Full output in §8.

classification.target has an email-message value and it is noisy in a way an enum disguises. It fires on 54 papers, 12 of which are outside this page's pool: The Matter of Heartbleed, The State of the SameSite, Traveling the Silk Road, an unsafe-image-generation paper, a conversational-search paper, an Android-app-reviews paper. None measures a message. If a future run is tempted to use this enum as the population, that is the reason not to.

All 73 outside-the-pool papers were read at title level; the 12 from the email-message enum were opened. Exactly one was a genuine miss:

• [10Starov, Oleksii; Gill, Phillipa; Nikiforakis, Nick (2016): "Are You Sure You Want to Contact Us? Quantifying the Leakage of PII via Website Contact Forms", in: Proceedings on Privacy Enhancing Technologies. (DOI)] — Starov, Gill and Nikiforakis, Are You Sure You Want to Contact Us? Quantifying the Leakage of PII via Website Contact Forms, PoPETs 2016. Added to EID and now the oldest paper in that slice, six years before [11Senol, Asuman; Acar, Gunes; Humbert, Mathias; Zuiderveen Borgesius, Frederik (2022): "Leaky Forms: A Study of Email and Password Exfiltration Before Form Submission", in: 31st USENIX Security Symposium (USENIX Security 22). (Link)]. Read the addition as the measured cost of a title-and-density pool: about one paper in seventy, in the direction of missing the oldest work in a slice.

Read and rejected from those lists, recorded because they are the ones most likely to be re-proposed:

Nine further papers surfaced by earlier exploratory probes were read and rejected before the pool rule was fixed, and are recorded in NEAR_MISSES in the fold script so a future run does not re-add them. They include What WeChat Knows (super-app first-party tracking — platforms), the PII-removal-service papers, and the three “opt-out” homograph papers.

verify_email_figures.mjs checks every literal per-paper figure quoted on the content page against that paper's own paper.cols.txt, not against evidence.quote — detection[].prevalence is a model summary, so a figure can be right in the paper and wrong in the extraction. Whitespace is collapsed on both sides.

78 figures across 24 papers, all present. (It was 60 across 22 when this section was first written and grew twice during review; the count here is now taken from the script's own summary line in §12, not retyped.) The pass earned its keep three times:

1. [12Agarwal, Sharad; Papasavva, Antonis; Suarez-Tangil, Guillermo; Vasek, Marie (2025): "Fishing for Smishing: Understanding SMS Phishing Infrastructure and Strategies by Mining Public User Reports", in: Proceedings of the ACM Internet Measurement Conference. (DOI)]: the mathematical-italic k. The paper typesets “27.7𝑘 smishing messages, 19.3𝑘 sender IDs, and 20𝑘 URLs” with U+1D458, not ASCII k. An obvious ASCII needle reported three present figures as missing. The needles now carry the real codepoint and a comment saying why.
2. [13Venkatadri, Giridhari; Andreou, Athanasios; Liu, Yabing; Mislove, Alan; Gummadi, Krishna P.; Loiseau, Patrick; Goga, Oana (2018): "Privacy Risks with Facebook's PII-Based Targeting: Auditing a Data Broker's Advertising Interface", in: Proceedings of the IEEE Symposium on Security and Privacy. (DOI)]: a paraphrased denominator. The extraction summarises the result as “18 of 20 visitors inferred”. The paper says “we also successfully inferred that 18 of the volunteers who did visit the webpage were part of the tracking pixel audience”, of 20 who visited. Correct, but a paraphrase, and the page quotes the paper.
3. [2Englehardt, Steven; Han, Jeffrey; Narayanan, Arvind (2018): "I never signed up for this! Privacy implications of email tracking", Proceedings on Privacy Enhancing Technologies 2018(1):109-126. (DOI)] and [14Pitsillidis, Andreas; Kanich, Chris; Voelker, Geoffrey M.; Levchenko, Kirill; Savage, Stefan (2012): "Taster's choice: a comparative analysis of spam feeds", in: Proceedings of the ACM Internet Measurement Conference. (DOI)]: two more paraphrases. “62% of leaks” is the paper's “The majority of these leaks (62%) are intentional”; “ten spam feeds” is its “ten different data feeds”.

Also checked and confirmed present, because the content page leans on them: the same paper's 27.7k of 64,284 image attachments and 220,585 posts — i.e. the 27.7k is post-extraction, and the abstract's larger 64.5k/28.6k/25.9k figures count raw attachments and reports. The page quotes the extracted figures and says they are extracted.

Every detection evidence.quote on the 70 on-page papers, located in the rendering the extractor read, with whitespace and smart quotes normalised. An elided quote passes if every 12+-character fragment appears in the paper's own order; otherwise a 5-word-window coverage of 60% counts as located.

Below threshold is not “unsupported”. All 62 are listed in the report output with their coverage figure and the first 110 characters, so each can be read by hand.

Eight were read by hand against paper.cols.txt — the longest contiguous run of each was located and the surrounding 300 characters printed. All eight are present, and every one is split by two-column interleaving that the .cols repair did not fully fix: the quote's own words are there, with a sentence from the neighbouring column spliced into the middle.

The pattern is uniform enough to be worth stating as a rule: a coverage figure in the 40–60% band on a two-column paper is a column splice, not a fabrication. A coverage figure below about 25% is worth opening, and the rule above was inferred from the middle of the distribution, so the tail was checked separately. The two lowest were read by hand as well, and both are present:

Four of the 62 are below 25% — [19Tu, Guan-Hua; Li, Chi-Yu; Peng, Chunyi; Li, Yuanjie; Lu, Songwu (2016): "New Security Threats Caused by IMS-based SMS Service in 4G LTE Networks", in: Proceedings of the ACM SIGSAC Conference on Computer and Communications Security. (DOI)] (23%), [20Utz, Christine; Michels, Matthias; Degeling, Martin; Marnau, Ninja; Stock, Ben (2023): "Comparing Large-Scale Privacy and Security Notifications", in: Proceedings on Privacy Enhancing Technologies. (DOI)] (20%), [21Reaves, Bradley; Scaife, Nolen; Tian, Dave; Blue, Logan; Traynor, Patrick; Butler, Kevin R. B. (2016): "Sending Out an SMS: Characterizing the Security of the SMS Ecosystem with Public Gateways", in: Proceedings of the IEEE Symposium on Security and Privacy. (DOI)] (20%) and [12Agarwal, Sharad; Papasavva, Antonis; Suarez-Tangil, Guillermo; Vasek, Marie (2025): "Fishing for Smishing: Understanding SMS Phishing Infrastructure and Strategies by Mining Public User Reports", in: Proceedings of the ACM Internet Measurement Conference. (DOI)] (18%). Checking that claim is how verify_email_figures.mjs gained two more papers: [12Agarwal, Sharad; Papasavva, Antonis; Suarez-Tangil, Guillermo; Vasek, Marie (2025): "Fishing for Smishing: Understanding SMS Phishing Infrastructure and Strategies by Mining Public User Reports", in: Proceedings of the ACM Internet Measurement Conference. (DOI)] was already covered, [21Reaves, Bradley; Scaife, Nolen; Tian, Dave; Blue, Logan; Traynor, Patrick; Butler, Kevin R. B. (2016): "Sending Out an SMS: Characterizing the Security of the SMS Ecosystem with Public Gateways", in: Proceedings of the IEEE Symposium on Security and Privacy. (DOI)] was not and the page quotes three of its figures, so it was added (386,327 messages, 522 containing email addresses, 14 months, over 400 numbers — all present); [22Jiang, Nan; Jin, Yu; Skudlark, Ann; Zhang, Zhi-Li (2013): "Greystar: Fast and Accurate Detection of SMS Spam Numbers in Large Cellular Networks Using Gray Phone Space", in: Proceedings of the USENIX Security Symposium. (Link)] was added in the same pass. The page quotes no figure from the remaining two, so nothing on it rests on those two quotes.

Every external fact on the content page, its primary source, and the check. Re-run with bash scripts/external_checks_email_tracking.sh; unedited output in §11.

• How much Gmail's proxy and Apple's MPP changed measured tracking prevalence. The mechanisms are documented by the vendors; the effect on a measurement is not measured anywhere the page could cite. It is stated as the page's largest methodological gap rather than estimated.
• Whether the recent crowdsourced SMS/smishing papers draw on overlapping sources. [23Murynets, Ilona; Jover, Roger Piqueras (2012): "Crime scene investigation: SMS spam data analysis", in: Proceedings of the ACM Internet Measurement Conference. (DOI)], [16Tang, Siyuan; Mi, Xianghang; Li, Ying; Wang, XiaoFeng; Chen, Kai (2022): "Clues in Tweets: Twitter-Guided Discovery and Analysis of SMS Spam", in: Proceedings of the ACM SIGSAC Conference on Computer and Communications Security. (DOI)], [24Agarwal, Sharad; Harvey, Emma; Vasek, Marie (2024): "Poster: A Comprehensive Categorization of SMS Scams", in: Proceedings of the ACM Internet Measurement Conference. (DOI)] and [12Agarwal, Sharad; Papasavva, Antonis; Suarez-Tangil, Guillermo; Vasek, Marie (2025): "Fishing for Smishing: Understanding SMS Phishing Infrastructure and Strategies by Mining Public User Reports", in: Proceedings of the ACM Internet Measurement Conference. (DOI)] all use public reports and at least two name overlapping forums. Quantifying the overlap would need their datasets. Stated as an open question.
• The size of the anti-spam literature outside these seven venues. CEAS, TREC spam track, APWG and the telephony venues are entirely absent. For this topic the seven-venue bound is weaker than on most pages of this wiki, and the page says so in those words rather than with a number.
• Field stability for this slice specifically. The ~20% free-text and 58% classification.method agreement figures were measured on the previous 4,322-paper run over 100 papers, not on this population. Quoted as an order of magnitude and labelled as such.
• Whether U is really zero or merely unsearchable. Six full-text probes were run, including the header name, the RFC numbers and the statute. The strongest evidence that zero is real rather than an artefact is that List-Unsubscribe — a fixed string that any paper measuring the mechanism would have to print — appears once in 5,859 papers, in a table of signable header fields.

1. privacy: rather than a new namespace, for a page that covers robocalls. The SMS and voice slice is not a web-platform topic and does not fit this namespace's own description. It is here because it shares the instrument — an address you own, messages that arrive unbidden — and because the alternative was a second namespace with one page in it. The page says this in its own words rather than pretending robocalls are web measurement.
2. The page keeps the id privacy:email_tracking and a broader title. The brief fixed the id. The title is Email Tracking and Message-Channel Measurement so that the SMS, voice and messaging content is not hidden behind a narrower name. A reader arriving from the id gets more than they expected, which is the safer direction.
3. EID was split out rather than folded into T. They share a subject (the address) and share no method (a web crawl versus a mailbox). Folding them would have made a 20-paper slice that looked healthy and hidden that the mailbox half is 8 papers and flat while the crawl half is 12 and growing.
4. The 2010–2012 spam-economics cluster is dated historical, not superseded. Superseded implies something replaced it. Nothing did: the method has no current successor and the ecosystem it measured is gone. [14Pitsillidis, Andreas; Kanich, Chris; Voelker, Geoffrey M.; Levchenko, Kirill; Savage, Stefan (2012): "Taster's choice: a comparative analysis of spam feeds", in: Proceedings of the ACM Internet Measurement Conference. (DOI)] is exempted explicitly because its feed-bias lesson is method, not ecosystem.
5. llm is reported as 2 papers, not as a percentage of a trend. Both are 2025, which is one of the provisional venue-years. The page gives the count and the years and refuses the trend line. A reasonable person could have written “LLM classification is emerging at 5.4% of recent papers”; that would be two data points in the corpus's thinnest years dressed as a rate.
6. The unsubscription section states a zero and stops. Karel is writing an IEEE S&P paper on unsubscription. Nothing unpublished is cited, summarised or anticipated; the <WRAP todo> names the four measurable claims that have no papers and says work is under way, which is a fact about the record, not a preview of a result.
7. The ethics section leads with the legal-assessment rate. 14 of 70 (20.0%) against 6.9% corpus-wide. That is a real signal about the slice and it is the sort of comparison that goes wrong without a base rate, so the base rate is printed beside it. ²⁾
8. Instrument families are tagged by question, not vendor. telephony-rig contains Asterisk, Twilio, an ip.access nanoBTS and an AT-command interface, which have nothing in common as products and everything in common as answers to “how do I receive a call or an SMS I did not ask for”.

node scripts/report_email_tracking.mjs — every figure on the page with its denominator, the full 181-paper off-topic residue, the 159-string instrument-fold residue, the recall probes, and all 62 below-threshold quotes.

report_email_tracking-output.txt

==============================================================================
A. CORPUS AND POOL
==============================================================================
extraction records          : 5859
  ... with no paper.cols.txt: 4 (counted as negatives by every full-text probe here)
candidate pool (screened)   : 354
  admitted by hand          : 3
hand-mapped verdicts        : 183
  ... of which carry a written reason      : 125 (182 MAP entries; the rest are
                                             PHISH / SOC / WEBPIXEL rows where the verdict IS the reason)
NOT in the hand map, verdict OFF by default: 171
  ... given a reason by an OFF_FAMILIES rule: 46
  ... carrying only "not individually annotated": 125
  These were screened at TITLE level. Nothing further was recorded per paper.
ON the page                 : 70
rejected outside the pool   : 9 (documented, see section H)
 
--- Verdicts over the pool
Verdict   Papers  Share of pool  On page  What it means                                                                                                                                                            
--------  ------  -------------  -------  -------------------------------------------------------------------------------------------------------------------------------------------------------------------------
T         8       2.3%           yes      message tracking — the measured object is tracking content carried inside a message, or the mailbox is the experimental unit                                             
EID       12      3.4%           yes      the email address (or phone number) as the tracking identifier — exfiltration from forms, hashed-email matching, alias collision                                         
U         0       0.0%           yes      marketing mail, consent for it, and opting out of it                                                                                                                     
SE        15      4.2%           yes      email spam / scam mail as the object of study, including filters and blocklists                                                                                          
SM        23      6.5%           yes      SMS, MMS, voice-call and messaging-app spam, scams and abuse                                                                                                             
INST      12      3.4%           yes      the mailbox or phone number is the INSTRUMENT and the object is elsewhere — honey accounts, canary addresses, notification campaigns, DSAR mail                          
INFRA     31      8.8%           no       email transport, authentication and encryption deployment (SPF/DKIM/DMARC/DANE/MTA-STS/STARTTLS/S-MIME, spoofing, delivery) — a page-sized slice of its own, not tracking
PHISH     28      7.9%           no       phishing and its interventions, with email as the vector — security:phishing                                                                                             
SOC       37      10.5%          no       social-platform, review, forum, SEO or ad-click spam — the platform, not a message channel                                                                               
WEBPIXEL  7       2.0%           no       the tracking pixel on a WEB page rather than in a message — privacy:requests                                                                                             
OFF       181     51.1%          no       homograph or unrelated                                                                                                                                                   
 
--- The four slices the page brief asked about, as measured
Brief                                                           Verdict(s)  Papers  Carries a page?   
--------------------------------------------------------------  ----------  ------  ------------------
(a) tracking inside a message / mail as a measurement artefact  T           8       section           
(a′) the address itself as the identifier                       EID         12      section           
(b) unsubscription / opt-out / List-Unsubscribe                 U           0       NO — see section E
(c) spam as the object of study (email)                         SE          15      section           
(d) SMS, phone, messaging-app spam / lead-gen                   SM          23      section           
(—) the mailbox as the instrument                               INST        12      section           
 
==============================================================================
B. SHAPE OF THE ON-PAGE POPULATION (denominator: the 70 on-page papers)
==============================================================================
 
--- By venue
Venue    Papers  Share of 70
-------  ------  -----------
USENIX   19      27.1%      
NDSS     13      18.6%      
PETS     11      15.7%      
IEEE-SP  9       12.9%      
IMC      8       11.4%      
CCS      7       10.0%      
WWW      3       4.3%       
 
--- By year and slice. 2025 and especially 2026 are PROVISIONAL venue-years.
Year  T  EID  SE  SM  INST  Total
----  -  ---  --  --  ----  -----
2010          1             1    
2011          4   1         5    
2012          4   1         5    
2013              1         1    
2014  1       1             2    
2015              1         1    
2016     1        2   2     5    
2017              1   1     2    
2018  1  1                  2    
2019  1  2                  3    
2020     1        3   2     6    
2021     1        1   1     3    
2022  1  1    1   2   3     8    
2023  1           3   1     5    
2024  1       2   1         4    
2025  2  2    2   5   1     12   
2026     3        1   1     5    
 
--- The currency question, by slice: first year, last year, and how many since 2022
Slice  Papers  First  Last  ≤2014  2015–2021  2022–2024  2025–2026 (prov.)
-----  ------  -----  ----  -----  ---------  ---------  -----------------
T      8       2014   2025  1      2          3          2                
EID    12      2016   2026  0      6          1          5                
SE     15      2010   2025  10     0          3          2                
SM     23      2011   2026  3      8          6          6                
INST   12      2016   2026  0      6          4          2                
 
==============================================================================
C. WHAT THESE PAPERS DO (denominator named per row)
==============================================================================
 
--- classification.method over the on-page papers that classified anything
denominator: 61 of 70 on-page papers have a classification tuple
Method (enum)        Papers  Share of 61
-------------------  ------  -----------
heuristic-rules      36      59.0%      
manual-labelling     23      37.7%      
third-party-service  15      24.6%      
supervised-ml        13      21.3%      
curated-database     12      19.7%      
regex-or-signature   12      19.7%      
other                7       11.5%      
unsupervised-ml      5       8.2%       
blocklist            4       6.6%       
dynamic-analysis     3       4.9%       
graph-analysis       2       3.3%       
static-analysis      2       3.3%       
llm                  2       3.3%       
 
--- classification.method, on-page papers, 2010–2019 vs 2020–2026
Method               2010–2019 (n=24)  2020–2026 (n=37)
-------------------  ----------------  ----------------
heuristic-rules      14 (58.3%)        22 (59.5%)      
manual-labelling     6 (25.0%)         17 (45.9%)      
supervised-ml        3 (12.5%)         10 (27.0%)      
third-party-service  6 (25.0%)         9 (24.3%)       
regex-or-signature   4 (16.7%)         8 (21.6%)       
curated-database     7 (29.2%)         5 (13.5%)       
other                3 (12.5%)         4 (10.8%)       
unsupervised-ml      2 (8.3%)          3 (8.1%)        
blocklist            2 (8.3%)          2 (5.4%)        
dynamic-analysis     1 (4.2%)          2 (5.4%)        
graph-analysis       0 (0.0%)          2 (5.4%)        
static-analysis      0 (0.0%)          2 (5.4%)        
llm                  0 (0.0%)          2 (5.4%)        
 
--- Where the on-page papers go quiet. Sentinels are NOT counted as answers.
Field                             Population (on-page papers with the family)  Stated  Share
--------------------------------  -------------------------------------------  ------  -----
classification.taxonomy           61                                           43      70.5%
classification.groundTruthSource  61                                           44      72.1%
classification.validation         61                                           58      95.1%
population.n                      70                                           69      98.6%
population.listVersion            70                                           33      47.1%
humanAnnotation.annotatorCount    51                                           18      35.3%
humanAnnotation.agreementMetric   51                                           4       7.8% 
ethics.reviewOutcome              66                                           36      54.5%
ethics.harmMitigation             66                                           63      95.5%
artifacts.availability            67                                           36      53.7%
NOTE: ethics and artifacts are OBJECTS with several keys sharing one evidence quote.
      A share here says how often the key carries a non-sentinel value, nothing more.
 
