Working log for blocking_and_geodifference. Every figure on that page, the query that produced it, the population it is a share of, the folds and their residue, the quotes that were checked against the source PDFs, the external sources that were fetched and the ones that were rejected, and the judgement calls — including the ones a reasonable person would have made differently.

Corpus-level caveats — venue scope, the selection funnel, the provisional 2025–2026 years, extraction stability — are on Corpus and are not restated here.

Run: 2026-09-10. Corpus at the time: data/extract/run1/extractions.jsonl, 5,859 extracted papers, 7 venues (CCS, IMC, NDSS, PoPETs, USENIX Security, TheWebConf, IEEE S&P), 2010–2026. Authored by Claude (Opus 5) with five review passes (see §12). No part of this page was carried over from an earlier run: this page did not exist before, and the queued estimate that scoped it (74 candidate papers) was re-derived rather than reused — see §3.

The extraction has no field meaning “this paper measures blocking”. It has:

So the population is a hand-audited candidate set, and the audit is the artefact. §2 is the rule, §3 the probes, §4 the verdicts.

Written down before any count was published, and reproduced verbatim on the content page:

A paper is IN the population if it measures whether some client could reach some
content or service AND attributes the failures to a deliberate blocking decision
taken by someone other than the client — a state, an ISP, a resolver operator, a
CDN, a platform, or the content owner — where that decision is keyed on WHO OR
WHERE the client is, on the content's acceptability to an authority or platform,
or on the traffic looking like an attempt to evade such a decision.

Explicitly OUT, each with its own off-topic family:
  * blocking keyed on the content being malicious (malware, phishing, spam);
  * blocking keyed on the client looking automated (programming:crawler_detection);
  * blocking the client chose (its own ad blocker or filter list);
  * a system PROPOSED to evade blocking, and a DETECTOR proposed for finding
    circumvention traffic (both counted as the ADJACENT family "circumvention");
  * the effect of a takedown or of moderating user posts on later behaviour.

The third key was added after a review pass: this literature also measures
blocking keyed on THE TRAFFIC LOOKING LIKE AN ATTEMPT TO EVADE a blocking
decision — Shadowsocks, fully encrypted flows, an SNI. Measuring how a DEPLOYED
censor detects and blocks such traffic is in the population; proposing a
detector for it is not. The two sides are one sentence apart and the first
version of this rule did not separate them, so two papers sat in the population
under a rule that appeared to exclude them.

This is the version the script prints and the content page reproduces, and it is the second version. The first said only “who or where the client is, or the content's acceptability”, which did not cover blocking keyed on what the traffic looks like — and the population contained four such papers all along: the GFW's active probing of hidden circumvention servers (IMC 2015), its blocking of Shadowsocks (IMC 2020), its blocking of fully encrypted traffic [1Wu, Mingshi; Sippe, Jackson; Sivakumar, Danesh; Burg, Jack; Anderson, Peter; Wang, Xiaokang; Bock, Kevin; Houmansadr, Amir; Levin, Dave; Wustrow, Eric (2023): "How the Great Firewall of China Detects and Blocks Fully Encrypted Traffic", in: Proceedings of the USENIX Security Symposium. (Link)], and SNI-based QUIC censorship (USENIX Security 2025). The generic review pass found the mismatch. The choice made here was to amend the rule to the population rather than to cut the papers, because protocol-shaped blocking is a real and growing part of what censors do, and a page that excluded it would be describing a smaller phenomenon than the one it is named after. The alternative — moving the four to circumvention and publishing a population of 59 — is a defensible reading, and the effect of it can be read off §4.

One boundary case is deliberately in: liu2024_implementation, the protective-DNS study, whose motive is security rather than geography, because its claim has this page's shape — control resolver against treatment resolver, and a definition of “blocked” that has to survive an NXDOMAIN and a sinkhole. It is the whole filter family, so the effect of that decision on any figure can be checked by subtracting one.

Each probe is a candidate generator, never a population. Counts are papers.

Union: 275 candidates, 4.7% of the corpus. Probes 6, 7 and 8 exist because probes 1–5 are built around the words censor and geoblock, and each was added after a later check found a paper the candidate set could not see:

• Probe 6 came from the jurisdiction section: four wall-shaped patterns caught 7 papers no censorship-shaped probe had.
• Probe 7 came from asking what a paper would call this if it never used the word censorship. Its yield is the honest measure of the hole: 2 in-population papers ([2Lu, Chaoyi; Liu, Baojun; Li, Zhou; Hao, Shuang; Duan, Hai-Xin; Zhang, Mingming; Leng, Chunying; Liu, Ying; Zhang, Zaifeng; Wu, Jianping (2019): "An End-to-End, Large-Scale Measurement of DNS-over-Encryption: How Far Have We Come?", in: Proceedings of the ACM Internet Measurement Conference. (DOI)], the 2019 DNS-over-encryption reachability measurement, and [3Sokoto, Saidu; Balduf, Leonhard; Trautwein, Dennis; Wei, Yiluo; Tyson, Gareth; Castro, Ignacio; Ascigil, Onur; Pavlou, George; Korczyński, Maciej; Scheuermann, Björn; Król, Michał (2024): "Guardians of the Galaxy: Content Moderation in the InterPlanetary File System", in: Proceedings of the USENIX Security Symposium. (Link)], IPFS moderation) against 44 off-topic hits.
• Probe 9 is not a probe at all. The generic review pass named 2022/USENIX/openvpn-is-open-to-vpn-fingerprinting as having the same shape as two papers in the population. Its title and summary contain no blocking vocabulary, so no regex could ever reach it; the script now has a hand-added candidate list, with the reason recorded per paper, because a candidate set has to be able to admit one rather than pretend the regexes are complete. Its verdict is circumvention (it proposes the fingerprinting method rather than measuring a deployed censor).
• Probe 8 came from reading the shared bibliography's newest entry. holzbauer2025_tracking, Tracking Internet Disruptions in Ukraine (IMC 2025), was in literature:bibliography already, is squarely about whether clients could reach the Internet, and no probe caught it — while a draft of the page asserted that shutdowns as an event class were absent from the corpus. They are not: there are 12 outage-detection papers, and what they lack is not measurement but attribution. The distinction became an adjacent family (§4) and the page's claim was rewritten around it. This is the finding to remember: a candidate set built from one vocabulary cannot see a literature that uses another, and the counter-example was already on the wiki.