--- The same silence, against the whole-corpus base rate. A slice can look bad and be average.
Field                             On-page papers  Whole corpus     
--------------------------------  --------------  -----------------
classification.validation         58/61 = 95.1%   3977/4439 = 89.6%
classification.groundTruthSource  44/61 = 72.1%   3234/4439 = 72.9%
classification.taxonomy           43/61 = 70.5%   2854/4439 = 64.3%
population.n                      69/70 = 98.6%   5496/5712 = 96.2%
population.listVersion            33/70 = 47.1%   2455/5712 = 43.0%
humanAnnotation.annotatorCount    18/51 = 35.3%   1288/3318 = 38.8%
humanAnnotation.agreementMetric   4/51 = 7.8%     512/3318 = 15.4% 
artifacts.availability            36/67 = 53.7%   2890/4854 = 59.5%
artifacts.availability is computed over the EMPIRICAL corpus, as OVERVIEW.md does,
because a non-empirical paper releasing nothing is not a reporting gap.
 
--- Study shape: platforms, study types, crawl configuration, participants
Platform              Papers  Share of 70
--------------------  ------  -----------
other-online-service  54      77.1%      
web                   33      47.1%      
mobile                18      25.7%      
offline               4       5.7%       
iot                   1       1.4%       
Study type                  Papers  Share of 70
--------------------------  ------  -----------
manual-audit                42      60.0%      
system-or-defence-proposal  35      50.0%      
existing-dataset-analysis   33      47.1%      
automated-web-crawl         28      40.0%      
network-scan-or-probe       15      21.4%      
code-or-binary-analysis     12      17.1%      
mobile-app-analysis         11      15.7%      
interview-or-survey         11      15.7%      
user-study                  10      14.3%      
simulation-or-theory-only   1       1.4%       
ran an automated web crawl (crawlConfig OR studyTypes): 29/70 = 41.4%
have a recorded crawlConfig object              : 27
crawlConfig field  Stated (of 27)  Share
-----------------  --------------  -----
consentAction      14              51.9%
statefulness       7               25.9%
headless           5               18.5%
browsers           16              59.3%
interactionDepth   22              81.5%
authentication     22              81.5%
recruited human participants: 12/70 = 17.1%
temporal.mode (papers, multi-valued)  Papers
------------------------------------  ------
active-probing                        27    
live-crawl                            26    
passive-collection                    25    
existing-dataset                      19    
 
--- Ethics and legal assessment, on-page papers against the corpus base rate
assessed compliance with a law: 14/70 = 20.0% on-page
                                402/5859 = 6.9% corpus-wide
Law assessed                                     On-page papers
-----------------------------------------------  --------------
GDPR                                             9             
ePrivacy Directive                               2             
Chinese law                                      1             
German competition law                           1             
Act against Unfair Competition (UWG)             1             
German Telemedia Act (TMG)                       1             
US federal law                                   1             
law of the state where the honeypot is operated  1             
HIPPA                                            1             
Gramm-Leach-Bliley Act                           1             
CCPA                                             1             
Digital Markets Act                              1             
TRACED Act                                       1             
 
ethics.reviewOutcome stated: 36/66 = 54.5% on-page
                             1728/4472 = 38.6% corpus, empirical papers with an ethics object
NOTE: OVERVIEW.md reports 33.8% for this field over ALL 5,118 empirical papers,
      counting a null ethics object as not-stated. Both denominators are
      defensible; the page must not mix them, and says which it uses.
      empirical papers with NO ethics object: 646
      over all empirical papers: 1728/5118 = 33.8%
 
--- Instruments: tools[] used or produced by the on-page papers, folded by FAMILY
denominator: 70 on-page papers
Instrument family        Papers  Share of 70  Distinct strings  What it answers                                                               
-----------------------  ------  -----------  ----------------  ------------------------------------------------------------------------------
browser-automation       18      25.7%        18                driving a browser or app                                                      
network-intelligence     16      22.9%        19                geolocating or attributing an address (see design:ip_classification)          
ml-model                 16      22.9%        32                a classifier or clustering algorithm                                          
telephony-rig            13      18.6%        21                placing, receiving or fingerprinting calls and SMS                            
browser                  13      18.6%        11                the browser itself                                                            
maliciousness-oracle     13      18.6%        13                a maliciousness verdict from a vendor (see security:virustotal)               
study-apparatus          9       12.9%        10                recruiting participants or analysing their answers                            
mail-plumbing            7       10.0%        16                sending, receiving or parsing mail                                            
language-tooling         7       10.0%        4                 translating or parsing non-English text (see programming:multilingual_support)
spam-blocklist           6       8.6%         7                 a spam / abuse blocklist as ground truth                                      
mail-filter              6       8.6%         5                 a spam filter run by the authors as a labeller                                
traffic-capture          6       8.6%         8                 recording traffic (see programming:traffic_files)                             
mail-endpoint            5       7.1%         8                 a mailbox, provider or mail-specific detector used as the measurement point   
number-parsing           5       7.1%         5                 parsing and validating phone numbers                                          
llm-or-weak-supervision  5       7.1%         11                an LLM or a weak-supervision labeller                                         
crawler-framework        4       5.7%         4                 a web-measurement framework (see programming:crawler)                         
tracker-blocklist        4       5.7%         7                 a tracking blocklist (see programming:filter_lists)                           
audio-pipeline           4       5.7%         7                 turning call audio into text or fingerprints                                  
messenger-client         4       5.7%         6                 driving a messaging app programmatically                                      
 
--- The exact-string undercount this fold repairs
Name               Family                Distinct spellings  Tuples  The spellings                                                                         
-----------------  --------------------  ------------------  ------  --------------------------------------------------------------------------------------
Spamhaus           spam-blocklist        4                   4       Spamhaus / Spamhaus blacklist / Spamhaus.org / Spamhaus passive DNS API               
MaxMind / GeoIP    network-intelligence  4                   6       GeoIP / MaxMind GeoIP / MaxMind / MaxMind GeoIP and GeoLite Databases and Web Services
HLR lookup         telephony-rig         2                   2       HLR Lookup / HLR lookup                                                               
Chrome / Chromium  browser               5                   9       Google Chrome / Chrome / Chromium / custom Chrome extensions / chrome.webNavigation   
Selenium           browser-automation    2                   7       Selenium / Selenium Python framework                                                  
libphonenumber     number-parsing        4                   5       libphonenumber / Google libphonenumber / libephonenumber / phonenumbers               
 
--- Instrument-fold residue: 159 distinct strings the families do not cover
Printed in full, because a residue that lives only in a local file is a residue nobody reads.
   3  Python
   2  Google
   2  Twitter Academic API
   2  VirtualBox
   1  2Captcha
   1  Alexa
   1  Alibaba Cloud
   1  Amazon EC2
   1  Amazon Web Services
   1  Amazon Web Services (AWS)
   1  Android emulator
   1  Android Runner
   1  AndroZoo
   1  Anubis
   1  App instrumentation framework by Koch et al.
   1  app-downloader
   1  automated conversation engine
   1  automated data collection system
   1  B@bel
   1  Baidu
   1  Beautiful Soup
   1  BeautifulSoup
   1  BigQuery
   1  Bitly
   1  Bluetooth
   1  BOTMAGNIFIER
   1  browser developer tools
   1  BrowserMob Proxy
   1  Calendly
   1  canary tokens
   1  CheckGA
   1  content clustering tool
   1  custom browser extension
   1  custom browser-driving scripts
   1  custom differential grammar-based fuzzer
   1  custom malware sandbox system
   1  custom probing program
   1  custom script
   1  custom scripts
   1  custom template-attack framework
   1  custom vulnerability-specific probes
   1  custom web-based report interface
   1  CYNEX/Conexus-S
   1  DDS
   1  DeCaptcher
   1  DonateARC
   1  Emulab
   1  Euphony
   1  Facebook advertising interface
   1  Facebook advertising platform
   1  Facebook marketing API
   1  Facebook tracking pixel
   1  Facebook's ad transparency tool “Why Am I Seeing This?”
   1  Fake Name Generator
   1  Faker
   1  FCA Handbook
   1  FingerprintJS2
   1  Flask
   1  FlowDroid
   1  Formlock
   1  free domain provider
   1  full-fidelity Web crawler
   1  GhostDriver
   1  Ghostery
   1  Google AdWords
   1  Google Analytics
   1  Google Apps Script
   1  Google Cloud
   1  Google Contacts
   1  Google Form
   1  Google geolocation
   1  Google reCAPTCHA
   1  Google Search
   1  Google Search API
   1  Google Search APIs
   1  Google system fingerprinting system
   1  Google Vision Text Documentation Detection API
   1  Grammarly
   1  HackFacebook
   1  Hadoop Distributed File System
   1  hashcat
   1  IAB Europe's CMP Compliance Programme
   1  iban.com API
   1  image manipulation software
   1  ip xfrm state
   1  iptables
   1  J2ME echo server
   1  JavaScript Deobfuscator
   1  Karma
   1  Kestrel
   1  keyword-based approach
   1  keyword-search and pattern matching script (custom)
   1  LEAK INSPECTOR
   1  Lin
   1  Logcat
   1  London Borough of Camden council website
   1  Luhn algorithm
   1  Magento
   1  Marvel App
   1  MDN Web Docs
   1  mobile application
   1  Monarch
   1  Monitoring module
   1  named-entity recognition (NER)
   1  nsILoginManager
   1  OneTrust
   1  OpenAI Vision API
   1  OpenCV
   1  OpenMP
   1  OpenSSL
   1  page classifiers
   1  PlanetLab
   1  plecost
   1  Polisis
   1  postcodes.io API
   1  Postgres
   1  ProVerif
   1  Pytesseract
   1  Python scripts
   1  Python string similarity matching
   1  RAIDAR
   1  RainbowCrack
   1  RAKE (Rapid Automatic Keyword Extraction)
   1  Reddit API
   1  Redis
   1  regular expressions
   1  relational database
   1  Reporting module
   1  Scrapy
   1  Screengrab!
   1  semi-automated PDF table parsing tools
   1  Sending module
   1  ShowAPI
   1  SimilarWeb
   1  Spark
   1  SQL
   1  SSH
   1  t-test
   1  Tencent Map Geocoding API
   1  Tesseract
   1  This person does not exist
   1  tld
   1  tldextract
   1  Traceability Analyzer
   1  Transparency-enhancing advertisements (Treads)
   1  Trend Micro Site Safety Center
   1  Tripwire crawler
   1  Twitter Streaming API
   1  Ubercart
   1  UK Companies House register
   1  UK Office for National Statistics
   1  URL crawler
   1  Web crawler
   1  Web scraper
   1  X-Cart
   1  XRay
   1  Zen-cart
   1  zlib
   1  Zoom
 
==============================================================================
D. MEASURED RESULTS, quoted with the paper’s own denominator
==============================================================================
Source: detection[].prevalence. These are the per-paper results, not corpus-wide.
Each has been checked against the paper text by verify_email_figures.mjs.
 
--- T — 8 papers
  USENIX/2014/xray-enhancing-the-web-s-transparency-with-differential-correlation  [2 stated prevalence]
      personal-data targeting: 80-90% precision and recall across Gmail, Amazon, and YouTube
      sensitive-topic ad targeting: All disease-related emails except HIV were strongly correlated with ads
  PETS/2018/i-never-signed-up-for-this-privacy-implications-of-email-tracking  [7 stated prevalence]
      embedded third-party email content: 85% of emails
      email-address leakage when viewing emails: about 29% of emails
      intentional email leakage: 62% of leaks
      email-address leakage through clicked links: about 11% of links
  IEEE-SP/2019/characterizing-pixel-tracking-through-the-lens-of-disposable-email-services  [7 stated prevalence]
      PII leakage in emails: 1,399 credit card numbers, 926 SSNs, 701 EINs, and 40K VINs
      Email usage categories: 94.75% spam, 3.63% registration, 0.86% password reset, 0.75% authentication
      Email tracking pixels: 573,244 emails (24.6%) contained tracking links
      Third-party email tracking: 149,303 emails with third-party tracking
  PETS/2022/checking-websites-gdpr-consent-compliance-for-marketing-emails  [8 stated prevalence]
      GDPR consent violations: 17.3% of websites sending marketing emails had at least one potential violation.
      Potential violations overall: 21.9% of websites contained potential violations.
      Email content violations: 17.7% of services sent emails potentially violating content requirements.
      Double opt-in: 59% first sent a double opt-in email.
  CCS/2023/poster-towards-a-dataset-for-the-discrimination-between-warranted-and-unwarrante  [1 stated prevalence]
      Provider-labeled spam in collected emails: Primary: 1.2K (2.0%); Ad-Hoc: 0.6K (2.90%); Forwarded: 1.4K (2.60%), through 20-Aug-2023
  PETS/2024/a-black-box-privacy-analysis-of-messaging-service-providers-chat-message-process  [6 stated prevalence]
      server-side message analysis: 34% of messaging services conducted server-side URL access
      client-side token leakage: CNAME redirect chains were recognized for 60% of analyzed messengers
      URL crawling: LinkedIn followed links to one level of child pages
      delayed message processing: One service accessed a page 30 days after the chat
  NDSS/2025/cascading-spy-sheets-exploiting-the-complexity-of-modern-css-for-email-and-browser-fingerprinting  [8 stated prevalence]
      browser–OS fingerprinting: 1152 of 1176 (97.95 %)
      font fingerprinting: 831 of 861 (96.5 %)
      CSS arithmetic fingerprinting: 1116 OS-browser combination pairs (94.9 %)
      translation identification: 127 of 131 languages
  USENIX/2025/doubly-dangerous-evading-phishing-reporting-systems-by-leveraging-email-tracking  [5 stated prevalence]
      email-open tracking: 7 of 8 services
      email prefetching: Gmail, Outlook, and Proton prefetch tracking vectors
      phishing-report tracking: Gmail 72.3%, Outlook 99.4%, Proton 2.9% for identified systems
      security-crawler visits: 317 Outlook visits; 16 unique IP addresses; 214/317 had one fingerprint
 
--- EID — 12 papers
  PETS/2016/are-you-sure-you-want-to-contact-us-quantifying-the-leakage-of-pii-via-website-c  [5 stated prevalence]
      Contact-form discovery and submission: 29.6% had a contact page; 17% had a located contact form; 82.5% of discovered forms were successfully submitted
      Potential PII leakage: 29% of sites with contact forms
      Accidental PII leakage: 423 forms, or 2.5%, leaked PII to 3,573 unique third-party domains
      Intentional PII leakage: 6.1% (1,035) of all submitted contact forms
  IEEE-SP/2018/privacy-risks-with-facebooks-pii-based-targeting-auditing-a-data-brokers-adverti  [6 stated prevalence]
      Facebook audience-statistic rounding: over 58K different combinations of custom audiences and targeting attributes
      PII deduplication: Facebook de-duplicated records within and across audiences
      website-visitor de-anonymization: 18 of 20 visitors inferred; all 20 non-visitors correctly excluded
      email-to-phone-number inference: 11 of 14 Boston users and all 8 French users
  PETS/2019/investigating-sources-of-pii-used-in-facebook-s-targeted-advertising  [4 stated prevalence]
      Potential-reach obfuscation: Noise uniformly distributed between 0 and 20, followed by rounding in steps of ten.
      PII source use in targeted advertising: Five of seven investigated sources enabled PII-based advertising.
      Security-purpose PII reuse: 2FA phone numbers became targetable after 22 days; login-alert PII after 17 days.
      Contact-sync PII reuse: The phone number became targetable in 36 days.
  WWW/2019/auditing-offline-data-brokers-via-facebooks-advertising-platform  [8 stated prevalence]
      online-offline identity linkage: over 90% in the U.S.; 81.3% Australia; 74.4% U.K.
      data-broker coverage: 48.0%–90.5% across the seven countries
      age variation in broker coverage: Coverage generally increases up to age 30.
      county-level coverage variation: Acxiom coverage ranged roughly from 65% to 95%.
  PETS/2020/no-boundaries-data-exfiltration-by-third-parties-embedded-on-web-pages  [4 stated prevalence]
      login-manager credential exfiltration: two scripts found on 1,110 Alexa top-million sites
      social data exfiltration: seven scripts on 434 Alexa top-million sites; six collected user IDs
      whole-DOM exfiltration: 28 scripts identified; eight only by chunk injection
      sensitive-data leakage in session replay: passwords, medical, student, credit-card, billing, and prescription data observed
  NDSS/2021/all-the-numbers-are-us-large-scale-abuse-of-contact-discovery-in-mobile-messengers  [7 stated prevalence]
      mobile-phone hash reversal: over 99.99% hit rate with 52 ms amortized lookup time
      contact-discovery enumeration: 10% of US numbers for WhatsApp and 100% for Signal
      WhatsApp account registration: 5.0 million users found among 46.2 million checked numbers
      Signal account registration: 2.5 million US users from 505.7 million checked numbers
  USENIX/2022/leaky-forms-a-study-of-email-and-password-exfiltration-before-form-submission  [4 stated prevalence]
      pre-submission email exfiltration: 1,844 EU websites and 2,950 US websites
      pre-submission password exfiltration: 52 websites in follow-up crawls without password input-type swapping
      previously unlisted tracker domains: 41 tracker domains; 13.4% EU and 4.2% US increase in leaky websites
      received emails after abandoned forms: 290 emails from 88 desktop-crawl sites in six weeks
  PETS/2025/tracker-installations-are-not-created-equal-understanding-tracker-configuration  [6 stated prevalence]
      Meta Pixel and Google Tag installation: Google Tag on 72.6% and Meta Pixel on 28.2% of 40,150 websites
      Form data collection: Meta Pixel 62.3% versus Google Tag 11.6%
      Meta FDC configuration: 51.3% used the default configuration collecting all fields
      Collected PII fields: Email 99.5%, name 93.7%, phone number 93.5%
  NDSS/2026/connecting-the-dots-an-investigative-study-on-linking-private-user-data-across-messaging-apps  [8 stated prevalence]
      contact-discovery profile enumeration: 105,366 of 155,147 KakaoTalk numbers; 67.9%
      OAuth access-token exposure: 63 websites exposed tokens; 16 exposed them in URLs
      third-party token access: 56 of 63 token-exposing websites deployed third-party scripts
      Tinder location inference: 324 meters with 40 queries
  NDSS/2026/hey-there-you-are-using-whatsapp-enumerating-three-billion-accounts-for-security-and-privacy  [8 stated prevalence]
      WhatsApp account enumeration: 3,546,479,731 accounts across 245 countries
      Profile-picture availability: More than 57 % of global users
      Faces in profile pictures: roughly two-thirds (66 %) of 500,000 images
      Key reuse: 2.3 M distinct keys reused in bundles of over 2.9 M devices
  USENIX/2025/addressing-the-address-books-interdependent-privacy-issues  [4 stated prevalence]
      DAB contact-field completeness: Birthday mean 3.6%; address mean 7.7%
      Willingness to share DAB data: 19.9% chose access without a financial incentive
      Third-party DAB access: 90.5% reported at least one app
      DAB provider access-request responses: No provider supplied the requested contact-card data
  NDSS/2026/one-email-many-faces-a-deep-dive-into-identity-confusion-in-email-aliases  [6 stated prevalence]
      email-provider alias mechanisms: 28 providers tested; 12 supported additional aliasing beyond case variation and backslash handling
      platform alias acceptance: 9 platforms were completely unable to counter any tested alias mechanisms
      public alias usage: 310,136 of 2,141,447 addresses (14.48%)
      alias multiplicity abuse: one base address had 139 alias accounts and published 3,904 packages
 
--- SE — 15 papers
  USENIX/2011/botmagnifier-locating-spambots-on-the-internet  [6 stated prevalence]
      spambot IP magnification: 2,031,110 bot IP addresses during 114 evaluation days
      Cutwail bot detection: 33,550 of 144,317 candidates (approximately 23%) were C&C-listed bots
      false positives: 35,680, approximately 1.6% of the total
      botnet population tracking: Lethic 887,852; Rustock 676,905; Cutwail 319,355; MegaD 68,117; Waledac 36,058
  USENIX/2011/show-me-the-money-characterizing-spam-advertised-revenue  [4 stated prevalence]
      Spam-advertised order volume: over 82,000 pharmaceutical orders and over 37,000 software orders monthly
      Customer basket contents: 289 distinct products; 38% of cart additions were outside the ED and sexually-related category
      Cart-addition conversion: 0.5% overall on an IP basis
      Geographic customer distribution: 80% originated from the U.S. and Canada; Europe contributed 6%
  IMC/2012/tasters-choice-a-comparative-analysis-of-spam-feeds  [6 stated prevalence]
      spam-advertised registered domains: over a billion messages distributed over three months
      feed purity: DNS and HTTP success rates varied substantially across feeds
      spam-domain coverage: 60% of live domains and 19% of tagged domains were exclusive to one feed
      spam campaign timing: Hu saw over 75% of domains within one day and 95% within three days; dbl saw over 95% within one day
  USENIX/2012/b-bel-leveraging-email-delivery-for-spam-mitigation  [5 stated prevalence]
      SMTP dialect discrimination: 60 dialects from 13 legitimate MUAs/MTAs and 91 malware samples; 48 unique malware dialects
      Bot-based spam detection: 99.32% true-positive rate, 99.3% precision, and 21% false-negative rate
      Unrecognized SMTP conversations: 98% of 24,757 unrecognized conversations were flagged as spam by existing methods
      Spam feedback poisoning: Three campaigns reduced subsequent spam; 550,776 emails, or 19%, were from those campaigns
  USENIX/2012/impact-of-spam-exposure-on-user-engagement  [2 stated prevalence]
      Spam exposure effect on engagement: Spam exposure leads to significantly lower user engagement.
      Matched user pairs: 486,102 matched pairs
  USENIX/2012/pharmaleaks-understanding-the-business-of-online-pharmaceutical-affiliate-progra  [7 stated prevalence]
      customer orders and revenue: over 2M sales records and over $170M in settled revenue
      repeat customer purchasing: repeat orders constituted 27% of GlavMed and 38% of SpamIt average revenue
      drug demand: ED products generated 75% of GlavMed revenue and 82% of SpamIt revenue
      affiliate revenue concentration: the top 10% of affiliates accounted for 75-90% of program revenue
  NDSS/2014/scambaiter-understanding-targeted-nigerian-scams-on-craigslist  [8 stated prevalence]
      Craigslist scam attempts: 13,215 scam-related first responses; 9.6 scam trials per ad
      Scammer IP geolocation: 50.3% Nigeria and 37.6% United States
      Scammer group concentration: Top 10 groups accounted for 48% of received scam threads
      Shipping-address geolocation: 70% Nigeria, 23% United States, and 7% other countries
  WWW/2022/left-or-right-a-peek-into-the-political-biases-in-email-spam-filtering-algorithm  [4 stated prevalence]
      Political bias in email spam filtering: Gmail marked 67.6% of right emails and 8.2% of left emails as spam; Outlook marked 95.8% of left and 75.4% of right emails as spam
      Affiliation effect after covariate matching: Matched-group differences remained above 40% for reported Gmail examples
      User-interaction effects on political bias: After five spam-to-inbox interactions, Gmail marked 5.34% of right and 0% of left emails as spam
      Demographic effects on spam filtering: No observed impact of age, ethnicity, or gender
  CCS/2024/inbox-invasion-exploiting-mime-ambiguities-to-evade-email-attachment-detectors  [4 stated prevalence]
      MIME parsing ambiguity bypass: 180 of 237 samples achieved detection bypass in at least one product-client combination, yielding 75.95%.
      Detector-client vulnerability combinations: 102 of 128 combinations exhibited parsing ambiguities leading to successful bypasses.
      New evasion methods: 24 unique bypass methods, including 19 newly discovered methods.
      Virus attachment detection bypass: All 16 tested email products exhibited virus detection capability before protocol-level modifications.
  WWW/2024/unfiltered-measuring-cloud-based-email-filtering-bypasses  [3 stated prevalence]
      cloud-based email filtering deployment: 1,601 domains after filtering: 673 edu and 928 com
      email hosting provider: 673 edu and 928 com domains remained after active-use filtering
      email filtering bypassability: 80% of domains; 1,262 of 1,577 evaluated configurations
  IEEE-SP/2010/investigation-of-triangular-spamming-a-stealthy-and-efficient-spamming-technique  [8 stated prevalence]
      outbound SMTP blocking: 97% of the blocking networks are vulnerable to triangular spamming
      inbound versus outbound SMTP blocking: only 22 prefixes out of 688 blocking prefixes (3%) appear to deploy IN-traffic-blocking policy
      IP spoofing feasibility: 313 successes to the local mail server; 312 successes to Yahoo and Gmail in the US
      triangular-spamming deployment: triangular spamming successfully deployed on PlanetLab
  NDSS/2025/hades-attack-understanding-and-evaluating-manipulation-risks-of-email-blocklists  [8 stated prevalence]
      DNSBL adoption: 307,244 domains; 90.06% relied on Spamhaus
      DNSBL capture-server operation: 29 DNSBL providers
      Spamtrap discovery: 140,449 domains
      DNSBL manipulation: 14 DNSBL providers were successfully manipulated
  IEEE-SP/2011/design-and-evaluation-of-a-real-time-url-spam-filtering-service  [7 stated prevalence]
      URL spam: 90.78% accuracy with 0.87% false positives
      Feature persistence: 77% of initial URL features in email spam disappear after 15 days
      Email/Twitter feature overlap: 90% of email and tweet features never overlap
      Redirect behavior: 67% of tweet spam URLs and 20% of email spam URLs contain redirects
  IEEE-SP/2011/click-trajectories-end-to-end-analysis-of-the-spam-value-chain  [6 stated prevalence]
      spam-advertised URL coverage: 98.1% of received URLs
      web redirection: 32% of crawled URLs redirected at least once
      affiliate-program storefronts: 45 affiliate programs across 69,002 registered domains
      payment-bank concentration: Just three banks provided payment servicing for over 95%
  IMC/2025/do-spammers-dream-of-electric-sheep-characterizing-the-prevalence-of-llm-generat  [2 stated prevalence]
      LLM-generated malicious emails: At least 51% of spam and 14.4% of BEC emails in April 2025
      Rewritten spam variants: The five largest clusters contained 78.9%, 52.1%, 8.4%, 8.4%, and 6.6% LLM-generated emails
 
--- SM — 23 papers
  USENIX/2011/sms-of-death-from-analyzing-to-attacking-mobile-phones-on-a-large-scale  [3 stated prevalence]
      SMS-triggered phone crashes and reboots: vulnerabilities in feature phones from six major manufacturers
      SMS-triggered bricking: two phones could no longer be reset or restored
      Network-assisted SMS retransmission: all tested operators retransmitted after one minute and several times every five minutes
  IMC/2012/crime-scene-investigation-sms-spam-data-analysis  [7 stated prevalence]
      SMS spam messaging volume: Spammers generate two orders of magnitude more messages than cell-phone users and one order of magnitude greater than most M2M systems.
      Geographical sources of messaging abuse: Sacramento, Los Angeles-Orange County, and Miami Beach were identified as major U.S. hot-spots.
      Spammer mobility: The maximum observed displacement between cell sites was 15 miles.
      Hardware concentration among spammers: 83% of identified spammers used one of the top five devices; 65% used the top device.
  USENIX/2013/greystar-fast-and-accurate-detection-of-sms-spam-numbers-in-large-cellular-netwo  [3 stated prevalence]
      SMS spam numbers: over 34K spam numbers in five months and two false positives
      SMS spam detection delay: median detection time of 1.2 hours; at least one day ahead in 50% of cases
      Spam traffic reduction: 75% reduction during peak hours
  NDSS/2015/phoneypot-data-driven-understanding-of-telephony-threats  [7 stated prevalence]
      Unsolicited telephony calls: 1,297,517 calls from 252,621 unique sources
      Phone-number age effect: Older blocks received significantly more calls; t-test p=0.005
      Geographic call-volume effects: No significant population, elderly, Asian-immigrant, or Hispanic/Latino effects
      Telemarketer calling patterns: Telemarketer pattern showed increasing calls and increasing unique targets
  CCS/2016/new-security-threats-caused-by-ims-based-sms-service-in-4g-lte-networks  [7 stated prevalence]
      IMS-based SMS vulnerabilities: Vulnerabilities were validated in two top-tier US carriers.
      SMS-powered service vulnerabilities: 28 out of 40 SMS-powered services were vulnerable.
      SMS client denial of service: Samsung S5 failed at least 3,825 messages/second; LG G3 crashed after more than 140K messages.
      Unauthorized donation: A validated unauthorized $10 donation was charged to the victim's bill.
  NDSS/2017/fbs-radar-uncovering-fake-base-stations-at-scale-in-the-wild  [7 stated prevalence]
      High-signal-strength fake base stations: 0.23%
      Invalid fake-base-station identifiers: 0.15%
      Fraud SMS from fake base stations: 0.16%
      Geographically inconsistent base stations: at least 4.1%
  CCS/2020/lies-in-the-air-characterizing-fake-base-station-spam-ecosystem-in-china  [8 stated prevalence]
      FBS spam messages: 279,017 logs spanning 97 days
      Spam business categories: Illegal businesses accounted for over 75% of FBS messages.
      Spam campaigns: 7,884 spam campaigns associated with 8,316 unique contacts
      FBS spammer locations: Spammers were concentrated near main roads and populated regions.
  NDSS/2020/are-you-going-to-answer-that-measuring-user-responses-to-anti-robocall-application-indicators  [5 stated prevalence]
      Anti-robocall warning designs: 80% used red to indicate an incoming spam call
      User preferences for warning indicators: Participants favored blue backgrounds, checkmarks, X marks, and authenticated Caller ID
      Responses to mock robocall warnings: Answered calls decreased by 43% with the Avail-Spam warning
      Authenticated Caller ID effect: Focus-AID produced 42% acceptance versus 35% for Control
  USENIX/2020/whos-calling-characterizing-robocalls-through-audio-and-metadata-analysis  [8 stated prevalence]
      unsolicited phone calls: 1,481,201 unsolicited phone calls
      high-volume call storms: 648 instances across 223 inbound lines
      voicemail spam: 2.91% of calls
      neighbor spoofing: 6.12% used matching NPA and NXX
  NDSS/2021/on-the-insecurity-of-sms-one-time-password-messages-against-local-attackers-in-modern-mobile-devices  [4 stated prevalence]
      Vulnerable SMS-authentication apps: 36 vulnerable apps sharing more than 230 million installations
      Phishing-induced OTP disclosure: 45% to 71% across scenarios
      SMS permission misunderstanding: 39% knew SMS access could compromise other application accounts
      Modern SMS API usage: 20 confirmed and 16 fixed apps among 140,586 analyzed apps
  CCS/2022/clues-in-tweets-twitter-guided-discovery-and-analysis-of-sms-spam  [8 stated prevalence]
      SMS spam discovery: 21,918 unique messages in 75 languages
      SMS screenshot detection: 24,831 SMS images identified from 50,545 images
      SMS text recovery: 90% exact message recovery on 1,000 sampled images
      Spam URL threat detection: 16.53% of spam URLs flagged by at least one engine; 6.50% by five
  IMC/2022/exploring-the-security-and-privacy-risks-of-chatbots-in-messaging-services  [6 stated prevalence]
      administrator permission requests: 8,521 of 15,525 valid chatbots (54.86%)
      missing privacy policies: 95.67% lacked a privacy policy
      incomplete privacy-policy traceability: No chatbot had complete traceability; policies showed partial traceability
      unauthorized canary-token access: 1 of 500 chatbots; Melonian accessed the word document and URL
  NDSS/2023/hope-of-delivery-extracting-user-locations-from-mobile-instant-messengers  [5 stated prevalence]
      receiver location classification: more than 80% accuracy for three locations within one city
      receiver country classification: up to 86% for five timings per sample in Threema
      network connection classification: up to 95% in Germany for individual devices
      messenger server infrastructure: 410 WhatsApp server instances using 143 location identifiers
  USENIX/2023/diving-into-robocall-content-with-snorcall  [8 stated prevalence]
      robocall campaigns: 26,791 campaigns consisting of 232,723 calls
      English versus non-English calls: 25,206 (94.08%) English campaigns and 1,585 (5.92%) Non-English campaigns
      robocall topics: Financial campaigns were 18.40%; tech support campaigns were 10.70%
      call-to-action verbs: 72.79% (18,348) campaigns used one of 131 call-to-action verbs
  USENIX/2023/freaky-leaky-sms-extracting-user-locations-by-analyzing-sms-timings  [3 stated prevalence]
      SMS delivery-report timing location inference: up to 96% for locations across different countries; 86% for two locations within Belgium
      SMS timing side-channel: 96% overseas-versus-domestic accuracy
      Unknown-location detection: 88% accuracy
  IMC/2024/poster-a-comprehensive-categorization-of-sms-scams  [3 stated prevalence]
      SMS scam sector distribution: Delivery was the most prevalent category: 830.2k recipients, 132.4k successful messages, and 1.3m blocked messages.
      Firewall evasion: Telecom impersonation scams were the most successful in evading detection.
      SMS scam campaign grouping: 2.3k unique campaign IDs containing 3.58m SMS messages.
  IEEE-SP/2016/sending-out-an-sms-characterizing-the-security-of-the-sms-ecosystem-with-public  [8 stated prevalence]
      SMS gateway activity: 386,327 messages over 14 months
      Sensitive information in SMS: 522 messages with email addresses
      Authentication-code entropy: 65% of services that use SMS to deliver codes
      Phone-verified-account evasion: 67.6% of messages contained a code
  IEEE-SP/2025/blind-users-really-do-heed-aural-telephone-scam-warnings  [2 stated prevalence]
      Responses to aural telephone scam warnings: Only two legally blind participants pressed one, both after the contextual Social Security warning.
      Telephone scam detection cues: 52 participants used an unknown number as an indicator; 34 used the area code.
  USENIX/2025/hey-mum-i-dropped-my-phone-down-the-toilet-investigating-hi-mum-and-dad-sms-scam  [5 stated prevalence]
      interactive SMS scam ecosystem: 582 unique mule accounts; more than £577k requested over 13 weeks
      scam lure principles: All scammers used kindness and distraction; 749 messages contained urgency cues
      mobile-number lifetime: 14 days for originating sender IDs; 46 days for scammer mobile numbers
      scammer communities: Six communities, plus three smaller communities requesting over £10k
  NDSS/2026/sipconfusion-exploiting-sip-semantic-ambiguities-for-caller-id-and-sms-spoofing  [7 stated prevalence]
      ambiguity-based SIP identity spoofing: 8442 instances across 16,200,000 samples
      server-user-agent spoofing vulnerabilities: 47 of 54 combinations, over 80%
      authenticated identity spoofing: All tested SIP servers permitted some authenticated spoofing
      unauthenticated identity spoofing: Kamailio, OpenSIPS, and ejabberd forwarded some unauthenticated requests
  IEEE-SP/2025/characterizing-robocalls-with-multiple-vantage-points  [8 stated prevalence]
      Robocall campaigns: WavLM clustered 76.55% of benchmark calls into campaigns
      Campaign similarity across honeypots: A 90% similarity threshold identified common campaigns
      Ringless voicemail: 362 campaigns used voicemail injection techniques
      Callback numbers: 6,010 callback numbers from 17,447 Robocall Observatory campaigns
  IMC/2025/fishing-for-smishing-understanding-sms-phishing-infrastructure-and-strategies-by  [8 stated prevalence]
      Public smishing reports and attachments: 220,585 posts and 64,284 image attachments
      Extracted smishing messages: 27,718 unique messages, 19,314 sender IDs, and 20,060 URLs
      Scam categories: Banking was 45.1% (15,277 of 33,869 messages).
      Message languages: 66 languages; English comprised 65.2%.
  USENIX/2025/preventing-artificially-inflated-sms-attacks-through-large-scale-traffic-inspect  [3 stated prevalence]
      AIT attack traffic: 4,947,099 of about 9.4 million samples
      AIT attack detection: 89.6% recall at 0.2% FPR for web; 91.1% recall at 0.1% FPR for native
      Scattered AIT attack behavior: 96% of native bots submitted one or two requests before switching identifiers
 