The queued estimate was wrong in both directions and is not comparable. roadmap queued this page on a title-and-summary probe that matched 74 papers, 38 of them on the web platform, 30 in 2020–2023 and 22 in 2024–2026. That probe is not in scripts/gap_probe_roadmap.mjs — it came from the 2026-09-02 brainstorm pass and only its alternation is recorded in prose — so it cannot be re-run as such. Running the recorded alternation (censorship|censored|OONI|Censored Planet|GFW|geoblock|blockpage|internet shutdown) over the current corpus gives 68 / 38 / 27 / 21. The web column matches exactly and the other three do not, so either the corpus moved under it or the recorded alternation is not quite what was run; this run cannot distinguish those, and does not claim to. And 68 is not a floor for this page's population either. Of those 68 hits, 38 are in the population and 30 are not — circumvention systems, homonyms and incidental mentions. In the other direction, 16 of the population's 63 papers are caught by no title probe at all, and probe 1 (a wider title probe than the queued one) excludes 36 of its own 83 hits, 43.4%. A candidate count justifies looking, and nothing more.

Every candidate carries a hand verdict in the VERDICT map inside scripts/report_blocking_geodifference.mjs (§16). The script throws if a candidate has no verdict, if a verdict names a paper no probe caught, or if a family has no label — both directions, because a one-directional check rots silently the next time a probe moves.

Three papers carry two measurement families: [4Ramesh, Reethika; Raman, Ram Sundara; Virkud, Apurva; Dirksen, Alexandra; Huremagic, Armin; Fifield, David; Rodenburg, Dirk; Hynes, Rod; Madory, Doug; Ensafi, Roya (2023): "Network Responses to Russia's Invasion of Ukraine in 2022: A Cautionary Tale for Internet Freedom", in: Proceedings of the USENIX Security Symposium. (Link)] (net+server), [5Knockel, Jeffrey; Dałek, Jakub; Aljizawi, Noura; Ahmed, Mohamed; Meletti, Levi; Lau, Justin (2026): "Banned Books: Analysis of Censorship on Amazon.com", Proceedings on Privacy Enhancing Technologies 2026(3):200-214. (DOI)] (server+platform) and [6Lipphardt, Friedemann; Ali, Moonis; Banzer, Martin; Feldmann, Anja; Gosain, Devashish (2026): "There is No War in Ba Sing Se: A Global Analysis of Content Moderation in Large Language Models", in: Proceedings of the Network and Distributed System Security Symposium. (Link)] (platform+server), which is why the family counts sum to 66 and the union is 63.

161 of 275 candidates (58.5%) are off topic, and every one has a named reason:

That 58.5% is the page's own headline caution: in this subject area a keyword probe is roughly 40% precise, and the words that do the damage are ordinary English ones.

vantage[].locations is free text and is folded through scripts/geo.mjs (shared with Crawling location and IP classification, so the pages agree). Unmapped strings are still counted as distinct locations — an unmapped string is a place the paper named, and dropping it would make “two or more vantage points” an artefact of the alias list — and every one is printed:

  4x  "Guangzhou"
  2x  "Dominican Republic"
  2x  "Jamaica"
  2x  "Puerto Rico"
  2x  "Saint Barthelemy"
  2x  "Saint Kitts and Nevis"
  2x  "Guadeloupe"
  2x  "Trinidad and Tobago"
  2x  "Martinique"
  2x  "Grenada"
  2x  "around the world"
  2x  "Taichung"
  2x  "Hangzhou"
  2x  "Longmont?"
  2x  "San Jose"

Twelve of the fifteen are real places geo.mjs has no alias for — the Caribbean cluster comes from the Cuba-embargo study's control set, the Chinese and Taiwanese cities from the GFW work. geo.mjs was not extended: it is shared with two published pages whose figures would move. The consequence is disclosed rather than fixed, and it is small — every paper in the residue already has two or more mapped locations.

detection[].phenomenon was used only as probe 2. It is unusable as a count: probe 2's own pattern matches 157 distinct phenomenon strings across 71 papers (printed by the report), and a wider exploratory pattern over phenomenon + technique + metric during scoping matched several hundred more, almost all off topic. Reading either set shows CPU throttling, GPS interference, WLAN interference, ad blocking, image filtering, reviewer-assignment manipulation and “clip-on memory manipulator attack” alongside the real ones. The figures on the page that look like phenomenon counts are not: they are hand verdicts (§4) and full-text signal probes (§6).