--- INST — 12 papers
  IMC/2016/what-happens-after-you-are-pwnd-understanding-the-use-of-leaked-webmail-credenti  [7 stated prevalence]
      malicious Gmail-account activity: 326 unique accesses; 147 emails opened, 845 sent, and 12 drafts
      attacker activity types: 224 curious, 82 gold-digger, 36 hijacker accesses; 8 accounts sent email
      Tor-based access: 132 of 326 accesses came from Tor exit nodes
      access-location malleability: Paste-site UK-location accesses had a 1400-kilometer radius versus 1784 without location
  USENIX/2016/hey-you-have-a-problem-on-the-feasibility-of-large-scale-web-vulnerability-notif  [4 stated prevalence]
      WordPress vulnerability exploitability: 74.5% of Web applications remained exploitable after the month-long experiment.
      Client-Side XSS exploitability: Notification groups had 8.6%–16.8% fixed domains, versus 2.2% in control.
      Notification report reachability: 2,064 of 35,832 transmitted vulnerability reports were received successfully (5.8%).
      Post-notification vulnerability fixing: About 40% fixed vulnerabilities after reading a notification report.
  IMC/2017/tripwire-inferring-internet-site-compromise  [3 stated prevalence]
      Site compromise: 19 compromises among approximately 2,302 sites with successful registrations
      Password compromise type: Hard-password accounts were accessed at ten sites; easy-only access occurred at eight sites.
      Attacker email-account activity: 1,316 distinct IPs across approximately 1,792 login attempts
  CCS/2020/the-cookie-hunter-automated-black-box-auditing-for-web-authentication-and-author  [5 stated prevalence]
      authentication-cookie exposure over HTTP: 12,014 domains (48.43%) initially; 10,921 remained vulnerable after HSTS analysis
      JavaScript cookie stealing exposure: 5,680 domains (22.9%) lacked httpOnly protection; 5,099 also embedded third-party scripts
      personal-data leakage after cookie hijacking: 9,324 domains according to the abstract; Table 3 reports 9,122 (78%)
      HSTS deployment errors: 10,495 had no HSTS; 426 had faulty HSTS and remained vulnerable
  PETS/2020/canarytrap-detecting-data-misuse-by-third-party-apps-on-online-social-networks  [7 stated prevalence]
      Unrecognized email use of honeytokens: 422 unrecognized emails associated with 20 apps; 16 apps had characterized unrecognized use
      Malicious honeytoken misuse: 76 malicious emails received on honeytokens shared with 3 apps
      Unrelated email use: 79 unrelated emails received on honeytokens shared with 9 apps
      Advertiser use of honeytokens: 9 of 47 advertisers were unrecognized
  USENIX/2021/effective-notification-campaigns-on-the-web-a-matter-of-trust-framing-and-suppor  [5 stated prevalence]
      Google Analytics without IP Anonymization: 12.7% of approximately 1.3 million German websites
      Website remediation after notification: 56.6% of notified owners versus 9.2% of controls
      Long-term recurrence of misconfiguration: 78 of 2224 previously compliant websites were non-compliant in April
      CheckGA usage: 46.9% of notified owners used CheckGA at least once
  CCS/2022/cart-ology-intercepting-targeted-advertising-via-ad-network-identity-entanglemen  [5 stated prevalence]
      victim merchant inference: target website received 167% more ads in normalized space
      victim product inference: victim-interacted item was served 64% more than any other item
      entanglement latency: roughly 15 hours
      cross-network entanglement: victim activity was 35% higher than any other advertiser
  PETS/2022/leave-no-data-behind-empirical-insights-into-data-erasure-from-online-services  [5 stated prevalence]
      Manual account-data deletion: 76 of 83 services (92%) enabled deletion of at least one category
      Incomplete account erasure: 24 of 90 services (27%)
      Residual personal data after erasure: 14 of 78 responding services
      Newsletter persistence: 12 of 67 services (18%)
  PETS/2022/revisiting-identification-issues-in-gdpr-right-of-access-policies-a-technical-an  [3 stated prevalence]
      vulnerable SAR authentication policies: 17 of 40 organizations were vulnerable in 2021
      persistent policy weaknesses after disclosure: 53% of previously vulnerable organizations remained unimproved; 27% of previously non-vulnerable organizations weakened their policies
      sensitive personal-data leakage: All vulnerable organizations leaked at least some sensitive personal information
  PETS/2023/comparing-large-scale-privacy-and-security-notifications  [7 stated prevalence]
      missing privacy policy: 17.44% of sites with at least one privacy issue
      third-party cookies without consent: 7.34% of sites with at least one privacy issue
      third-party cookies before consent: 7.57% of sites with at least one privacy issue
      personal-data input fields without HTTPS: 2.85% of sites with at least one privacy issue
  IEEE-SP/2026/consumer-beware-exploring-data-brokers-ccpa-compliance  [7 stated prevalence]
      VCR response compliance: 57% responded; 43% did not respond
      CCPA response deadlines: 51.5% responded within 45 calendar days
      Identity-verification PII requests: Email requested by 444/454 brokers and name by 431/454
      VCR submission burden: 79 seconds per broker; 9 hours 57 minutes total
  USENIX/2025/am-i-infected-lessons-from-operating-a-large-scale-iot-security-diagnostic-servi  [5 stated prevalence]
      Malware infection: 171 users (0.15%)
      IoT vulnerabilities: 417 users (0.36%)
      Remediation after malware notification: 59 of 67 (88%)
      Remediation after vulnerability notification: 75 of 151 (50%)
 
==============================================================================
E. SLICE (b): UNSUBSCRIPTION. Full-text probes over all 5,859 papers.
==============================================================================
Probe                                                     Papers (of 5,859)
--------------------------------------------------------  -----------------
List-Unsubscribe (header, RFC 2369)                       1                
one-click unsubscribe / RFC 8058 / List-Unsubscribe-Post  0                
RFC 2369 (cited by number)                                0                
CAN-SPAM                                                  2                
the word "unsubscrib*"                                    56               
"opt out of the mailing/newsletter/list"                  2                
 
The one `List-Unsubscribe` hit, read by hand:
  USENIX/2022/a-large-scale-and-longitudinal-measurement-study-of-dkim-deployment
  — names the header in a table of RFC 6376 signable headers. Not measured.
 
Both `CAN-SPAM` hits, read by hand:
  WWW/2020/finding-a-choice-in-a-haystack…  — opt-out statements extracted from privacy-policy TEXT
  CCS/2023/poster-towards-a-dataset…        — motivational aside in the introduction
 
The 56 `unsubscrib*` papers, split by hand:
   10  MQTT / pub-sub / SDN protocol verb
    3  CCPA / CPRA "do not sell", or an advertising opt-out cookie
    4  turning off a security or account notification
   39  remainder: one passing mention each, in papers about something else
 
Papers whose OBJECT of measurement is an unsubscription mechanism: 0.
The nearest thing in the corpus is one compliance sub-check:
  PETS/2022  Missing unsubscribe or legal notice: 16% of websites did not provide an unsubscribe method or legal notice.
 
==============================================================================
F. THE LEAD-MARKETING PAPER THE BRIEF NAMED IS NOT IN THE EXTRACTION
==============================================================================
extraction record for understanding-data-collection-brokerage-and-spam-in-the-lead-marketing-ecosystem: ABSENT
bibliographic index record  : PRESENT
  Understanding Data Collection, Brokerage, and Spam in the Lead Marketing Ecosystem.
  Yash Vekaria, Nurullah Demir, Konrad Kollnig, Zubair Shafiq
  doi 10.1109/sp63933.2026.00162
selection (labels.jsonl)    : NEVER SCREENED
why: IEEE-SP 2026 has 252 index records and only 58 with an OpenAlex abstract;
     selection screens on the abstract, so 194 of that venue-year were never screened.
     the same gap in WWW 2026: 954 records, 630 with an abstract.
 
==============================================================================
G. QUOTE SPOT-CHECKS
==============================================================================
detection evidence quotes on the 70 on-page papers: 400
  verbatim                     : 220
  located after normalisation  : 118
  below threshold              : 62
Below threshold is NOT "unsupported" — each is listed so it can be read by hand:
  USENIX/2011/show-me-the-money-characterizing-spam-advertised-revenue
      53% of 5-word windows: Based on the total number of visitors where we have referrer information, the conversion percentage on an IP b…
  USENIX/2011/sms-of-death-from-analyzing-to-attacking-mobile-phones-on-a-large-scale
      55% of 5-word windows: Through our analysis we discovered vulnerabilities in the feature phone platforms of all major manufacturers.…
  IMC/2012/crime-scene-investigation-sms-spam-data-analysis
      50% of 5-word windows: On average, spammers make many more phone calls than legitimate users, however the average number of phone cal…
  USENIX/2012/impact-of-spam-exposure-on-user-engagement
      42% of 5-word windows: We find that indeed spam exposure leads to significantly, both statistically and economically, lower user enga…
  USENIX/2012/pharmaleaks-understanding-the-business-of-online-pharmaceutical-affiliate-progra
      56% of 5-word windows: repeat orders are an important part of the business, constituting 27% and 38% of average program revenue for G…
  USENIX/2012/pharmaleaks-understanding-the-business-of-online-pharmaceutical-affiliate-progra
      54% of 5-word windows: The graph shows that just 10% of the highest-revenue affiliates account for 75-90% of total program revenue…
  USENIX/2012/pharmaleaks-understanding-the-business-of-online-pharmaceutical-affiliate-progra
      38% of 5-word windows: the median annualized affiliate commissions for GlavMed, SpamIt, and RX-Promotion are just $292, $3,320, and $…
  NDSS/2015/phoneypot-data-driven-understanding-of-telephony-threats
      56% of 5-word windows: There were 1,500 calls made on 30th March 2014 and another 1300 on 31st March. Incidentally, the attacks laste…
  NDSS/2014/scambaiter-understanding-targeted-nigerian-scams-on-craigslist
      56% of 5-word windows: the majority of the phone numbers, 91%, are registered within the United States…
  CCS/2016/new-security-threats-caused-by-ims-based-sms-service-in-4g-lte-networks
      45% of 5-word windows: when the malware sends SIP messages to the SMS client with the speed at least 3,825 messages ... per second, t…
  CCS/2016/new-security-threats-caused-by-ims-based-sms-service-in-4g-lte-networks
      50% of 5-word windows: We employ DonateARC to make an unauthorized $10 donation to ARC from the victim's mobile bill.…
  CCS/2016/new-security-threats-caused-by-ims-based-sms-service-in-4g-lte-networks
      23% of 5-word windows: The snapshot of the victim account's activity logs ... confirms that those three attack actions are successful…
  IMC/2016/what-happens-after-you-are-pwnd-understanding-the-use-of-leaked-webmail-credenti
      50% of 5-word windows: within the first 25 days after leak, we recorded 80% of all unique accesses to accounts leaked to paste sites,…
  PETS/2016/are-you-sure-you-want-to-contact-us-quantifying-the-leakage-of-pii-via-website-c
      43% of 5-word windows: Our crawler identified a contact page on 29.6% of the 100K websites... successfully located contact forms on 1…
  PETS/2016/are-you-sure-you-want-to-contact-us-quantifying-the-leakage-of-pii-via-website-c
      54% of 5-word windows: we discovered that 6.1% (1,035) of all submitted contact forms leak a user's PII to third parties.…
  PETS/2016/are-you-sure-you-want-to-contact-us-quantifying-the-leakage-of-pii-via-website-c
      47% of 5-word windows: Over the duration of our experiment we received 309 emails from third parties, that is, domains which our craw…
  USENIX/2016/hey-you-have-a-problem-on-the-feasibility-of-large-scale-web-vulnerability-notif
      56% of 5-word windows: out of 35,832 transmitted vulnerability reports, only 2,064 (5.8%) were actually received successfully…
  USENIX/2016/hey-you-have-a-problem-on-the-feasibility-of-large-scale-web-vulnerability-notif
      57% of 5-word windows: If a notification report is read by the owner of the vulnerable application, the likelihood of a subsequent re…
  NDSS/2017/fbs-radar-uncovering-fake-base-stations-at-scale-in-the-wild
      58% of 5-word windows: we conclude that Set-4 includes at least 4.1% of suspicious messages, corresponding to δ = 5.0.…
  NDSS/2017/fbs-radar-uncovering-fake-base-stations-at-scale-in-the-wild
      50% of 5-word windows: Set-5 includes at least 0.39% of suspicious messages, using the most conservative combination Vmin + threshold…
  NDSS/2017/fbs-radar-uncovering-fake-base-stations-at-scale-in-the-wild
      56% of 5-word windows: the median (mean) deviation distance of FBS localization is only 11 (149) meters.…
  PETS/2018/i-never-signed-up-for-this-privacy-implications-of-email-tracking
      40% of 5-word windows: The average Jaccard similarity between the sets of third parties loaded during the first and second views of t…
  PETS/2019/investigating-sources-of-pii-used-in-facebook-s-targeted-advertising
      50% of 5-word windows: We found that the previously-unused phone number became targetable in 36 days, showing that it had indeed been…
  PETS/2020/no-boundaries-data-exfiltration-by-third-parties-embedded-on-web-pages
      52% of 5-word windows: We found seven loosely-integrated third-party scripts collecting Facebook user data using the first party's Fa…
  USENIX/2021/effective-notification-campaigns-on-the-web-a-matter-of-trust-framing-and-suppor
      58% of 5-word windows: 56.6 % of notified owners remediated the issue, compared to 9.2 % in the control group.…
  USENIX/2021/effective-notification-campaigns-on-the-web-a-matter-of-trust-framing-and-suppor
      56% of 5-word windows: Under this assumption, half of the notified owners (46.9 %) have used the tool at least once for their site(s)…
  CCS/2022/clues-in-tweets-twitter-guided-discovery-and-analysis-of-sms-spam
      59% of 5-word windows: On randomly sampled 1000 SMS images, STR achieves an accuracy of 90%, i.e. accurately recognizing 90% of messa…
  CCS/2022/clues-in-tweets-twitter-guided-discovery-and-analysis-of-sms-spam
      45% of 5-word windows: Among these 280 spam campaigns, 53 of them contain SMS spam messages in different languages.…
  CCS/2022/clues-in-tweets-twitter-guided-discovery-and-analysis-of-sms-spam
      56% of 5-word windows: As a result, it falsely reported 103 out of 124 benign SMS messages.…
  PETS/2022/revisiting-identification-issues-in-gdpr-right-of-access-policies-a-technical-an
      55% of 5-word windows: 53% of the previously vulnerable organizations have not corrected their policy and an additional 27% of previo…
  PETS/2022/revisiting-identification-issues-in-gdpr-right-of-access-policies-a-technical-an
      50% of 5-word windows: We conclude that all data responses of the vulnerable organizations contained sensitive personal information…
  WWW/2022/left-or-right-a-peek-into-the-political-biases-in-email-spam-filtering-algorithm
      50% of 5-word windows: After the five S→I interactions, on average, Gmail marks just 5.34% of the right emails as spam...and 0% of th…
  WWW/2022/left-or-right-a-peek-into-the-political-biases-in-email-spam-filtering-algorithm
      53% of 5-word windows: Neither the age group, nor the ethnicity, nor the gender of the account holder had any impact on how SFAs trea…
  NDSS/2023/hope-of-delivery-extracting-user-locations-from-mobile-instant-messengers
      56% of 5-word windows: In the setup in Germany, we can detect the receiver's Internet connection type with high accuracy for all devi…
  NDSS/2023/hope-of-delivery-extracting-user-locations-from-mobile-instant-messengers
      54% of 5-word windows: In total, our set of discovered WhatsApp servers comprises 410 server instances using 143 different location i…
  PETS/2023/comparing-large-scale-privacy-and-security-notifications
      57% of 5-word windows: followed by Before Consent (7.57 %) and No Consent (7.34 %).…
  PETS/2023/comparing-large-scale-privacy-and-security-notifications
      57% of 5-word windows: followed by Before Consent (7.57 %) and No Consent (7.34 %).…
  PETS/2023/comparing-large-scale-privacy-and-security-notifications
      20% of 5-word windows: No HTTPS was rarest, with 2.85 % of sites.…
  PETS/2023/comparing-large-scale-privacy-and-security-notifications
      53% of 5-word windows: For a given day 𝑡𝑖, we considered a domain 𝑑 to be problematic if at any point in time within a 7-day windo…
  USENIX/2023/diving-into-robocall-content-with-snorcall
      56% of 5-word windows: 881 callback numbers were used in more than one campaign, and 2344 callback numbers were used in only one camp…
  CCS/2024/inbox-invasion-exploiting-mime-ambiguities-to-evade-email-attachment-detectors
      29% of 5-word windows: We conducted tests using the anti-virus test file EICAR, the exe virus WannaCry, and a PDF virus, respectively…
  IEEE-SP/2016/sending-out-an-sms-characterizing-the-security-of-the-sms-ecosystem-with-public
      58% of 5-word windows: Our analysis tracks over 400 phone numbers in 28 countries over the course of 14 months, resulting in a datase…
  IEEE-SP/2016/sending-out-an-sms-characterizing-the-security-of-the-sms-ecosystem-with-public
      50% of 5-word windows: We found 1.0% of tagged messages across 32 clusters related to advertising.…
  IEEE-SP/2016/sending-out-an-sms-characterizing-the-security-of-the-sms-ecosystem-with-public
      20% of 5-word windows: VirusTotal returned 417 URLs with at least one detection.…
  USENIX/2025/doubly-dangerous-evading-phishing-reporting-systems-by-leveraging-email-tracking
      40% of 5-word windows: each reporting service gives a unique 'post-report' signal-Gmail's via its User-Agent header, Outlook's via it…
  USENIX/2025/hey-mum-i-dropped-my-phone-down-the-toilet-investigating-hi-mum-and-dad-sms-scam
      55% of 5-word windows: Our research finds that the median lifetime of originating sender ID mobile numbers abused to send initial 'hi…
  NDSS/2025/hades-attack-understanding-and-evaluating-manipulation-risks-of-email-blocklists
      53% of 5-word windows: we conducted an empirical study on 29 DNSBL providers by reviewing their websites and public reports, as summa…
  NDSS/2025/hades-attack-understanding-and-evaluating-manipulation-risks-of-email-blocklists
      53% of 5-word windows: We find that 39,201 (76.88%) outgoing mail servers of popular ESPs and websites have been included in at least…
  NDSS/2025/hades-attack-understanding-and-evaluating-manipulation-risks-of-email-blocklists
      50% of 5-word windows: Among these blocklisted domains, the status of 398 (5.67%) domains is serverHold.…
  NDSS/2026/connecting-the-dots-an-investigative-study-on-linking-private-user-data-across-messaging-apps
      56% of 5-word windows: Among the 63 websites that expose their tokens, 56 websites deploy third-party scripts.…
  NDSS/2026/sipconfusion-exploiting-sip-semantic-ambiguities-for-caller-id-and-sms-spoofing
      50% of 5-word windows: over 80% (47 out of 54) of server and user agent combinations are susceptible to ambiguity-based spoofing atta…
  NDSS/2026/sipconfusion-exploiting-sip-semantic-ambiguities-for-caller-id-and-sms-spoofing
      58% of 5-word windows: We found that all the VoIP devices are vulnerable to ambiguity-based spoofing attacks to some extent.…
  NDSS/2026/sipconfusion-exploiting-sip-semantic-ambiguities-for-caller-id-and-sms-spoofing
      33% of 5-word windows: real-world deployed services such as Linphone and Linhome are affected by the P-Preferred-Identity preferred a…
  NDSS/2026/sipconfusion-exploiting-sip-semantic-ambiguities-for-caller-id-and-sms-spoofing
      46% of 5-word windows: all three operators are vulnerable to A4 attack, and one of them is vulnerable to B3 attack.…
  NDSS/2026/hey-there-you-are-using-whatsapp-enumerating-three-billion-accounts-for-security-and-privacy
      55% of 5-word windows: We found that 0.8 % of main devices and 13 % of companion devices did not have a one-time prekey at the scan t…
  NDSS/2026/hey-there-you-are-using-whatsapp-enumerating-three-billion-accounts-for-security-and-privacy
      57% of 5-word windows: Our analysis across 3.5 B accounts reveals an Android share of 81 % and iOS at 19 %.…
  IEEE-SP/2011/design-and-evaluation-of-a-real-time-url-spam-filtering-service
      50% of 5-word windows: We identify web service spam with 90.78% accuracy (0.87% false positives), with a median feature collection an…
  IEEE-SP/2011/design-and-evaluation-of-a-real-time-url-spam-filtering-service
      56% of 5-word windows: Notably, 77% of initial URL features appearing in email disappear after 15 days.…
  IEEE-SP/2011/click-trajectories-end-to-end-analysis-of-the-spam-value-chain
      43% of 5-word windows: We attempted 120 purchases, of which 76 authorized and 56 settled.…
  IMC/2025/fishing-for-smishing-understanding-sms-phishing-infrastructure-and-strategies-by
      47% of 5-word windows: Table 1: Overview of our smishing dataset collected using posts (n = 220, 585) and image attachments (n = 64, …
  IMC/2025/fishing-for-smishing-understanding-sms-phishing-infrastructure-and-strategies-by
      18% of 5-word windows: Overall, we collect a dataset with 27.7k smishing messages, 19.3k sender IDs, and 20k URLs.…
  USENIX/2025/addressing-the-address-books-interdependent-privacy-issues
      54% of 5-word windows: even without any financial incentive, 19.9% of the respondents chose to grant access to their DAB data…
 
==============================================================================
H. THE RESIDUE. Everything screened and not put on the page.
==============================================================================
 
--- INFRA — 31 papers — email transport, authentication and encryption deployment (SPF/DKIM/DMARC/DANE/MTA-STS/STARTTLS/S-MIME, spoofing, delivery) — a page-sized slice of its own, not tracking
  CCS/2015/security-by-any-other-name-on-the-effectiveness-of-provider-based-email-security
      provider-side mail security
  IMC/2015/neither-snow-nor-rain-nor-mitm-an-empirical-analysis-of-email-delivery-security
      STARTTLS / SPF / DKIM / DMARC deployment
  IMC/2017/email-typosquatting
      doppelganger mail domains catching misdirected mail
  USENIX/2018/efail-breaking-s-mime-and-openpgp-email-encryption-using-exfiltration-channels
      S/MIME and OpenPGP exfiltration channels
  USENIX/2018/end-to-end-measurements-of-email-spoofing-attacks
      spoofing reaching the inbox and what the UI shows
  USENIX/2019/johnny-you-are-fired-spoofing-openpgp-and-s-mime-signatures-in-emails
      signature spoofing in mail clients
  USENIX/2020/a-longitudinal-and-comprehensive-study-of-the-dane-ecosystem-in-email
      DANE for SMTP
  USENIX/2020/composition-kills-a-case-study-of-email-sender-authentication
      sender-authentication composition flaws
  USENIX/2021/weak-links-in-authentication-chains-a-large-scale-analysis-of-email-sender-spoof
      sender spoofing across 30 providers
  USENIX/2021/why-tls-is-better-without-starttls-a-security-analysis-of-starttls-in-the-email
      STARTTLS
  IMC/2022/spfail-discovering-measuring-and-remediating-vulnerabilities-in-email-sender-val
      SPF implementation vulnerabilities
  USENIX/2022/a-large-scale-and-longitudinal-measurement-study-of-dkim-deployment
      DKIM deployment
  USENIX/2022/under-the-hood-of-dane-mismanagement-in-smtp
      DANE mismanagement
  WWW/2022/revisiting-email-forwarding-security-under-the-authenticated-received-chain-prot
      ARC and forwarding
  PETS/2023/heads-in-the-clouds-measuring-universities-migration-to-public-clouds-implicatio
      MX-record concentration as the measurement
  IMC/2023/lazy-gatekeepers-a-large-scale-study-on-spf-configuration-in-the-wild
      SPF configuration
  USENIX/2023/youve-got-report-measurement-and-security-implications-of-dmarc-reporting
      DMARC aggregate reporting
  USENIX/2023/content-type-multipart-oracle-tapping-into-format-oracles-in-email-end-to-end-en
      format oracles in mail end-to-end encryption
  USENIX/2023/extended-hell-o-a-comprehensive-large-scale-study-on-email-confidentiality-and-i
      confidentiality and integrity mechanisms in the wild
  IMC/2024/bounce-in-the-wild-a-deep-dive-into-email-delivery-failures-from-a-large-email-s
      delivery failures at a large ESP
  NDSS/2024/breakspf-how-shared-infrastructures-magnify-spf-vulnerabilities-across-the-internet
      SPF at shared infrastructure
  USENIX/2024/fakebehalf-imperceptible-email-spoofing-attacks-against-the-delegation-mechanism
      spoofing via the on-behalf-of delegation
  USENIX/2024/spf-beyond-the-standard-management-and-operational-challenges-in-practice-and-pr
      SPF operations
  IMC/2025/unraveling-the-complexities-of-mta-sts-deployment-and-management-in-securing-ema
      MTA-STS deployment
  USENIX/2025/email-spoofing-with-smtp-smuggling-how-the-shared-email-infrastructures-magnify
      SMTP smuggling
  NDSS/2025/a-multifaceted-study-on-the-use-of-tls-and-auto-detect-in-email-ecosystems
      TLS and autoconfiguration in mail clients
  USENIX/2025/s-mine-collecting-and-analyzing-s-mime-certificates-at-scale
      S/MIME certificates at scale
  NDSS/2026/coordmail-exploiting-smtp-timeout-and-command-interaction-to-coordinate-email-middleware-for-convergence-amplification-attack
      SMTP middleware amplification
  IMC/2025/understanding-and-characterizing-intermediate-paths-of-email-delivery-the-hidden
      the hops between sender and recipient
  NDSS/2025/automatic-insecurity-exploring-email-auto-configuration-in-the-wild
      mail autoconfiguration
  IEEE-SP/2022/27-years-and-81-million-opportunities-later-investigating-the-use-of-email-encry
      27 years of mail encryption at one university
 
--- PHISH — 28 papers — phishing and its interventions, with email as the vector — security:phishing
  USENIX/2014/targeted-threat-index-characterizing-and-quantifying-politically-motivated-targe
  USENIX/2014/when-governments-hack-opponents-a-look-at-actors-and-technology
  USENIX/2014/a-look-at-targeted-attacks-through-the-lense-of-an-ngo
  CCS/2016/phisheye-live-monitoring-of-sandboxed-phishing-kits
  CCS/2017/data-breaches-phishing-or-malware-understanding-the-risks-of-stolen-credentials
  USENIX/2017/detecting-credential-spearphishing-in-enterprise-settings
  USENIX/2019/cognitive-triaging-of-phishing-attacks
  USENIX/2019/detecting-and-characterizing-lateral-phishing-at-scale
  USENIX/2019/high-precision-detection-of-business-email-compromise
  WWW/2019/hack-for-hire-exploring-the-emerging-market-for-account-hijacking
  IMC/2020/who-is-targeted-by-email-based-phishing-and-malware-measuring-factors-that-diffe
  USENIX/2020/sunrise-to-sunset-analyzing-the-end-to-end-life-cycle-and-effectiveness-of-phish
  USENIX/2021/is-real-time-phishing-eliminated-with-fido-social-engineering-downgrade-attacks
  IEEE-SP/2022/phishing-in-organizations-findings-from-a-large-scale-and-long-term-study
  CCS/2024/content-nudges-and-incentives-a-study-on-the-effectiveness-and-perception-of-emb
  CCS/2024/employees-attitudes-towards-phishing-simulations-its-like-when-a-child-reaches-o
  CCS/2024/poster-different-victims-same-layout-email-visual-similarity-detection-for-enhan
  USENIX/2024/assessing-suspicious-emails-with-banner-warnings-among-blind-and-low-vision-user
  USENIX/2024/simulated-stress-a-case-study-of-the-effects-of-a-simulated-phishing-campaign-on
  IEEE-SP/2025/restricting-the-link-effects-of-focused-attention-and-time-delay-on-phishing-war
  USENIX/2025/url-inspection-tasks-helping-users-detect-phishing-links-in-emails
  PETS/2026/linguistic-hooks-investigating-the-role-of-language-triggers-in-phishing-emails
  USENIX/2026/a-large-scale-study-of-personalized-phishing-using-large-language-models
  CCS/2025/phishing-susceptibility-and-the-in-effectiveness-of-common-anti-phishing-interve
  IEEE-SP/2025/understanding-the-efficacy-of-phishing-training-in-practice
  USENIX/2025/phishing-attacks-against-password-manager-browser-extensions
  IEEE-SP/2023/beyond-phish-toward-detecting-fraudulent-e-commerce-websites-at-scale
  IEEE-SP/2023/design-and-evaluation-of-inclusive-email-security-indicators-for-people-with-vis
 
--- SOC — 37 papers — social-platform, review, forum, SEO or ad-click spam — the platform, not a message channel
  CCS/2010/exploiting-social-networking-sites-for-spam
  CCS/2010/detecting-and-characterizing-social-spam-campaigns
      Facebook wall posts
  CCS/2010/spam-the-underground-on-140-characters-or-less
      Twitter
  IMC/2010/detecting-and-characterizing-social-spam-campaigns
      Facebook wall posts
  WWW/2010/the-social-honeypot-project-protecting-online-communities-from-spammers
  IMC/2011/suspended-accounts-in-retrospect-an-analysis-of-twitter-spam
  IMC/2011/uncovering-social-network-sybils-in-the-wild
  NDSS/2012/warningbird-detecting-suspicious-urls-in-twitter-stream
  USENIX/2012/efficient-and-scalable-socware-detection-in-online-social-networks
  CCS/2013/viceroi-catching-click-spam-in-search-ad-networks
      ad click spam
  WWW/2012/spotting-fake-reviewer-groups-in-consumer-reviews
      review spam
  NDSS/2013/compa-detecting-compromised-accounts-on-social-networks
  WWW/2012/understanding-and-combating-link-farming-in-the-twitter-social-network
  WWW/2013/an-analysis-of-socware-cascades-in-online-social-networks
  WWW/2012/analyzing-spammers-social-networks-for-fun-and-profit-a-case-study-of-cyber-crim
  WWW/2012/serf-and-turf-crowdturfing-for-fun-and-profit
  NDSS/2013/neighborwatcher-a-content-agnostic-comment-spam-inference-system
      web comment spam
  USENIX/2013/you-are-how-you-click-clickstream-analysis-for-sybil-detection
  USENIX/2014/towards-detecting-anomalous-user-behavior-in-online-social-networks
  WWW/2015/early-detection-of-spam-mobile-apps
      app-store spam
  WWW/2015/opinion-spam-detection-in-web-forum-a-real-case-study
  WWW/2016/characterizing-long-tail-seo-spam-on-cloud-web-hosting-services
      SEO spam
  WWW/2016/in-a-world-that-counts-clustering-and-detecting-fake-social-engagement-at-scale
  CCS/2017/poised-spotting-twitter-spam-off-the-beaten-paths
  WWW/2017/bimodal-distribution-and-co-bursting-in-review-spam-detection
  WWW/2017/can-you-spot-the-fakes-on-the-limitations-of-user-feedback-in-online-social-netw
  WWW/2017/detecting-collusive-spamming-activities-in-community-question-answering
  WWW/2018/collective-classification-of-spam-campaigners-on-twitter-a-hierarchical-meta-pat
  WWW/2018/detecting-crowdturfing-add-to-favorites-activities-in-online-shopping
  WWW/2019/measurement-and-early-detection-of-third-party-application-abuse-on-twitter
  USENIX/2021/deep-entity-classification-abusive-account-detection-for-online-social-networks
  USENIX/2023/token-spammers-rug-pulls-and-sniper-bots-an-analysis-of-the-ecosystem-of-tokens
      ERC-20 token spam
  WWW/2023/simplistic-collection-and-labeling-practices-limit-the-utility-of-benchmark-data
      Twitter bot benchmarks
  WWW/2025/pirates-of-charity-exploring-donation-based-abuses-in-social-media-platforms
  NDSS/2026/tickets-to-hide-an-inside-look-into-the-anti-abuse-ecosystem-through-internal-abuse-data
      platform abuse tickets across many abuse types
  USENIX/2025/predictive-response-optimization-using-reinforcement-learning-to-fight-online-so
  WWW/2026/visual-content-moderation-in-messaging-systems
      moderation of images in messaging systems, not message abuse measurement
 
--- WEBPIXEL — 7 papers — the tracking pixel on a WEB page rather than in a message — privacy:requests
  USENIX/2019/leaky-images-targeted-privacy-attacks-in-the-web
  PETS/2020/missed-by-filter-lists-detecting-unknown-third-party-trackers-with-invisible-pix
  WWW/2023/the-hitchhikers-guide-to-facebook-web-tracking-with-invisible-pixels-and-click-i
  PETS/2024/johnny-still-cant-opt-out-assessing-the-iab-ccpa-compliance-framework
      CCPA opt-out signalling on web pages, not mail opt-out
  PETS/2024/website-data-transparency-in-the-browser
  CCS/2025/piixel-leaks-passive-identification-of-personally-identifiable-information-leaka
      Meta Pixel on web pages
  WWW/2026/tgnn-enhancing-pixel-tracking-detection-via-llm-driven-annotation-and-gat-powere
      web pixel detection with LLM-assisted labelling
 