The “what blocked is operationalised as” table is eleven full-text regexes evaluated over the 63 papers. Each is:

• an upper bound on use — a sentence in related work matches; and
• a lower bound on the idea — a paper can compare against a control and never write the word.

Both bounds are stated on the content page, in the table's own caption. Median signals matched per paper: 4; 45 of 63 match three or more; 6 papers match none and are named in the output (two geodifference studies, a CCPA study, an IPFS attack paper, a connectivity characterisation and a P2P bootstrapping study — all of them papers whose blocking claim is not about a page body).

Two probes were repaired after reading their hits, and both repairs changed a published number:

• the LLM row matched an unanchored BERT, which matches the surname Deibert — Ron Deibert of the Citizen Lab, who is in the reference list of a large share of this literature. Before anchoring, the row read 3 papers with two of the three being reference-list noise. Every term in both the row and the widened probe is now \b-anchored.
• the widened probe's first draft included prompt\w*, which matches “censorship prompted”. It came out.

The control-vantage probe is deliberately published at two widths, because the width decides the claim: tight gives 32 of 63 and loose gives 47 of 63. A review pass found that the first version of the loose probe did not contain the tight one — 6 of the 32 tight hits matched no loose pattern, because the tight probe's “uncensored control” and “vantage point outside country” branches contain no literal “control” near a vantage word. So “tight” and “loose” were two different questions whose counts could not be compared, and the guard could not see it: it compared sizes, and a size comparison passes on two disjoint sets. Both loose probes are now built from their tight counterparts by construction ([…TIGHT, …extra]), and both guards assert containment: they throw and name the escaping papers. The published loose figure moved from 41 to 47 as a result.

The instrument table crosses tools[] (used or produced) against a full-text mention sweep, corpus-wide:

The page uses this table for one claim — that a structured tool query undercounts observatory reuse by about an order of magnitude — and for no usage figure. The bottom half of the table is why: Satellite is satellite Internet, Geneva is a city and a convention, Iris is an iOS analyser and a given name, ONI is a substring, and Encore and Augur are ordinary words. Any per-instrument adoption figure would need the same hand audit as §4.

“Block” alone is useless in this corpus: it is overwhelmingly blockchain, blocklist and ad blocking. An early draft of this probe used (block|deny|refus)[^.]{0,80}(EU|European|GDPR) and returned 42 papers, of which most were about Ethereum. The published probe is four tight patterns requiring a wall-shaped construction; the loose variant drops the requirement that the two halves be close together.

The same containment bug applied here and was found by the same review pass: with the first loose probe, 6 of the 9 tight hits were not loose hits, the union was 24 and the page said “all 27 distinct hits” — 9 + 18, added as if the sets nested. The loose probe now contains the tight one by construction, the guard asserts containment, and the union it prints is 24. All 24 were read, and the disposition of each is in the VERDICT map or printed as “not a candidate” in the output. That is the evidence behind the page's claim that the prevalence of GDPR-driven blocking has not been measured in these seven venues: it is a claim about 5,859 papers under two probe widths and a hand read of every hit, and it is stated on the page with that scope.

scripts/bgd_quotecheck.py (§16) looks for 94 strings — every phrase the page quotes, and every paper-sourced figure a reader could pull on — in the cited paper's own text, trying paper.cols.txt, paper.norm.txt and a live pypdf extraction of paper.pdf, each verbatim after normalisation and then folded to lowercase alphanumerics. It exits non-zero on any miss. Final run: 94 checked, 0 NOT FOUND.

Three strings failed on the first run, and all three were the same mistake — detection[].prevalence is a model summary, not a quote:

Two further quotes were replaced before publication because they were not in the papers at all — they were the extraction's summary field, which is model-written:

• ICLab was quoted as comparing “DNS, TCP, HTTP, TLS, traceroute, and packet observations to controls”. That is the summary. The paper's own words are “compares them with responses to matching DNS queries from our control node”, and its architecture figure labels a “Control vantage”. Both are now on the page and both are in the checker.
• Censored Planet was described in the summary's terms; the page now quotes “four remote measurement techniques (Augur, Satellite/Iris, Quack, and Hyperquack)” and “synchronized measurements on 6 different Internet protocols (IP, DNS, HTTP, HTTPS, Echo and Discard)”.

One BibTeX record was wrong in the corpus index and was corrected by hand: IMC/2014/censorship-in-the-wild… carried author = {Abdelberi, Chaabane and …}, with given and family name swapped. The paper's own title block reads “Abdelberi Chaabane” and Crossref agrees, so the entry is chaabane2014_censorship with {Chaabane, Abdelberi}. bibgen.mjs derived the citekey from the swapped field and would have published abdelberi2014_censorship.

node scripts/check_page_numbers.mjs pages/design_blocking_and_geodifference.txt work_bg/all_reports.txt — where all_reports.txt is the concatenation of the four committed outputs — ends “OK — every figure in the page traces”, with no ALLOW entries added for this page.

Two figures had to be made traceable rather than allow-listed, which is the point of the guard:

• the Censored Planet dashboard prints its counters with dots as thousands separators (116.924.640.072), so a page quoting them with commas could not be matched. bgd_render_checks.mjs now prints both forms.
• the page's “13 of 62 (21.0%)” single-vantage share was computed on the page and not by the script. The script now prints the percentage.

The guard is a presence check, not a binding check: it proves each number appears in an output, not that it is the right number for its sentence. The prose was therefore re-read by hand after every fold or population change, and by the review passes in §12.