--- OFF — 181 papers — homograph or unrelated
  IMC/2010/estimating-and-sampling-graphs-with-multidimensional-random-walks
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2010/inference-and-analysis-of-formal-models-of-botnet-command-and-control-protocols
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2010/dissecting-one-click-frauds
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2010/building-a-dynamic-reputation-system-for-dns
      network-layer work that uses spam as a label source
  USENIX/2010/searching-the-searchers-with-searchaudit
      not individually annotated: screened at title level, matched no off-topic family rule
  IMC/2011/gq-practical-containment-for-measuring-modern-malware-systems
      systems security; the message vocabulary is incidental
  IMC/2011/monitoring-the-initial-dns-behavior-of-malicious-domains
      network-layer work that uses spam as a label source
  USENIX/2011/dirty-jobs-the-role-of-freelance-labor-in-web-service-abuse
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2011/forensic-triage-for-mobile-phones-with-dec0de
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2011/measuring-and-analyzing-search-redirection-attacks-in-the-illicit-online-prescri
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2012/detecting-money-stealing-apps-in-alternative-android-markets
      systems security; the message vocabulary is incidental
  CCS/2012/innocent-by-association-early-recognition-of-legitimate-users
      legitimacy of webmail accounts; the unit is the account, not the message
  CCS/2012/mobile-data-charging-new-attacks-and-countermeasures
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2012/guess-who-s-texting-you-evaluating-the-security-of-smartphone-messaging-applicat
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2012/hey-you-get-off-of-my-market-detecting-malicious-apps-in-official-and-alternativ
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2013/cross-origin-pixel-stealing-timing-attacks-using-css-filters
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2013/whyper-towards-automating-risk-assessment-of-mobile-applications
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2014/dialing-back-abuse-on-phone-verified-accounts
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2014/are-you-ready-to-lock
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2013/detecting-passive-content-leaks-and-pollution-in-android-applications
      systems security; the message vocabulary is incidental
  USENIX/2013/jekyll-on-ios-when-benign-apps-become-evil
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2013/automatically-inferring-the-evolution-of-malicious-activity-on-the-internet
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2013/protecting-sensitive-web-content-from-client-side-vulnerabilities-with-cryptons
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2013/vetting-undesirable-behaviors-in-android-apps-with-permission-use-analysis
      systems security; the message vocabulary is incidental
  CCS/2014/consequences-of-connectivity-characterizing-account-hijacking-on-twitter
      not individually annotated: screened at title level, matched no off-topic family rule
  WWW/2013/two-years-of-short-urls-internet-measurement-security-threats-and-countermeasure
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2014/autocog-measuring-the-description-to-permission-fidelity-in-android-applications
      systems security; the message vocabulary is incidental
  CCS/2014/real-threats-to-your-data-bills-security-loopholes-and-defenses-in-mobile-data-c
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2014/understanding-the-dark-side-of-domain-parking
      not individually annotated: screened at title level, matched no off-topic family rule
  IMC/2014/handcrafted-fraud-and-extortion-manual-account-hijacking-in-the-wild
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2015/leakage-abuse-attacks-against-searchable-encryption
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2015/copperdroid-automatic-reconstruction-of-android-malware-behaviors
      systems security; the message vocabulary is incidental
  NDSS/2015/mind-your-blocks-on-the-stealthiness-of-malicious-bgp-hijacks
      network-layer work that uses spam as a label source
  NDSS/2015/what-s-in-your-dongle-and-bank-account-mandatory-and-discretionary-protection-of
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2015/android-permissions-remystified-a-field-study-on-contextual-integrity
      systems security; the message vocabulary is incidental
  USENIX/2015/cloudy-with-a-chance-of-breach-forecasting-cyber-security-incidents
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2014/on-the-mismanagement-and-maliciousness-of-networks
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2014/man-vs-machine-practical-adversarial-detection-of-malicious-crowdsourcing-worker
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2014/dspin-detecting-automatically-spun-content-on-the-web
      not individually annotated: screened at title level, matched no off-topic family rule
  IMC/2015/affiliate-crookies-characterizing-affiliate-marketing-abuse
      not individually annotated: screened at title level, matched no off-topic family rule
  IMC/2015/detecting-malicious-activity-with-dns-backscatter
      network-layer work that uses spam as a label source
  IMC/2015/leveraging-internet-background-radiation-for-opportunistic-network-analysis
      not individually annotated: screened at title level, matched no off-topic family rule
  IMC/2015/the-dark-menace-characterizing-network-based-attacks-in-the-cloud
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2015/evilcohort-detecting-communities-of-malicious-accounts-on-online-services
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2016/featuresmith-automatically-engineering-features-for-malware-detection-by-mining
      systems security; the message vocabulary is incidental
  CCS/2016/over-the-top-bypass-study-of-a-recent-telephony-fraud
      interconnect (SIM-box) bypass fraud: the object is call routing revenue, not a message delivered to a user
  CCS/2016/predator-proactive-recognition-and-elimination-of-domain-abuse-at-time-of-regist
      not individually annotated: screened at title level, matched no off-topic family rule
  IMC/2016/bdrmap-inference-of-borders-between-ip-networks
      not individually annotated: screened at title level, matched no off-topic family rule
  PETS/2016/tales-from-the-dark-side-privacy-dark-strategies-and-privacy-dark-patterns
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2016/the-ever-changing-labyrinth-a-large-scale-analysis-of-wildcard-dns-powered-black
      network-layer work that uses spam as a label source
  IEEE-SP/2017/obstacles-to-the-adoption-of-secure-communication-tools
      not individually annotated: screened at title level, matched no off-topic family rule
  IMC/2017/understanding-the-role-of-registrars-in-dnssec-deployment
      network-layer work that uses spam as a label source
  NDSS/2017/delta-a-security-assessment-framework-for-software-defined-networks
      pub/sub or SDN "unsubscribe" — protocol verb, not marketing mail
  USENIX/2017/characterizing-the-nature-and-dynamics-of-tor-exit-blocking
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2017/malton-towards-on-device-non-invasive-mobile-malware-analysis-for-art
      systems security; the message vocabulary is incidental
  CCS/2018/detecting-attacks-against-robotic-vehicles-a-control-invariant-approach
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2018/towards-paving-the-way-for-large-scale-windows-malware-analysis-generic-binary-u
      systems security; the message vocabulary is incidental
  IEEE-SP/2018/computer-security-and-privacy-for-refugees-in-the-united-states
      not individually annotated: screened at title level, matched no off-topic family rule
  IMC/2018/is-the-web-ready-for-ocsp-must-staple
      not individually annotated: screened at title level, matched no off-topic family rule
  PETS/2018/toward-distribution-estimation-under-local-differential-privacy-with-small-sampl
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2018/from-patching-delays-to-infection-symptoms-using-risk-profiles-for-an-early-disc
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2018/o-single-sign-off-where-art-thou-an-empirical-analysis-of-single-sign-on-account
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2019/a-systematic-framework-to-generate-invariants-for-anomaly-detection-in-industrial-control-systems
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2019/cleaning-up-the-internet-of-evil-things-real-world-evidence-on-isp-and-consumer-efforts-to-remove-mirai
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2019/privacy-attacks-to-the-4g-and-5g-cellular-paging-protocols-using-side-channel-information
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2019/a-billion-open-interfaces-for-eve-and-mallory-mitm-dos-and-tracking-attacks-on-i
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2019/policylint-investigating-internal-privacy-policy-contradictions-on-google-play
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2019/reading-the-tea-leaves-a-comparative-analysis-of-threat-intelligence
      not individually annotated: screened at title level, matched no off-topic family rule
  WWW/2019/evaluating-login-challenges-as-adefense-against-account-takeover
      not individually annotated: screened at title level, matched no off-topic family rule
  WWW/2019/exploring-user-behavior-in-email-re-finding-tasks
      information retrieval inside a mailbox — no privacy or security measurement
  WWW/2019/studying-preferences-and-concerns-about-information-disclosure-in-email-notifica
      what a notification should say, not what it discloses to third parties
  WWW/2019/understanding-the-evolution-of-mobile-app-ecosystems-a-longitudinal-measurement
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2020/towards-attribution-in-mobile-markets-identifying-developer-account-polymorphism
      not individually annotated: screened at title level, matched no off-topic family rule
  IEEE-SP/2020/the-many-kinds-of-creepware-used-for-interpersonal-attacks
      not individually annotated: screened at title level, matched no off-topic family rule
  IEEE-SP/2020/karonte-detecting-insecure-multi-binary-interactions-in-embedded-firmware
      systems security; the message vocabulary is incidental
  IMC/2020/revisiting-transactional-statistics-of-high-scalability-blockchains
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2020/blag-improving-the-accuracy-of-blacklists
      network-layer work that uses spam as a label source
  NDSS/2020/hotfuzz-discovering-algorithmic-denial-of-service-vulnerabilities-through-guided-micro-fuzzing
      systems security; the message vocabulary is incidental
  NDSS/2020/surfingattack-interactive-hidden-attack-on-voice-assistants-using-ultrasonic-guided-waves
      not individually annotated: screened at title level, matched no off-topic family rule
  PETS/2020/sok-anatomy-of-data-breaches
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2020/security-analysis-of-unified-payments-interface-and-payment-apps-in-india
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2020/teerex-discovery-and-exploitation-of-memory-corruption-vulnerabilities-in-sgx-en
      not individually annotated: screened at title level, matched no off-topic family rule
  WWW/2020/fairrec-two-sided-fairness-for-personalized-recommendations-in-two-sided-platfor
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2021/dont-forget-the-stuffing-revisiting-the-security-impact-of-typo-tolerant-passwor
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2021/alchemist-fusing-application-and-audit-logs-for-precise-attack-provenance-without-instrumentation
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2021/from-whois-to-whowas-a-large-scale-measurement-study-of-domain-registration-privacy-under-the-gdpr
      not individually annotated: screened at title level, matched no off-topic family rule
  PETS/2021/the-motivated-can-encrypt-even-with-pgp
      not individually annotated: screened at title level, matched no off-topic family rule
  PETS/2021/we-three-brothers-have-always-known-everything-of-each-other-a-cross-cultural-st
      not individually annotated: screened at title level, matched no off-topic family rule
  PETS/2021/defining-privacy-how-users-interpret-technical-terms-in-privacy-policies
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2021/alpaca-application-layer-protocol-confusion-analyzing-and-mitigating-cracks-in-t
      network-layer work that uses spam as a label source
  USENIX/2021/effect-of-mood-location-trust-and-presence-of-others-on-video-based-social-authe
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2021/now-im-a-bit-angry-individuals-awareness-perception-and-responses-to-data-breach
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2021/privatedrop-practical-privacy-preserving-authentication-for-apple-airdrop
      contact discovery, but the contribution is a protocol; no measurement of an address space
  USENIX/2021/strategies-and-perceived-risks-of-sending-sensitive-documents
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2021/the-hijackers-guide-to-the-galaxy-off-path-taking-over-internet-resources
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2021/why-older-adults-dont-use-password-managers
      not individually annotated: screened at title level, matched no off-topic family rule
  WWW/2021/an-investigation-of-identity-account-inconsistency-in-single-sign-on
      not individually annotated: screened at title level, matched no off-topic family rule
  WWW/2021/privacy-policies-over-time-curation-and-analysis-of-a-million-document-dataset
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2022/stolenencoder-stealing-pre-trained-encoders-in-self-supervised-learning
      not individually annotated: screened at title level, matched no off-topic family rule
  IMC/2022/stop-drop-and-roa-effectiveness-of-defenses-through-the-lens-of-drop
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2022/auto-draft-238
      not individually annotated: screened at title level, matched no off-topic family rule
  PETS/2022/athena-probabilistic-verification-of-machine-unlearning
      not individually annotated: screened at title level, matched no off-topic family rule
  PETS/2022/user-perceptions-of-gmail-s-confidential-mode
      usability of one provider feature
  USENIX/2022/foap-fine-grained-open-world-android-app-fingerprinting
      systems security; the message vocabulary is incidental
  USENIX/2022/many-roads-lead-to-rome-how-packet-headers-influence-dns-censorship-measurement
      network-layer work that uses spam as a label source
  USENIX/2022/pre-hijacked-accounts-an-empirical-study-of-security-failures-in-user-account-cr
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2022/sgxfuzz-efficiently-synthesizing-nested-structures-for-sgx-enclave-fuzzing
      systems security; the message vocabulary is incidental
  CCS/2023/black-ostrich-web-application-scanning-with-string-solvers
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2023/comprehension-from-chaos-towards-informed-consent-for-private-computation
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2023/fetchbench-systematic-identification-and-characterization-of-proprietary-prefetc
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2023/i-still-know-what-you-watched-last-sunday-privacy-of-the-hbbtv-protocol-in-the-european-smart-tv-landscape
      not individually annotated: screened at title level, matched no off-topic family rule
  PETS/2023/designing-a-location-trace-anonymization-contest
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2023/weve-disabled-mfa-for-you-an-evaluation-of-the-security-and-usability-of-multi-f
      not individually annotated: screened at title level, matched no off-topic family rule
  PETS/2023/no-privacy-among-spies-assessing-the-functionality-and-insecurity-of-consumer-an
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2023/a-large-scale-measurement-of-website-login-policies
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2023/a-two-decade-retrospective-analysis-of-a-universitys-vulnerability-to-attacks-ex
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2023/security-and-privacy-failures-in-popular-2fa-apps
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2023/basecomp-a-comparative-analysis-for-integrity-protection-in-cellular-baseband-so
      systems security; the message vocabulary is incidental
  USENIX/2023/work-from-home-and-covid-19-trajectories-of-endpoint-security-management-in-a-se
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2023/anatomy-of-a-high-profile-data-breach-dissecting-the-aftermath-of-a-crypto-walle
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2023/instructions-unclear-undefined-behaviour-in-cellular-network-specifications
      systems security; the message vocabulary is incidental
  USENIX/2023/notice-the-imposter-a-study-on-user-tag-spoofing-attack-in-mobile-apps
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2023/policycomp-counterpart-comparison-of-privacy-policies-uncovers-overbroad-persona
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2024/maginot-line-assessing-a-new-cross-app-threat-to-pii-as-factor-authentication-in-chinese-mobile-apps
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2024/on-precisely-detecting-censorship-circumvention-in-real-world-networks
      network-layer work that uses spam as a label source
  WWW/2023/unsupervised-anomaly-detection-on-microservice-traces-through-graph-vae
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2024/a-first-look-at-security-and-privacy-risks-in-the-rapidapi-ecosystem
      not individually annotated: screened at title level, matched no off-topic family rule
  IEEE-SP/2024/combing-for-credentials-active-pattern-extraction-from-smart-reply
      ML-privacy work whose training data happens to be mail
  IEEE-SP/2024/deeptheft-stealing-dnn-model-architectures-through-power-side-channel
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2024/a-mixed-methods-study-on-user-experiences-and-challenges-of-recovery-codes-for-a
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2024/guardians-of-the-galaxy-content-moderation-in-the-interplanetary-file-system
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2024/landscape-exploring-ldap-weaknesses-and-data-leaks-at-internet-scale
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2024/malla-demystifying-real-world-large-language-model-integrated-malicious-services
      LLM-for-hire services, one output of which is spam; the unit is the service
  USENIX/2024/pixel-thief-exploiting-svg-filter-leakage-in-firefox-and-chrome
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2024/simurai-slicing-through-the-complexity-of-sim-card-security-research
      systems security; the message vocabulary is incidental
  IEEE-SP/2012/dissecting-android-malware-characterization-and-evolution
      systems security; the message vocabulary is incidental
  IEEE-SP/2016/the-cracked-cookie-jar-http-cookie-hijacking-and-the-exposure-of-private-informa
      not individually annotated: screened at title level, matched no off-topic family rule
  CCS/2025/noise-and-stress-dont-help-with-learning-a-qualitative-study-to-inform-design-of
      not individually annotated: screened at title level, matched no off-topic family rule
  IMC/2025/congestion-patterns-in-a-large-scale-rdma-datacenter
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2025/all-your-database-are-belong-to-us-characterizing-database-ransomware-attacks
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2025/whats-done-is-not-whats-claimed-detecting-and-interpreting-inconsistencies-in-app-behaviors
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2025/who-is-trying-to-access-my-account-exploring-user-perceptions-and-reactions-to-risk-based-authentication-notifications
      same; its 7 `unsubscri` hits are about turning off security alerts
  NDSS/2025/powerradio-manipulate-sensor-measurement-via-power-gnd-radiation
      systems security; the message vocabulary is incidental
  NDSS/2025/was-this-you-investigating-the-design-considerations-for-suspicious-login-notifications
      the notification's wording is the treatment; the channel is incidental
  NDSS/2025/attributing-open-source-contributions-is-critical-but-difficult-a-systematic-analysis-of-github-practices-and-their-impact-on-software-supply-chain-security
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2025/evaluating-llm-based-personal-information-extraction-and-countermeasures
      ML-privacy work whose training data happens to be mail
  USENIX/2025/scanned-and-scammed-insecurity-by-obsqrity-measuring-user-susceptibility-and-awa
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2025/private-investigator-extracting-personally-identifiable-information-from-large-l
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2025/generated-data-with-fake-privacy-hidden-dangers-of-fine-tuning-large-language-mo
      ML-privacy work whose training data happens to be mail
  USENIX/2025/mbfuzzer-a-multi-party-protocol-fuzzer-for-mqtt-brokers
      pub/sub or SDN "unsubscribe" — protocol verb, not marketing mail
  USENIX/2025/when-llms-go-online-the-emerging-threat-of-web-enabled-llms
      ML-privacy work whose training data happens to be mail
  USENIX/2025/effective-pii-extraction-from-llms-through-augmented-few-shot-learning
      ML-privacy work whose training data happens to be mail
  NDSS/2026/tipso-gan-malicious-network-traffic-detection-using-a-novel-optimized-generative-adversarial-network
      not individually annotated: screened at title level, matched no off-topic family rule
  PETS/2026/because-i-didnt-touch-these-and-even-dont-know-why-i-should-to-change-these-why
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2026/trojpix-electromagnetic-covert-channels-via-imperceptible-pixel-modulation
      not individually annotated: screened at title level, matched no off-topic family rule
  WWW/2026/truth-with-a-twist-the-rhetoric-of-persuasion-in-professional-vs-community-autho
      not individually annotated: screened at title level, matched no off-topic family rule
  IEEE-SP/2015/vetting-ssl-usage-in-applications-with-sslint
      not individually annotated: screened at title level, matched no off-topic family rule
  IEEE-SP/2016/triggerscope-towards-detecting-logic-bombs-in-android-applications
      systems security; the message vocabulary is incidental
  IEEE-SP/2018/the-spyware-used-in-intimate-partner-violence
      not individually annotated: screened at title level, matched no off-topic family rule
  IEEE-SP/2017/the-password-reset-mitm-attack
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2025/a-large-scale-measurement-study-of-the-proxy-protocol-and-its-security-implications
      network-layer work that uses spam as a label source
  NDSS/2025/starshields-for-ios-navigating-the-security-cosmos-in-satellite-communication
      systems security; the message vocabulary is incidental
  PETS/2025/gig-work-at-what-cost-exploring-privacy-risks-of-gig-work-platform-participation
      not individually annotated: screened at title level, matched no off-topic family rule
  PETS/2025/real-world-deniability-in-messaging
      not individually annotated: screened at title level, matched no off-topic family rule
  USENIX/2025/from-alarms-to-real-bugs-multi-target-multi-step-directed-greybox-fuzzing-for-st
      systems security; the message vocabulary is incidental
  PETS/2025/sok-web-authentication-and-recovery-in-the-age-of-end-to-end-encryption
      not individually annotated: screened at title level, matched no off-topic family rule
  PETS/2026/chatbot-confessions-large-scale-analysis-of-private-data-disclosure-in-shared-ai
      ML-privacy work whose training data happens to be mail
  USENIX/2026/opossum-attack-application-layer-desynchronization-using-opportunistic-tls
      network-layer work that uses spam as a label source
  IEEE-SP/2010/inspector-gadget-automated-extraction-of-proprietary-gadgets-from-malware-binari
      systems security; the message vocabulary is incidental
  IEEE-SP/2017/under-the-shadow-of-sunshine-understanding-and-detecting-bulletproof-hosting-on
      not individually annotated: screened at title level, matched no off-topic family rule
  NDSS/2025/isolategpt-an-execution-isolation-architecture-for-llm-based-agentic-systems
      ML-privacy work whose training data happens to be mail
  IEEE-SP/2022/desperate-times-call-for-desperate-measures-user-concerns-with-mobile-loan-apps
      not individually annotated: screened at title level, matched no off-topic family rule
  IEEE-SP/2022/device-fingerprinting-with-peripheral-timestamps
      not individually annotated: screened at title level, matched no off-topic family rule
  IEEE-SP/2022/towards-automated-auditing-for-account-and-session-management-flaws-in-single-si
      not individually annotated: screened at title level, matched no off-topic family rule
  IEEE-SP/2024/dnsbomb-a-new-practical-and-powerful-pulsing-dos-attack-exploiting-dns-queries-a
      network-layer work that uses spam as a label source
  IEEE-SP/2024/tcp-spoofing-reliable-payload-transmission-past-the-spoofed-tcp-handshake
      network-layer work that uses spam as a label source
  IEEE-SP/2024/understanding-the-privacy-practices-of-political-campaigns-a-perspective-from-th
      measures what campaign websites collect, including addresses; no mail was received or measured
  IEEE-SP/2023/d-dae-defense-penetrating-model-extraction-attacks
      not individually annotated: screened at title level, matched no off-topic family rule
  IEEE-SP/2022/scraping-sticky-leftovers-app-user-information-left-on-servers-after-account-del
      not individually annotated: screened at title level, matched no off-topic family rule
  IEEE-SP/2023/rulekeeper-gdpr-aware-personal-data-compliance-for-web-frameworks
      not individually annotated: screened at title level, matched no off-topic family rule
  IEEE-SP/2024/erasan-efficient-rust-address-sanitizer
      not individually annotated: screened at title level, matched no off-topic family rule
 
--- Rejected outside the pool — 9
  what-wechat-knows-pervasive-first-party-tracking-in-a-billion-user-super-app-eco
      first-party tracking inside a super-app — design:platforms, not the message channel
  measuring-the-accuracy-and-effectiveness-of-pii-removal-services
      people-search removal services; the brokerage side without the message side
  it-feels-like-whack-a-mole-user-experiences-of-data-removal-from-people-search-w
      same, as a user study
  demystifying-the-messaging-platforms-ecosystem-through-the-lens-of-twitter
      which messaging platforms people discuss on Twitter, not what the platforms do
  measuring-and-mitigating-oauth-access-token-abuse-by-collusion-networks
      OAuth token abuse; the "spam trap" is a Facebook one
  following-their-footsteps-characterizing-account-automation-abuse-and-defenses
      platform account automation — SOC-adjacent but the unit is the account
  two-steps-forward-and-one-step-back-the-right-to-opt-out-of-sale-under-cpra
      CCPA/CPRA "do not sell", the third homograph of opt-out
  do-opt-outs-really-opt-me-out
      advertising opt-out cookies, not mail
  finding-a-choice-in-a-haystack-automatic-extraction-of-opt-out-statements-from-p
      opt-out statements extracted from privacy policies — text, not mail
 
--- Probes measured and rejected
  classification.target == 'user-generated-text'
      419 papers — fires on 419 papers, nearly all of them social-media text; no email-vs-platform distinction
  detection[].phenomenon =~ /spam/i
      (not counted) — free text, ~20% run-to-run stable; and the phenomenon string does not say which channel
  full text =~ /\bunsubscrib/i
      56 papers — MQTT / SDN protocol verb in 10 of 56; CCPA "opt out of sale" or an advertising opt-out cookie in 3; turning off a security notification in 4; the remaining 39 have one passing mention each. Zero papers whose OBJECT is an unsubscription mechanism
  full text =~ /\bspam/i
      954 papers — a fact about these being security venues, not a population
  full text =~ /\be-?mail/i
      2502 papers — every paper that mentions contacting an author or recruiting by mail
  tools[].name =~ /mail|smtp|imap/i
      (not counted) — returns mail libraries used as plumbing by papers about something else
 
--- Recall probes: signals NOT used to build the pool, run against it afterwards
A pool built from titles and full-text density can miss a paper whose
topic only shows up in a schema field. These probes look for that.
 
  classification.target == 'email-message'
    54 papers; 12 already on the page; 30 in the pool but screened out; 12 OUTSIDE the pool
      outside: WWW/2013/traveling-the-silk-road-a-measurement-analysis-of-a-large-anonymous-online-marke
      outside: NDSS/2013/behind-the-scenes-of-online-attacks-an-analysis-of-exploitation-behaviors-on-the
      outside: IMC/2014/the-matter-of-heartbleed
      outside: USENIX/2019/the-betrayal-at-cloud-city-an-empirical-analysis-of-cloud-based-mobile-backends
      outside: WWW/2020/dynamic-composition-for-conversational-domain-exploration
      outside: CCS/2023/unsafe-diffusion-on-the-generation-of-unsafe-images-and-hateful-memes-from-text
      outside: WWW/2023/migration-reframed-a-multilingual-analysis-on-the-stance-shift-in-europe-during
      outside: USENIX/2024/a-decade-of-privacy-relevant-android-app-reviews-large-scale-trends
      outside: PETS/2026/the-role-of-online-forums-in-developer-understanding-of-privacy-law-a-reddit-cas
      outside: WWW/2026/doxing-as-a-service-demystifying-the-chinese-online-doxing-ecosystem
      outside: NDSS/2025/misdirection-of-trust-demystifying-the-abuse-of-dedicated-url-shortening-service
      outside: IEEE-SP/2022/the-state-of-the-samesite-studying-the-usage-effectiveness-and-adequacy-of-sames
 
  classification.targetDetail matches /sms|text message|robocall|phone call/i
    26 papers; 9 already on the page; 4 in the pool but screened out; 13 OUTSIDE the pool
      outside: USENIX/2011/why-special-agent-johnny-still-cant-encrypt-a-security-analysis-of-the-apco-proj
      outside: PETS/2018/power-to-peep-all-inference-attacks-by-malicious-batteries-on-mobile-devices
      outside: WWW/2021/towards-understanding-and-demystifying-bitcoin-mixing-services
      outside: PETS/2022/analyzing-the-monetization-ecosystem-of-stalkerware
      outside: NDSS/2023/navigating-murky-waters-automated-browser-feature-testing-for-uncovering-tracking-vectors
      outside: USENIX/2024/hermes-unlocking-security-analysis-of-cellular-network-protocols-by-synthesizing
      outside: USENIX/2024/dva-extracting-victims-and-abuse-vectors-from-android-accessibility-malware
      outside: USENIX/2024/defects-in-depth-analyzing-the-integration-of-effective-defenses-against-one-day
      outside: CCS/2025/lock-the-door-but-keep-the-window-open-extracting-app-protected-accessibility-in
      outside: CCS/2025/we-just-did-not-have-that-on-the-embedded-system-insights-and-challenges-for-sec
      outside: USENIX/2025/choicejacking-compromising-mobile-devices-through-malicious-chargers-like-a-deca
      outside: NDSS/2025/the-forking-way-when-tees-meet-consensus
      outside: IEEE-SP/2026/on-the-in-security-of-loading-machine-learning-models
 
  population.unit matches /e-?mail|message|sms|mailbox|phone number/i
    53 papers; 13 already on the page; 25 in the pool but screened out; 15 OUTSIDE the pool
      outside: NDSS/2012/host-fingerprinting-and-tracking-on-the-web-privacy-and-security-implications
      outside: WWW/2013/the-cost-of-annoying-ads
      outside: USENIX/2019/the-secret-sharer-evaluating-and-testing-unintended-memorization-in-neural-netwo
      outside: WWW/2020/ten-social-dimensions-of-conversations-and-relationships
      outside: USENIX/2022/khaleesi-breaker-of-advertising-and-tracking-request-chains
      outside: IEEE-SP/2023/analyzing-leakage-of-personally-identifiable-information-in-language-models
      outside: CCS/2023/skillscanner-detecting-policy-violating-voice-applications-through-static-analys
      outside: CCS/2024/precurious-how-innocent-pre-trained-language-models-turn-into-privacy-traps
      outside: USENIX/2024/exploiting-leakage-in-password-managers-via-injection-attacks
      outside: USENIX/2025/password-guessing-using-large-language-models
      outside: USENIX/2025/privacyxray-detecting-privacy-breaches-in-llms-through-semantic-consistency-and
      outside: NDSS/2025/skillpov-towards-accessible-and-effective-privacy-notice-for-amazon-alexa-skills
      outside: USENIX/2026/overcoming-the-retrieval-barrier-indirect-prompt-injection-in-the-wild-for-llm-s
      outside: IEEE-SP/2016/you-get-where-youre-looking-for-the-impact-of-information-sources-on-code-securi
      outside: NDSS/2026/expshield-safeguarding-web-text-from-unauthorized-crawling-and-llm-exploitation
 
  detection[].phenomenon matches /e-?mail|inbox|sms|robocall|smish|unsubscrib/i
    132 papers; 37 already on the page; 63 in the pool but screened out; 32 OUTSIDE the pool
      outside: NDSS/2012/insights-into-user-behavior-in-dealing-with-internet-attacks
      outside: WWW/2013/the-cost-of-annoying-ads
      outside: CCS/2013/beheading-hydras-performing-effective-botnet-takedowns
      outside: USENIX/2014/an-internet-wide-view-of-internet-wide-scanning
      outside: IMC/2015/temporal-and-spatial-classification-of-active-ipv6-addresses
      outside: IMC/2016/an-analysis-of-the-privacy-and-security-risks-of-android-vpn-permission-enabled
      outside: PETS/2016/access-denied-contrasting-data-access-in-the-united-states-and-ireland
      outside: IMC/2019/tales-from-the-porn-a-comprehensive-privacy-analysis-of-the-web-porn-ecosystem
      outside: NDSS/2019/time-does-not-heal-all-wounds-a-longitudinal-analysis-of-security-mechanism-support-in-mobile-browsers
      outside: USENIX/2020/firmscope-automatic-uncovering-of-privilege-escalation-vulnerabilities-in-pre-in
      outside: NDSS/2021/understanding-worldwide-private-information-collection-on-android
      outside: NDSS/2021/zoom-on-the-keystrokes-exploiting-video-calls-for-keystroke-inference-attacks
      outside: USENIX/2021/assessing-browser-level-defense-against-idn-based-phishing
      outside: WWW/2021/towards-understanding-and-demystifying-bitcoin-mixing-services
      outside: CCS/2022/poster-an-analysis-of-privacy-features-in-expert-approved-kids-apps
      outside: CCS/2022/freely-given-consent-studying-consent-notice-of-third-party-tracking-and-its-vio
      outside: IEEE-SP/2022/are-we-there-yet-timing-and-floating-point-attacks-on-differential-privacy-syste
      outside: PETS/2022/analyzing-the-monetization-ecosystem-of-stalkerware
      outside: IEEE-SP/2023/webspec-towards-machine-checked-analysis-of-browser-security-mechanisms
      outside: IMC/2023/the-prevalence-of-single-sign-on-on-the-web-towards-the-next-generation-of-web-c
      outside: NDSS/2023/navigating-murky-waters-automated-browser-feature-testing-for-uncovering-tracking-vectors
      outside: USENIX/2023/beyond-typosquatting-an-in-depth-look-at-package-confusion
      outside: USENIX/2023/erebus-access-control-for-augmented-reality-systems
      outside: USENIX/2023/hidden-reality-caution-your-hand-gesture-inputs-in-the-immersive-virtual-world-a
      outside: PETS/2024/two-steps-forward-and-one-step-back-the-right-to-opt-out-of-sale-under-cpra
      outside: USENIX/2024/dva-extracting-victims-and-abuse-vectors-from-android-accessibility-malware
      outside: USENIX/2024/tickets-or-privacy-understand-the-ecosystem-of-chinese-ticket-grabbing-apps
      outside: USENIX/2024/i-really-just-leaned-on-my-community-for-support-barriers-challenges-and-coping
      outside: IEEE-SP/2012/a-framework-to-eliminate-backdoors-from-response-computable-authentication
      outside: PETS/2026/exadprinter-semi-exhaustive-permissionless-device-fingerprinting-within-the-andr
      outside: WWW/2026/towards-token-level-text-anomaly-detection
      outside: IEEE-SP/2022/symbexcel-automated-analysis-and-understanding-of-malicious-excel-4-0-macros
 
  outside-the-pool mentions across the four probes: 72
  DISTINCT papers outside the pool that the probes surfaced: 67
 
  Verdict on the 2026-09-02 run. All four probes' outside-the-pool lists were
  read at title level; the twelve from the 'email-message' enum were also
  opened. That enum is noisy in a way its being an enum disguises: it fires on
  The Matter of Heartbleed, The State of the SameSite, Traveling the Silk Road
  and an unsafe-image-generation paper, none of which measures a message.
 
  ONE genuine miss came out of the phenomenon probe and was added:
  Starov et al., PoPETs 2016, "Are You Sure You Want to Contact Us?
  Quantifying the Leakage of PII via Website Contact Forms" — now the oldest
  paper in the EID slice. Two others were added from earlier probes. Read the
  addition as the measured cost of a title-and-density pool: about one paper in
  seventy, in the direction of missing the oldest work in a slice.
 
  Read and rejected from these lists, as the near misses most likely to be
  re-proposed: access-denied (PoPETs 2016, a survey about data-access
  attitudes, no mail), freely-given-consent (CCS 2022 — privacy:consent),
  IDN-based phishing (USENIX 2021 — security:phishing), stalkerware
  monetisation (PoPETs 2022, the object is the payment flow).
 
--- Papers admitted to the pool by hand
  auditing-offline-data-brokers-via-facebooks-advertising-platform
      recall probe /lead gener|lead broker|data broker/ (161 hits)
  exploring-the-security-and-privacy-risks-of-chatbots-in-messaging-services
      recall probe /spam ?trap|honey ?pot (e-?mail|account)|honey ?token/ (8 hits)
  are-you-sure-you-want-to-contact-us-quantifying-the-leakage-of-pii-via-website-c
      recall probe detection[].phenomenon =~ /e-?mail|inbox|.../ — a genuine miss, and the oldest paper in the EID slice: contact forms as the gateway from pseudonym to address, six years before Leaky Forms
 
==============================================================================
Z. EVERY NON-CORPUS NUMBER ON THE PAGE, with its primary source
==============================================================================
Checked by scripts/external_checks_email_tracking.sh on the date printed there.
 
  RFC 2369            List-Help/List-Unsubscribe/... header fields, 1998, Proposed Standard
  RFC 8058            List-Unsubscribe-Post one-click unsubscribe, 2017, Proposed Standard
  RFC 5321 / 5322     SMTP and the Internet Message Format
  RFC 6376            DKIM Signatures
  2024-02-01          the date Google and Yahoo began enforcing bulk-sender requirements,
                      including one-click unsubscribe within 2 days
  5,000               Google's bulk-sender threshold, messages/day to Gmail addresses
  0.3% / 0.10%        Google's stated spam-rate thresholds (avoid / stay under)
  4,322               the size of the PREVIOUS extraction run, on which the field-stability
                      figures (~20% free-text agreement, 58% classification.method) were
                      measured. Source: data/extract/README.md and OVERVIEW.md's own note
                      that the stability table is not recomputed by the overview script.
  12,618              EMAILS collected by Englehardt et al., PoPETs 2018, from 902
                      distinct senders on 15,700 sites crawled. NOT a count of mailing
                      lists — that wording was wrong here and on the page until review
                      on 2026-09-02. All four figures are needles in
                      verify_email_figures.mjs
  56.5%               data/extract/OVERVIEW.md's own artifacts.availability rate, quoted on
                      the page only to say it is NOT comparable with this report's 59.5%
                      (different denominator: all empirical papers vs papers with a
                      non-null artifacts object)
  33.8%               OVERVIEW.md's ethics.reviewOutcome rate, over all 5,118 empirical
                      papers; this report's on-page figure uses papers with a non-null
                      ethics object, and the section above prints both
  Llama-3.1-8B        a model name, not a figure
  0.30 / 0.10         Google's spam-rate thresholds, percent
  November 2025       Google's enforcement escalation to temporary and permanent
                      rejections, support.google.com/a/answer/14229414
  5,000 / 2025-05-05  Microsoft's high-volume-sender threshold and effective date for
                      Outlook.com consumer services (SPF + DKIM + DMARC, no RFC 8058).
                      Verified INDIRECTLY: the announcement (Defender for Office 365
                      blog post 4399730) is served behind a block that refuses this
                      host to curl and to a headless browser
  550 5.7.515         the Outlook.com NDR quoted in Microsoft's own Q&A threads
  Safari 17.0         the release WebKit names for Link Tracking Protection, described
                      there as a PRIVATE BROWSING protection
  arXiv:2604.06759    preprint of the IEEE S&P 2026 lead-marketing paper
  11573454            its IEEE Xplore document id, from resolving the DOI
  arXiv:2508.05276    Agarwal et al., 7726 user reports, 1.35 million reports — a
                      PREPRINT, cited as such, used for no figure
  arXiv:2606.31790    Altwlkany et al., robocalls, 8.7 million records, 65 countries —
                      likewise a preprint

node scripts/verify_email_figures.mjs –show

verify_email_figures-output.txt

  ok   i-never-signed-up-for-this :: 85% of emails ||| 85 % of emails
  ok   i-never-signed-up-for-this :: 29% of emails ||| 29 % of emails
  ok   i-never-signed-up-for-this :: (62%) are intentional
  ok   i-never-signed-up-for-this :: 11% of links ||| 11 % of links
  ok   i-never-signed-up-for-this :: 87% ||| 87 %
  ok   i-never-signed-up-for-this :: 12,618
  ok   i-never-signed-up-for-this :: 902
  ok   i-never-signed-up-for-this :: 15,700
  ok   i-never-signed-up-for-this :: 38%
  ok   i-never-signed-up-for-this :: 32%
  ok   characterizing-pixel-tracking-through-the-lens-of-disposable-email-services :: 24.6% ||| 24.6 %
  ok   characterizing-pixel-tracking-through-the-lens-of-disposable-email-services :: 573,244
  ok   characterizing-pixel-tracking-through-the-lens-of-disposable-email-services :: 50.4% ||| 50.4 %
  ok   characterizing-pixel-tracking-through-the-lens-of-disposable-email-services :: 91.7% ||| 91.7 %
  ok   characterizing-pixel-tracking-through-the-lens-of-disposable-email-services :: 1,399
  ok   characterizing-pixel-tracking-through-the-lens-of-disposable-email-services :: 926
  ok   characterizing-pixel-tracking-through-the-lens-of-disposable-email-services :: 89,329
  ok   characterizing-pixel-tracking-through-the-lens-of-disposable-email-services :: 94.75% ||| 94.75 %
  ok   characterizing-pixel-tracking-through-the-lens-of-disposable-email-services :: 3.63% ||| 3.63 %
  ok   checking-websites-gdpr-consent-compliance-for-marketing-emails :: 17.3% ||| 17.3 %
  ok   checking-websites-gdpr-consent-compliance-for-marketing-emails :: 21.9% ||| 21.9 %
  ok   checking-websites-gdpr-consent-compliance-for-marketing-emails :: 59% ||| 59 %
  ok   checking-websites-gdpr-consent-compliance-for-marketing-emails :: 2.3% ||| 2.3 %
  ok   checking-websites-gdpr-consent-compliance-for-marketing-emails :: 16% ||| 16 %
  ok   a-black-box-privacy-analysis-of-messaging-service-providers-chat-message-process :: 105
  ok   a-black-box-privacy-analysis-of-messaging-service-providers-chat-message-process :: 34% ||| 34 %
  ok   a-black-box-privacy-analysis-of-messaging-service-providers-chat-message-process :: 37% ||| 37 %
  ok   a-black-box-privacy-analysis-of-messaging-service-providers-chat-message-process :: 30 days
  ok   leaky-forms-a-study-of-email-and-password-exfiltration-before-form-submission :: 1,844
  ok   leaky-forms-a-study-of-email-and-password-exfiltration-before-form-submission :: 2,950
  ok   leaky-forms-a-study-of-email-and-password-exfiltration-before-form-submission :: 52 websites ||| 52 sites
  ok   leaky-forms-a-study-of-email-and-password-exfiltration-before-form-submission :: 290 emails
  ok   all-the-numbers-are-us-large-scale-abuse-of-contact-discovery-in-mobile-messengers :: 10% of US ||| 10 % of US ||| 10% of the US
  ok   all-the-numbers-are-us-large-scale-abuse-of-contact-discovery-in-mobile-messengers :: 100% ||| 100 %
  ok   all-the-numbers-are-us-large-scale-abuse-of-contact-discovery-in-mobile-messengers :: 5.0 million ||| 5 million
  ok   hey-there-you-are-using-whatsapp-enumerating-three-billion-accounts-for-security-and-privacy :: 3.5 B ||| 3.5B ||| 3.5 billion
  ok   hey-there-you-are-using-whatsapp-enumerating-three-billion-accounts-for-security-and-privacy :: 81 % ||| 81%
  ok   left-or-right-a-peek-into-the-political-biases-in-email-spam-filtering-algorithm :: 67.6% ||| 67.6 %
  ok   left-or-right-a-peek-into-the-political-biases-in-email-spam-filtering-algorithm :: 8.2% ||| 8.2 %
  ok   left-or-right-a-peek-into-the-political-biases-in-email-spam-filtering-algorithm :: 5.34% ||| 5.34 %
  ok   tasters-choice-a-comparative-analysis-of-spam-feeds :: 60% ||| 60 %
  ok   tasters-choice-a-comparative-analysis-of-spam-feeds :: ten different data feeds
  ok   unfiltered-measuring-cloud-based-email-filtering-bypasses :: 80% ||| 80 %
  ok   unfiltered-measuring-cloud-based-email-filtering-bypasses :: 1,601
  ok   unfiltered-measuring-cloud-based-email-filtering-bypasses :: 673
  ok   unfiltered-measuring-cloud-based-email-filtering-bypasses :: 928
  ok   fishing-for-smishing-understanding-sms-phishing-infrastructure-and-strategies-by :: 27.7𝑘 smishing messages
  ok   fishing-for-smishing-understanding-sms-phishing-infrastructure-and-strategies-by :: 19.3𝑘 sender IDs
  ok   fishing-for-smishing-understanding-sms-phishing-infrastructure-and-strategies-by :: 20𝑘 URLs
  ok   fishing-for-smishing-understanding-sms-phishing-infrastructure-and-strategies-by :: 220, 585 ||| 220,585
  ok   fishing-for-smishing-understanding-sms-phishing-infrastructure-and-strategies-by :: 64, 284 ||| 64,284
  ok   do-spammers-dream-of-electric-sheep-characterizing-the-prevalence-of-llm-generat :: LLM
  ok   characterizing-robocalls-with-multiple-vantage-points :: vantage point
  ok   diving-into-robocall-content-with-snorcall :: weak supervision ||| Snorkel
  ok   inbox-invasion-exploiting-mime-ambiguities-to-evade-email-attachment-detectors :: 75.95% ||| 75.95 %
  ok   inbox-invasion-exploiting-mime-ambiguities-to-evade-email-attachment-detectors :: 180
  ok   inbox-invasion-exploiting-mime-ambiguities-to-evade-email-attachment-detectors :: 237
  ok   privacy-risks-with-facebooks-pii-based-targeting-auditing-a-data-brokers-adverti :: 18 of the volunteers who did visit
  ok   privacy-risks-with-facebooks-pii-based-targeting-auditing-a-data-brokers-adverti :: 20 volunteers visit
  ok   investigating-sources-of-pii-used-in-facebook-s-targeted-advertising :: 22 days
  ok   investigating-sources-of-pii-used-in-facebook-s-targeted-advertising :: 36 days
  ok   cascading-spy-sheets-exploiting-the-complexity-of-modern-css-for-email-and-browser-fingerprinting :: 97.95 % ||| 97.95%
  ok   cascading-spy-sheets-exploiting-the-complexity-of-modern-css-for-email-and-browser-fingerprinting :: 1152 ||| 1,152
  ok   auditing-offline-data-brokers-via-facebooks-advertising-platform :: 90% ||| 90 %
  ok   auditing-offline-data-brokers-via-facebooks-advertising-platform :: 64.3% ||| 64.3 %
  ok   auditing-offline-data-brokers-via-facebooks-advertising-platform :: 13.8% ||| 13.8 %
  ok   hey-mum-i-dropped-my-phone-down-the-toilet-investigating-hi-mum-and-dad-sms-scam :: median lifetime
  ok   poster-a-comprehensive-categorization-of-sms-scams :: 830.2𝑘
  ok   poster-a-comprehensive-categorization-of-sms-scams :: 132.4𝑘
  ok   sending-out-an-sms-characterizing-the-security-of-the-sms-ecosystem-with-public :: 386,327
  ok   sending-out-an-sms-characterizing-the-security-of-the-sms-ecosystem-with-public :: 522
  ok   sending-out-an-sms-characterizing-the-security-of-the-sms-ecosystem-with-public :: 14 months
  ok   sending-out-an-sms-characterizing-the-security-of-the-sms-ecosystem-with-public :: 400 phone numbers ||| over 400 phone numbers
  ok   greystar-fast-and-accurate-detection-of-sms-spam-numbers-in-large-cellular-netwo :: 34K ||| 34k ||| 34,000
  ok   greystar-fast-and-accurate-detection-of-sms-spam-numbers-in-large-cellular-netwo :: 1.2 hours ||| 1.2 h
  ok   greystar-fast-and-accurate-detection-of-sms-spam-numbers-in-large-cellular-netwo :: gray phone space ||| grey phone space ||| Gray Phone Space
  ok   xray-enhancing-the-web-s-transparency-with-differential-correlation :: Gmail
  ok   xray-enhancing-the-web-s-transparency-with-differential-correlation :: 80% ||| 80 %
 