Two scripts, both committed with their unedited output. HTTP status is printed before any byte count or content, because a 302 to a redirect page and an empty 200 are indistinguishable in a size.

• scripts/bgd_external_checks.sh — curl checks on every instrument, dataset, standard and legal instrument the page names.
• scripts/bgd_render_checks.mjs — Playwright renders for the four sources that are single-page apps, where “HTTP 200” says nothing about the data behind them. /usr/bin/chromium cannot be driven in this container, so Playwright's own build is used via PLAYWRIGHT_BROWSERS_PATH=/workspace/.playwright.

What the checks established, all on 2026-09-10:

Sources considered and rejected:

Four passes. The page text, the report script, its unedited output, the quote checker, the external checks and these notes were handed to each one, with the instruction that the author's context may not be exhaustive. The three focused passes ran in parallel against a frozen snapshot (work_bg/review1/, page MD5 f4b2a2041d998d27134d1d1502921c8f); the generic pass saw the corrected text. Rejections are recorded as fully as fixes: they are the only evidence of whether a reviewer slot is worth its cost.

Instructed to fetch rather than recall, and it did: every figure in the currency table was independently re-fetched rather than re-read from the committed output, including all seven Censored Planet repo dates, the OONI aggregation query, the Citizen Lab commit timestamp, both stale GFW dashboards and both RFC status records. Three findings, all three accepted:

Nothing was rejected from this pass. Its own summary of what it re-confirmed clean (OONI, Citizen Lab, Censored Planet's URLs and repos, GFWatch/GFWeb staleness with a search for any fresher mirror, M-Lab, IODA, Cloudflare Radar's 403, both RFCs, and every literal URL on the page) is the reason the currency table is stated as flatly as it is.

Re-ran the report itself (byte-identical to the committed output), traced every number on the page and in this log to it, and mutation-tested all six guards in copies of the script: no-verdict, verdict-names-a-non-candidate, unlabelled family, tight-vs-loose on both probe pairs, and the QUOTED list naming a paper outside the population. All six fired when broken. Three findings, all three accepted:

Nothing was rejected. The pass also confirmed, by re-derivation rather than by re-reading: the population funnel, the vantage table, all 11 signal rows with their window splits, the venue and year tables (both summing to 63), the instrument table's three columns for all 14 instruments, and 49 of 50 citekey-to-family mappings. Its one housekeeping item — eight mutation-test copies left in scripts/ because rm was denied to it — was cleaned up here.

Extracted the page's quoted spans itself rather than trusting the checker's list, verified all of them against three renderings of each paper, and checked all 30 bibliography additions against the papers' own title blocks and Crossref. Quotes, attributions, sibling-project confusions (GFWatch vs GFWeb) and industry claims came back clean. Seven findings, all seven accepted, six of them in the bibliography:

Six of the seven are name defects that bibgen.mjs inherited from the corpus index or from a landing-page parse, and one of them was in an entry a human had already corrected by hand. The lesson recorded here for the next run: correcting one author in an entry is not correcting the entry, and a landing-page parse that has ever produced an affiliation in an author slot will do it again.

Between the focused passes and the generic one, the page was read end to end once more against the report. Six defects that no guard could see, all fixed:

• An inverted share. “a measurement that only looks at HTTP bodies sees a quarter of it” — block pages are 72.3% of the Cuban geoblocking, so a body-only measurement sees the largest share and misses a quarter. The sentence said the opposite of its own source.
• “IMC carries the largest share of any venue by a factor of three.” IMC is 3.0% and the next venue is 1.4%: a factor of two. Replaced with the two figures.
• “a third of the papers … state a single vantage location.” It is 13 of 62, a fifth. The table three lines above had it right.
• A window claim built by adding two overlapping families. “7 of the 16 server-and-platform papers are from 2025–2026” added 11 + 5 without subtracting the 2 papers in both, and then miscounted the window. The distinct union is 14, of which 5 are 2025–2026; the report now prints both, and the sentence names the papers.
• Three papers named as the 2025–2026 platform slice included a 2024 one (the IPFS moderation study). Corrected to the three that are actually in the window.
• Two ambiguous “et al.” — Wu et al. is both the 2023 fully-encrypted-traffic paper and the 2025 Henan one (the same first author), and Singh et al. is both the 2017 Tor-exit study and the 2025 tracking-flow one (different people). Years added.

Also caught here: five anchor links that would have silently failed. DokuWiki generates heading ids by stripping punctuation, so #Jurisdiction: geo-targeted consent is #jurisdictiongeo-targeted_consent and #4. Your packet headers change the answer is #your_packet_headers_change_the_answer. All five cross-page anchors were rewritten to the ids in the targets' rendered DOM, which was fetched to get them.

The lesson for the review layer: three focused passes and a generic pass do not replace reading your own page against your own output. Four of the six above are arithmetic a figures pass would accept (each number is in the report), and two are the kind of ambiguity only a reader notices.

No checklist; asked for whatever the three focused passes were not looking for. It returned ten findings and was right about all of them. Four were already fixed by the author's own re-read (§ above) and are marked as such; the rest changed the page materially, and two changed the population's definition.

Nothing from this pass was rejected. It found the two defects that decide whether the page's central number means anything — the rule that did not describe its own population, and a hand-tallied table that contradicted the script — and neither was visible to a pass that checks figures against the report, because neither figure was in the report.