ASCII RENDERINGS of figures the papers typeset with the mathematical-italic k
(U+1D458). The needles above carry the real codepoint; the wiki page quotes them
in ASCII, so they are listed here for scripts/check_page_numbers.mjs:
  27.7k  19.3k  20k   (Agarwal et al., IMC 2025)
  830.2k  132.4k       (Agarwal et al., IMC 2024 poster)
 
check_page_numbers.mjs tokenises "830.2k" as the bare prefix and then refuses to
match it against a "830.2" in this file, so the bare tokens are listed too:
  27 , 19 , 20 , 830 , 132
 
 
24 papers, 78 figures: 78 present, 0 MISSING

bash scripts/external_checks_email_tracking.sh

external_checks_email_tracking-output.txt

== IETF standards ==
PASS   RFC 2369 status: PROPOSED STANDARD | obsoleted_by: [] | updated_by: []
PASS   RFC 8058 status: PROPOSED STANDARD | obsoleted_by: [] | updated_by: []
PASS   RFC 6376 status: INTERNET STANDARD | obsoleted_by: [] | updated_by: ['RFC8301', 'RFC8463', 'RFC8553', 'RFC8616']
PASS   RFC 8058 title
PASS   RFC 2369 defines List-Unsubscribe
 
== mailbox-provider requirements (they set what a 2026 crawl can measure) ==
PASS   Google bulk-sender threshold 5,000/day
PASS   Google one-click unsubscribe cites RFC 8058
PASS   Google 48-hour unsubscribe deadline
PASS   Gmail image proxy announcement, Dec 2013
PASS   Gmail image proxy: no IP, no cookies
PASS   Gmail image proxy does NOT hide the open
PASS   Google Nov-2025 enforcement escalation
PASS   WebKit Link Tracking Protection is a PRIVATE BROWSING feature
PASS   arXiv preprint of the lead-marketing paper
PASS   Yahoo one-click + two days, Q1 2024
 
== client-side defences (BROWSER REQUIRED — expect FAILED under curl) ==
PASS   Apple Mail Privacy Protection hides IP and the open
 
== tools the page names ==
PASS   OpenWPM repository alive
PASS   Email Privacy Tester alive
 
ALL CHECKS PASSED

node scripts/report_email_tracking.mjs –list prints all 70 on-page papers with their verdict and its reason. It is not reproduced here because the same information, per paper, is in the residue and verdict sections of the report above.

What the number guard actually caught, in order. This is the part worth recording, because every one of these would have shipped:

1. Four base-rate columns in the silence table came from data/extract/OVERVIEW.md, not from this page's own script. The values were current, but a page whose figures its own script cannot reproduce cannot be refreshed. The report now computes all eight base rates itself.
2. And computing them exposed a denominator mismatch. artifacts.availability is 56.5% in the overview and 59.5% here, because the overview divides by all 5,118 empirical papers and this report divides by the 4,854 with a non-null artifacts object. The on-page column (36/67) uses the second rule, so pairing it with 56.5% would have compared two different things. The page now prints 59.5% and says explicitly why it differs from corpus.
3. The same mismatch, again, in the ethics section. The draft compared the slice's 54.5% against the overview's 33.8%. On the comparable denominator the corpus rate is 38.6%, so the draft overstated how much better this slice is by five points. Fixed, with the other figure in a footnote.
4. The whole-page run found seven more figures outside the audited window — in the ethics section, the address-as-identifier section and the SMS section — that no script produced. Five are now checked by verify_email_figures.mjs (13.8%, 64.3%, 830.2k, 926, 89,329, 94.75%, 3.63%) and two are external and listed in the report's own non-corpus block.
5. 830.2k broke the guard for a mechanical reason worth knowing. The guard tokenises “830.2k” as the bare prefix 830 and then refuses to match it against a 830.2 in the report, because its lookahead rejects a number followed by more digits. The verifier output now lists the bare tokens as well.
6. A loose “12,000 newsletters” in the residue commentary was replaced by the real 12,618 with its citation.

The guard is not useful on this provenance page and was not made to pass on it. Run whole-page against it and it flags eleven figures: three section numbers (2.1, 3.2, 9.1), five URL path components (81126, 145919, 15263077, 102320, the Yahoo post id), the character count in §8's own description, the marker count in the run log, and the 880 quoted from security's own page. A working log is made of identifiers; forcing them into a shared ALLOW map would make the map useless for the pages the guard is for.

Three focused reviewers ran in parallel against the published text, the scripts and their committed output; the generic pass ran after their findings were applied. Every finding is listed with whether it was accepted or rejected, because the rejections are the only record of whether a reviewer earns its slot.

The same reviewer independently re-verified the whole run of per-paper figures, the external quotes, and the author lists of the new PETS/USENIX/NDSS entries against each paper's first page, and found no other misattribution.

Two things were found before the review pass, by the guards rather than by reading, and both are recorded in section 13: four silence base rates copied out of OVERVIEW.md instead of computed, and the denominator mismatch that copying them concealed. Two more were found by the author while checking a claim the review would probably have accepted: “on by default since iOS 15” for Mail Privacy Protection, which Apple documents as a setting the user turns on and for which Apple publishes no take-up figure, and “Gmail's image proxy has been on since 2013”, which was recall until the December 2013 announcement was fetched. The provenance page's own assertion that eight below-threshold quotes had been read by hand was written before they had been; reading them is what produced the table in section 8, and checking the sentence after it is what added two papers to the verifier.

Four more things were changed after the review, found by reading the page top to bottom rather than by any check:

• The intro claimed “two of the four things you might want to measure have essentially never been measured”. One slice is zero; the other three all have papers. Rewritten to say what is true: one slice is a measured zero, and a second question — what the mailbox providers' own defences do to a tracking measurement — is unmeasured.
• “Four neighbouring clusters are larger than anything on this page” was false of one of the four: the web-pixel cluster is 7 papers against the SMS slice's 23. Now “three of them are larger than any slice that is”.
• “This topic has an unusual concentration of ethics problems, and the corpus reflects it” used a legal-assessment rate as evidence for an ethics claim. The rate (20.0% against 6.9%) is real and stays; the inference does not, and the page now says the number is about legal engagement and that the reporting figures at the foot of the same list say the ethics are not handled well.
• “Three of the recent smishing and SMS-spam papers draw on partly overlapping public sources” was not supported by anything. Two of them mine Twitter ([16Tang, Siyuan; Mi, Xianghang; Li, Ying; Wang, XiaoFeng; Chen, Kai (2022): "Clues in Tweets: Twitter-Guided Discovery and Analysis of SMS Spam", in: Proceedings of the ACM SIGSAC Conference on Computer and Communications Security. (DOI)], [12Agarwal, Sharad; Papasavva, Antonis; Suarez-Tangil, Guillermo; Vasek, Marie (2025): "Fishing for Smishing: Understanding SMS Phishing Infrastructure and Strategies by Mining Public User Reports", in: Proceedings of the ACM Internet Measurement Conference. (DOI)]); one uses a crowdsourced app and one carrier-side reports. Replaced with the two that actually overlap and the specific comparison nobody has run.

The pattern in all four is the same and worth naming: the number was right and the sentence around it claimed more than the number supports. Neither the number guard nor a figures reviewer can see that, because they check digits against a script. It is the failure mode the generic review slot exists for, and half of it was still found by re-reading rather than by a reviewer.

The generic reviewer's verdict was that the pages are well guarded on their headline figures and that the provenance page repeatedly failed to practise what it preaches about itself. That is correct, and the underlying cause is one this project has a memory note about: a fix survives in the provenance log. Every accepted review fix had been applied to the content page and almost none had been propagated into the script constants and the prose template that the provenance page republishes.

All eight claimed fixes re-derived independently and CONFIRMED: the five Englehardt figures against the paper; the study-shape figures recomputed from buildPool() (70 / 29 / 27 / 33 / 54 / 12 / 42 and 14-7-5-16-22-22 of 27); both untruncated tables; the verifier count; the non-zero exit; the browsers array fix at 16 of 27 (and a sweep confirming languages is the only other array field, and it is not in that table); the bibliography with the QA notes gone and 816 entries intact; and the U+FF20 title rendering with every marker resolving.

It also found two things nobody else did:

• “73 papers outside the pool” was wrong. The four probes produce 72 outside-the-pool mentions and 67 distinct papers, because five papers appear on more than one probe's list. Neither figure is 73. Worse, check_page_numbers.mjs passed it, because its shared ALLOW map contains a 73 written for design:platforms — a whitelist entry for one page silently blessing a wrong number on another. Fixed on the page (67 distinct, 72 mentions), and the report now prints both counts so neither has to be derived by hand. The shared ALLOW map is a cross-page liability and this is the first measured instance of it hiding a real error.
• The “all five citation markers” claim in 14.2 above. There is one. Corrected in place.

scripts/check_attributions.mjs already existed, and its own header said “a page that attributes in prose is not covered; report 0 checked rather than assume pass”. This page attributes entirely in prose, so it reported 0 checked — and a fabricated author name shipped anyway: the page said “Sharevski and Zettlemoyer” where the authors are Sharevski, Loop, Evans and Ponticello. No other guard could see it. The citekey resolved, the figure beside it was verified against the paper, and the number guard traced the digits.

A –prose mode was added: it matches Surname, Surname et al., Surname and Surname and Surname, Surname and Surname immediately before a marker, and requires every capitalised token to appear somewhere in that entry's full author string. Two iterations were needed and both are worth recording, because a guard that is wrong in either direction does not get run twice:

1. A first version scanned every capitalised token in a 90-character window and reported eight false positives on a clean page (XRay, SMTP-dialect, DNSBL, GDPR, URL, “Older”, “Since”).
2. Tightening it to the four name shapes brought it to 52 attributions checked and one suspect — “Kubicek et al.” against Kub{\'i}{\v{c}}ek, because the skeletoniser folded \v{c} to “vc”. A LaTeX-accent stripping step fixed that. 52 checked, 0 suspect.

Run it in both modes on any page it is pointed at. Table mode reports 0 on this page and says so loudly, which is the behaviour that let the error through in the first place.

Nothing on this page was carried over from an earlier run. The three bench-* dossiers and every published figure predating 2026-08-11 were treated as stale by construction and not consulted for numbers.

[1]

Kirchner, Robin; Koch, Simon; Kamangar, Noah; Klein, David; Johns, Martin (2024): "A Black-Box Privacy Analysis of Messaging Service Providers' Chat Message Processing", in: Proceedings on Privacy Enhancing Technologies. (DOI)

[2]

Englehardt, Steven; Han, Jeffrey; Narayanan, Arvind (2018): "I never signed up for this! Privacy implications of email tracking", Proceedings on Privacy Enhancing Technologies 2018(1):109-126. (DOI)

[3]

Chand, Anish; Nikiforakis, Nick; Vadrevu, Phani (2025): "Doubly Dangerous: Evading Phishing Reporting Systems by Leveraging Email Tracking Techniques", in: Proceedings of the USENIX Security Symposium. (Link)

[4]

Hagen, Christoph; Weinert, Christian; Sendner, Christoph; Dmitrienko, Alexandra; Schneider, Thomas (2021): "All the Numbers are US: Large-scale Abuse of Contact Discovery in Mobile Messengers", in: Proceedings of the Network and Distributed System Security Symposium. (Link)

[5]

Gegenhuber, Gabriel K.; Frenzel, Philipp E.; Günther, Maximilian; Ullrich, Johanna; Judmayer, Aljosha (2026): "Hey there! You are using WhatsApp: Enumerating Three Billion Accounts for Security and Privacy", in: Proceedings of the Network and Distributed System Security Symposium. (Link)

[6]

Kang, Junkyu; Lee, Soyoung; Kwon, Yonghwi; Son, Sooel (2026): "Connecting the Dots: An Investigative Study on Linking Private User Data Across Messaging Apps", in: Proceedings of the Network and Distributed System Security Symposium. (Link)

[7]

Niksirat, Kavous Salehzadeh; Velykoivanenko, Lev; Mätzler, Samuel; Mulders, Stephan; Tamò-Larrieux, Aurelia; Boldi, Marc-Olivier; Humbert, Mathias; Huguenin, Kévin (2025): "Addressing the Address Books' (Interdependent) Privacy Issues", in: Proceedings of the USENIX Security Symposium. (Link)

[8]

Venkatadri, Giridhari; Sapiezynski, Piotr; Redmiles, Elissa M.; Mislove, Alan; Goga, Oana; Mazurek, Michelle L.; Gummadi, Krishna P. (2019): "Auditing Offline Data Brokers via Facebook's Advertising Platform", in: Proceedings of the ACM Web Conference. (DOI)

[9]

Edu, Jide S.; Mulligan, Cliona; Pierazzi, Fabio; Polakis, Jason; Suarez-Tangil, Guillermo; Such, Jose M. (2022): "Exploring the security and privacy risks of chatbots in messaging services", in: Proceedings of the ACM Internet Measurement Conference. (DOI)

[10]

Starov, Oleksii; Gill, Phillipa; Nikiforakis, Nick (2016): "Are You Sure You Want to Contact Us? Quantifying the Leakage of PII via Website Contact Forms", in: Proceedings on Privacy Enhancing Technologies. (DOI)

[11]

Senol, Asuman; Acar, Gunes; Humbert, Mathias; Zuiderveen Borgesius, Frederik (2022): "Leaky Forms: A Study of Email and Password Exfiltration Before Form Submission", in: 31st USENIX Security Symposium (USENIX Security 22). (Link)

[12]

Agarwal, Sharad; Papasavva, Antonis; Suarez-Tangil, Guillermo; Vasek, Marie (2025): "Fishing for Smishing: Understanding SMS Phishing Infrastructure and Strategies by Mining Public User Reports", in: Proceedings of the ACM Internet Measurement Conference. (DOI)

[13]

Venkatadri, Giridhari; Andreou, Athanasios; Liu, Yabing; Mislove, Alan; Gummadi, Krishna P.; Loiseau, Patrick; Goga, Oana (2018): "Privacy Risks with Facebook's PII-Based Targeting: Auditing a Data Broker's Advertising Interface", in: Proceedings of the IEEE Symposium on Security and Privacy. (DOI)

[14]

Pitsillidis, Andreas; Kanich, Chris; Voelker, Geoffrey M.; Levchenko, Kirill; Savage, Stefan (2012): "Taster's choice: a comparative analysis of spam feeds", in: Proceedings of the ACM Internet Measurement Conference. (DOI)

[15]

Acar, Gunes; Englehardt, Steven; Narayanan, Arvind (2020): "No boundaries: data exfiltration by third parties embedded on web pages", Proceedings on Privacy Enhancing Technologies 2020(4):220-238. (DOI)

[16]

Tang, Siyuan; Mi, Xianghang; Li, Ying; Wang, XiaoFeng; Chen, Kai (2022): "Clues in Tweets: Twitter-Guided Discovery and Analysis of SMS Spam", in: Proceedings of the ACM SIGSAC Conference on Computer and Communications Security. (DOI)

[17]

Prasad, Sathvik; Dunlap, Trevor; Ross, Alexander; Reaves, Bradley (2023): "Diving into Robocall Content with SnorCall", in: Proceedings of the USENIX Security Symposium. (Link)

[18]

Li, Ruixuan; Lu, Chaoyi; Liu, Baojun; Zhang, Yunyi; Hong, Geng; Duan, Haixin; Lin, Yanzhong; Pan, Qingfeng; Yang, Min; Shao, Jun (2025): "HADES Attack: Understanding and Evaluating Manipulation Risks of Email Blocklists", in: Proceedings of the Network and Distributed System Security Symposium. (Link)

[19]

Tu, Guan-Hua; Li, Chi-Yu; Peng, Chunyi; Li, Yuanjie; Lu, Songwu (2016): "New Security Threats Caused by IMS-based SMS Service in 4G LTE Networks", in: Proceedings of the ACM SIGSAC Conference on Computer and Communications Security. (DOI)

[20]

Utz, Christine; Michels, Matthias; Degeling, Martin; Marnau, Ninja; Stock, Ben (2023): "Comparing Large-Scale Privacy and Security Notifications", in: Proceedings on Privacy Enhancing Technologies. (DOI)

[21]

Reaves, Bradley; Scaife, Nolen; Tian, Dave; Blue, Logan; Traynor, Patrick; Butler, Kevin R. B. (2016): "Sending Out an SMS: Characterizing the Security of the SMS Ecosystem with Public Gateways", in: Proceedings of the IEEE Symposium on Security and Privacy. (DOI)

[22]

Jiang, Nan; Jin, Yu; Skudlark, Ann; Zhang, Zhi-Li (2013): "Greystar: Fast and Accurate Detection of SMS Spam Numbers in Large Cellular Networks Using Gray Phone Space", in: Proceedings of the USENIX Security Symposium. (Link)

[23]

Murynets, Ilona; Jover, Roger Piqueras (2012): "Crime scene investigation: SMS spam data analysis", in: Proceedings of the ACM Internet Measurement Conference. (DOI)

[24]

Agarwal, Sharad; Harvey, Emma; Vasek, Marie (2024): "Poster: A Comprehensive Categorization of SMS Scams", in: Proceedings of the ACM Internet Measurement Conference. (DOI)

[25]

Stringhini, Gianluca; Egele, Manuel; Zarras, Apostolis; Holz, Thorsten; Kruegel, Christopher; Vigna, Giovanni (2012): "B＠bel: Leveraging Email Delivery for Spam Mitigation", in: Proceedings of the USENIX Security Symposium. (Link)