• Whether GFWatch's and GFWeb's pipelines still run. Both dashboards render, both stop in 2024. Whether the measurement stopped or only the dashboard did is not answerable from outside, and neither project publishes a status page. The page says exactly that.
• Whether ICLab's data is available anywhere. iclab.org cannot be reached by a client that verifies certificates, and no mirror was found. Not pursued further — the page treats the paper as the citation.
• The prevalence of GDPR-driven blocking, which is the page's own headline gap: 2 incidental measurements in 5,859 papers.
• How much bot management is misread as geoblocking. Both refuse a request on the basis of the client's address. No paper in the population separates them, and this run did not run the experiment that would.
• Any per-instrument adoption figure. §7's full-text column is a mention count over homograph-heavy names; turning it into a usage figure needs the §4 treatment and was not done.
• Whether the 2025–2026 growth in server-side and platform blocking is real. 7 of 16 such papers are in the two provisional venue-years. The page states the direction and the caveat, and does not test the trend — with 16 papers a trend test would be theatre.
• The false-positive rate of this page's own hand audit. 275 verdicts, one auditor. A second auditor on a sample would have given an agreement figure; none exists, and the page does not claim one. This is the largest single weakness in the population figure.

1. A live GH_TOKEN was printed into the run log. Testing whether the GitHub API had credentials, this run used ${GH_TOKEN:+set}${GH_TOKEN:-unset} — a form whose second half expands to the value whenever the variable is set. The token appeared in the transcript. The only safe form is ${VAR:+set} alone. It is disclosed here because a leak that is not disclosed cannot be rotated, and the check script now carries the warning inline.
2. otherToolsMentioned is an array of strings, not of objects. The first instrument probe read .name off each element, silently matched nothing, and reported 29 papers where the fixed version reports 35. Caught by reading a homonym hit, not by any guard.
3. Two probes matched an English word rather than a method: prompt\w* matched “censorship prompted”, and an unanchored BERT matched the surname Deibert. Both had reached a draft figure.
4. A first draft of the jurisdiction probe returned 42 papers, most of them about Ethereum, because “block” in this corpus is overwhelmingly blockchain. Replaced by four wall-shaped patterns and published at two widths.
5. Two “quotes” came from the extraction's model-written summary field and were replaced with the papers' own words (§9). Nothing was published in that state, but the checker had to be extended to catch the class rather than the instance.
6. scripts/sitemap.mjs reported a failure that was not one. It caches page source under out/pages/ and never expires it, so a roadmap row removed twenty minutes earlier still showed as “WRITTEN but still on the roadmap”. The cache entry was refreshed by hand; the script's cache policy is a known trap rather than a bug found here.
7. A literal <div id="root"> inside monospace truncated this page from section 11 onward, and it was live in that state for about two minutes. DokuWiki's inline code does not escape a tag: everything after it was dropped, silently, with no error and no failing check — check_wrap.mjs and check_tables.mjs both passed on the source. It was caught by fetching the rendered DOM and comparing its heading, table and <file> counts against the source's. Nothing but reading the render catches this class.
8. Every internal link on this page was namespace-relative. From provenance:design:, [[roadmap]] resolves to provenance:design:roadmap. Every link here is now root-anchored with a leading colon, and the rendered DOM was checked for wikilink2 (red-link) classes. Writing that sentence produced a third instance of the same bug: the example link, in monospace, was still a live link and rendered red, because inline code does not escape a wiki link any more than it escapes a tag. It needs … — which is what the sentence you are reading now uses.
9. report_blocking_geodifference.mjs exhausted the V8 heap on its first run, because it cached every paper's full text (0.5 GB on disk, over 4 GB as JS strings). It now does a single streaming pass for the whole-corpus regexes and caches only the 63 population texts.

Four scripts, each embedded below exactly as committed, followed by its output exactly as produced. The report script is long because the hand-verdict map for all 275 candidates lives inside it — that map is the audit trail, and putting it anywhere else would make the population unauditable without repo access.

The report. Every corpus figure on the content page comes from this file; the hand-verdict map for every candidate is the VERDICT object below, and section 4 of this page tabulates it.

report_blocking_geodifference.mjs

#!/usr/bin/env node
// Report for design:blocking_and_geodifference — every figure the page publishes,
// with the population it is a share of.
//
// Run:  node scripts/report_blocking_geodifference.mjs
//       node scripts/report_blocking_geodifference.mjs --list <family>
//
// Three rules from data/extract/README.md are load-bearing here:
//   * count PAPERS, never tuples;
//   * a sentinel is never an answer;
//   * `detection.phenomenon` / `.technique` are free text (~20% run-to-run
//     agreement on exact strings), so they are folded into families and reported
//     as rankings, never as precise percentages of a phenomenon.
//
// The population is NOT a probe output. Five probes build a CANDIDATE set; every
// candidate carries a hand verdict in VERDICT below, and the script throws if a
// candidate has no verdict or a verdict names a paper no probe caught. The
// families that make up the page's population are MEASURE_FAMILIES.
 
import fs from 'node:fs';
import path from 'node:path';
import { loadExtractions, dataRoot, isSentinel } from './lib.mjs';
import { normalizeLocation } from './geo.mjs';
 
const rows = loadExtractions();
const KEY = (p) => `${p.year}/${p.venue}/${p.slug}`;
const byKey = new Map(rows.map((p) => [KEY(p), p]));
const T = (p) => `${p.title ?? ''} • ${p.summary ?? ''}`;
const pct = (n, d) => (d === 0 ? 'n/a' : `${((100 * n) / d).toFixed(1)}%`);
const norm = (s) => s.replace(/\s+/g, ' ');
 
// ---------------------------------------------------------------- full text
// 5,855 papers of .cols text is ~0.5 GB on disk and more than 4 GB as JS strings,
// so nothing is cached corpus-wide: the whole-corpus regexes are evaluated in ONE
// streaming pass (corpusScan below) and only the population's texts are kept.
function ftRead(p) {
  const f = path.join(dataRoot(), 'fulltext', String(p.year), p.venue, p.slug, 'paper.cols.txt');
  // Whitespace is collapsed before any regex runs: a PDF line break inside a
  // phrase otherwise makes a multi-word probe silently undercount.
  return fs.existsSync(f) ? norm(fs.readFileSync(f, 'latin1')) : null;
}
const popCache = new Map();
function ftPop(p) {
  const k = KEY(p);
  if (!popCache.has(k)) popCache.set(k, ftRead(p));
  return popCache.get(k);
}
 
// ---- one streaming pass for every whole-corpus regex --------------------
const CORPUS_RE = {
  // A negative claim ("nobody classifies block pages with an LLM") must not rest
  // on the narrow signal probe of section 5, so the widened version runs over the
  // whole corpus. Two alternatives had to be repaired after reading the hits:
  // "prompt\\w*" matches "censorship prompted", and an unanchored BERT matches the
  // surname **Deibert** — Ron Deibert of the Citizen Lab, who is in the reference
  // list of a large share of this literature. Every term is now \\b-anchored.
  llmWide: /\b(LLMs?|large language model|GPT-?[0-9]|ChatGPT|Claude|Gemini|Llama|DeBERTa|BERTopic|BERT)\b[^.]{0,200}(block ?page|censor\w*|geo-?block\w*|blocked (page|site|domain|content))|((block ?page|censor\w*|geo-?block\w*)[^.]{0,200}\b(LLMs?|large language model|GPT-?[0-9]|ChatGPT|DeBERTa|BERTopic)\b)/i,
  ftinstr: /\bOONI\b|Open Observatory of Network Interference|Censored ?Planet|\bICLab\b|GFWatch|Citizen ?Lab('s)? (global )?test list|citizenlab\/test-lists/i,
  ftvocab: /block ?page|blocking page|geo-?block|geo-?differen|geo-?restrict|HTTP 451|unavailable for legal reasons/i,
};
const INSTRUMENTS = {
  OONI: /\bOONI\b|Open Observatory of Network Interference/i,
  'Censored Planet': /Censored ?Planet/i,
  ICLab: /\bICLab\b/i,
  'Citizen Lab test list': /Citizen ?Lab('s)? (global )?(test|citizen) ?lists?|citizenlab\/test-lists/i,
  Quack: /\bQuack\b/i,
  Hyperquack: /\bHyperquack\b/i,
  Augur: /\bAugur\b/i,
  'Iris (DNS)': /\bIris\b/i,
  Satellite: /\bSatellite\b/i,
  GFWatch: /\bGFWatch\b/i,
  Geneva: /\bGeneva\b/i,
  'M-Lab': /\bM-?Lab\b|Measurement Lab/i,
  'OpenNet Initiative': /OpenNet Initiative|\bONI\b/i,
  Encore: /\bEncore\b/i,
};
// Tight, wall-specific probes. "block" alone is useless here: in this corpus it is
// overwhelmingly blockchain, blocklist and ad blocking.
const WALL_TIGHT = [
  /(?:geo-?block|block(?:ed|ing|s)?|deny|denied|denies|refus\w+|unavailable|inaccessible)[^.]{0,70}\b(?:EU|EEA|European(?: Union)?)\b[^.]{0,50}(?:visitor|user|reader|resident|traffic|IP address|audienc)/i,
  /\b(?:EU|EEA|European(?: Union)?)\b[^.]{0,50}(?:visitor|user|reader|resident|traffic)s?[^.]{0,70}(?:are |were |is |was )?(?:geo-?block\w*|block\w*|denied access|refused|shut out|unavailable)/i,
  /GDPR[- ](?:induced|driven|related|motivated)[^.]{0,40}(?:geo-?block|block|withdrawal|exit)/i,
  /not available[^.]{0,40}in (?:the )?(?:EU|Europe|European Union|EEA)/i,
];
// The loose probe must CONTAIN the tight one, or "tight" and "loose" are two
// different questions rather than two widths of one, and the union is not
// max(|tight|,|loose|). The first version did not: 6 of 9 tight hits matched no
// loose pattern, so a review pass found the page reporting 9+18=27 "distinct"
// hits where the real union was 24. It is now built from WALL_TIGHT by
// construction, and the guard below asserts containment rather than size.
const WALL_LOOSE = [
  ...WALL_TIGHT,
  /\b(?:EU|EEA|European)\b[^.]{0,120}\b(?:geo-?block\w*|geo-?fenc\w*|block(?:ed|ing)?|unavailable|inaccessible|refus\w+|denied)\b/i,
  /\b(?:geo-?block\w*|geo-?fenc\w*)\b[^.]{0,120}\b(?:GDPR|CCPA|ePrivacy|California|EU|EEA|European)\b/i,
  /\b(?:GDPR|CCPA)\b[^.]{0,120}\b(?:geo-?block\w*|geo-?fenc\w*)\b/i,
];
const scan = new Map(); // key -> { ftinstr, ftvocab, instr: Set, wallTight, wallLoose }
const ftMissing = [];
for (const p of rows) {
  const t = ftRead(p);
  if (t === null) { ftMissing.push(p); scan.set(KEY(p), null); continue; }
  const rec = { instr: new Set() };
  for (const [n, re] of Object.entries(CORPUS_RE)) rec[n] = re.test(t);
  for (const [n, re] of Object.entries(INSTRUMENTS)) if (re.test(t)) rec.instr.add(n);
  rec.wallTight = WALL_TIGHT.some((re) => re.test(t));
  rec.wallLoose = WALL_LOOSE.some((re) => re.test(t));
  scan.set(KEY(p), rec);
}
const S = (p) => scan.get(KEY(p));
 
// ------------------------------------------------------------------- probes
// NOTE: otherToolsMentioned is an array of STRINGS, not of objects. Reading
// `.name` off it matched nothing at all in the first version of this probe.
const toolNames = (p) => [...p.tools.map((t) => t.name), ...p.otherToolsMentioned].map((n) => String(n ?? '').trim());
 
const INSTR_RE = /^(OONI|OONI ?Probe|OONI ?Explorer|OONI data|Censored ?Planet|ICLab|Quack|Hyperquack|Satellite|Augur|GFWatch|Encore|Geneva|Iris|CensorScope|Tor Metrics|Citizen ?Lab.*|Block(page)? ?(corpus|list of blockpages))$/i;
 
// Hand-added candidates, with the reason each one was added.
const MANUAL_CANDIDATES = new Map([
  ['2022/USENIX/openvpn-is-open-to-vpn-fingerprinting',
   'named by the 2026-09-10 generic review as the same shape as two papers in the population; no probe reaches it because its title and summary never say censorship, blocking or VPN blocking'],
]);
 
const PROBES = {
  title: (p) => /censorship|censored|censoring|censors\b|OONI|Censored Planet|ICLab|GFW\b|Great Firewall|geo-?block|blockpage|block page|internet shutdown|geo-?differen|geo-?restrict|geofenc|network interference|connection tampering|DNS manipulation|throttl/i.test(T(p)),
  detect: (p) => p.detection.some((t) => /censor|geo-?block|geo-?differen|block ?page|internet (filter|shutdown)|dns (manipulat|censor|interference|redirection|tamper)|network interference|connection tamper|url[- ]filter|keyword[- ]?(based )?filter|content (filter|censor)|over-?block|site blocking|blocked (site|domain|web|content|url|categor)|throttling of|traffic throttl|sanction/i.test(String(t.phenomenon ?? ''))),
  instr: (p) => toolNames(p).some((n) => INSTR_RE.test(n)),
  ftinstr: (p) => S(p) !== null && S(p).ftinstr,
  ftvocab: (p) => S(p) !== null && S(p).ftvocab,
  // Sixth probe, added after the first run: the jurisdiction-wall patterns of
  // section 8 caught seven papers no other probe saw. A probe that finds a paper
  // the candidate set missed is a recall hole, so it becomes a candidate probe.
  wall: (p) => S(p) !== null && S(p).wallTight,
  // Seventh probe, added after the second run. The first six are built around the
  // words "censor" and "geoblock", so they miss a paper that says "reachability",
  // "content moderation" or "blocking-resistant" instead. It found two papers that
  // belong in the population (the 2019 DoT/DoH reachability measurement and the
  // IPFS moderation study) and 44 that do not, all verdicted below.
  // Eighth probe, added after review. The bibliography's own newest entry —
  // "Tracking Internet Disruptions in Ukraine" (IMC 2025) — was caught by none of
  // probes 1-7, and the page had claimed that shutdowns as an event class were
  // absent from the corpus. They are not: there is an outage-detection literature
  // here, it just does not attribute unreachability to anybody's decision. That
  // distinction is the inclusion rule, so these papers are an ADJACENT family
  // rather than off topic.
  outage: (p) => /\b(outage|outages|shutdown|shutdowns|blackout|disruption|disruptions|internet resilience|connectivity loss|loss of connectivity|depeering|route withdrawal)\b/i.test(T(p)),
  // Ninth "probe": papers added by hand. A review pass named a paper that no
  // regex could reach — OpenVPN fingerprinting, whose title and summary contain
  // no blocking vocabulary at all — and a candidate set has to be able to admit
  // one, with the reason recorded, rather than pretend the regexes are complete.
  manual: (p) => MANUAL_CANDIDATES.has(KEY(p)),
  recall: (p) => /blocking-?resist|unblock|un-?censor|reachab|inaccessib|availability across|available across|not available in|content moderation|takedown|deplatform|shadow ?ban|across (\d+ )?(countries|regions|jurisdictions)|country-?level (difference|variation)|regional (variation|difference)|per-country/i.test(T(p)),
};
 
const candidates = new Map(); // key -> string[] of probe names
for (const p of rows) {
  const hit = Object.entries(PROBES).filter(([, f]) => f(p)).map(([n]) => n);
  if (hit.length) candidates.set(KEY(p), hit);
}
 
// --------------------------------------------------------------- verdicts
// One hand verdict per candidate, from reading title + summary and, where that
// was not enough, the matched sentence in paper.cols.txt. A paper may carry two
// families (Network Responses measures censorship AND corporate geoblocking), so
// families overlap and their counts do not sum to the population.
//
// MEASURE families = the page's population: the paper measures who could not
// reach what. Adjacent families are counted and named but excluded, because a
// circumvention proposal or an interview study is not a blocking measurement.
const MEASURE_FAMILIES = ['net', 'server', 'jurisdiction', 'platform', 'filter', 'probelist'];
const ADJACENT_FAMILIES = ['circumvention', 'outage', 'wall-encounter', 'interview', 'sok', 'vantage-instrument'];
const FAMILY_LABEL = {
  net: 'network-level interference with access (censorship, DNS manipulation, connection tampering, throttling, interception)',
  server: 'server- or platform-side refusal or differentiation by region (geoblocking, geodifference, sanctions)',
  jurisdiction: 'blocking or differentiation driven by a privacy law (GDPR wall, CCPA geofencing)',
  platform: 'moderation or censorship enforced inside one platform or service',
  filter: 'filtering products deployed on the client side (protective DNS), and their over-blocking',
  probelist: 'the censorship probe list itself as the object of study',
  circumvention: 'ADJACENT: evasion or circumvention system, or censor-side detection of it',
  interview: 'ADJACENT: human-subjects study of living with censorship',
  sok: 'ADJACENT: SoK or survey of the censorship literature',
  'vantage-instrument': 'ADJACENT: instrument for measuring FROM a place (see design:crawling_location)',
  'wall-encounter': 'ADJACENT: ran into, worked around, or reported others proposing an EU wall, without measuring how common it is',
  'outage': 'ADJACENT: measures unreachability without attributing it to a deliberate blocking decision — power, fibre, weather, war damage, misconfiguration',
};
const OFF_LABEL = {
  'off:incidental': 'one incidental mention, or a related-work citation only',
  'off:homonym-tor-metrics': 'homonym: names Tor Metrics, a Tor network dataset, not a blocking observatory',
  'off:homonym-iris': 'homonym: iRiS, an iOS private-API analyser, not Iris the DNS-manipulation platform',
  'off:homonym-augur': 'homonym: names Augur in a browser-fingerprinting context, not Augur the disruption prober',
  'off:homonym-throttling': 'homonym: bandwidth throttling as a defence, not throttling as censorship',
  'off:homonym-statistical-censoring': 'homonym: "censored"/"censoring" in the statistics or ML sense',
  'off:homonym-ml-concept-censorship': 'homonym: concept censorship in a generative model',
  'off:blockchain-sanctions': 'sanctions and blocklists on a blockchain, not access to content',
  'off:adblocking': 'content blocking by the user’s own ad blocker (see programming:filter_lists)',
  'off:security-blocking': 'blocking malware or phishing, not blocking by who or where you are',
  'off:vpn-proxy-ecosystem': 'the VPN or proxy ecosystem itself (see design:crawling_location)',
  'off:proxy-detection': 'detecting proxy or VPN clients (the server side of vantage choice)',
  'off:age-gate': 'age assurance, a different access gate (queued as privacy:age_assurance)',
  'off:bot-blocking': 'blocked for looking like a bot (see programming:crawler_detection)',
  'off:censor-as-attack-surface': 'censorship middleboxes abused as an amplifier, not measured as a censor',
  'off:llm-guardrails': 'model guardrails and refusals of what the user asked, not access blocking',
  'off:takedown-effects': 'the effect of a takedown or deplatforming on the target, not whether a client could reach content',
  'off:moderation-effects': 'the effect of moderating user posts on later user behaviour',
  'off:moderation-system': 'proposes or evaluates a moderation classifier — builds the blocker rather than measuring blocking',
  'off:resilience-modelling': 'models what would happen if countries disconnected, from routing graphs; no client reachability measured',
};
 
const VERDICT = {
  '2010/WWW/measurement-and-analysis-of-an-online-content-voting-network-a-case-study-of-dig': ['off:incidental'],
  '2011/IMC/analysis-of-country-wide-internet-outages-caused-by-censorship': ['net'],
  '2012/CCS/context-aware-web-security-threat-prevention': ['off:security-blocking'],
  '2012/USENIX/throttling-tor-bandwidth-parasites': ['off:homonym-throttling'],
  '2013/CCS/polyglots-crossing-origins-by-crossing-formats': ['off:incidental'],
  '2013/USENIX/automatic-mediation-of-privacy-sensitive-resource-access-in-smartphone-applicati': ['off:incidental'],
  '2013/CCS/protocol-misidentification-made-easy-with-format-transforming-encryption': ['circumvention'],
  '2013/CCS/users-get-routed-traffic-correlation-on-tor-by-realistic-adversaries': ['off:homonym-tor-metrics'],
  '2013/IMC/a-method-for-identifying-and-confirming-the-use-of-url-filtering-products-for-ce': ['net'],
  '2014/IMC/a-look-at-the-consequences-of-internet-censorship-through-an-isp-lens': ['net'],
  '2014/IMC/automated-detection-and-fingerprinting-of-censorship-block-pages': ['net'],
  '2014/IMC/censorship-in-the-wild-analyzing-internet-filtering-in-syria': ['net'],
  '2014/USENIX/brahmastra-driving-apps-to-test-the-security-of-third-party-components': ['off:incidental'],
  '2015/CCS/cachebrowser-bypassing-chinese-censorship-without-proxies-using-cached-content': ['circumvention'],
  '2015/CCS/iris-vetting-private-api-abuse-in-ios-applications': ['off:homonym-iris'],
  '2015/IMC/examining-how-the-great-firewall-discovers-hidden-circumvention-servers': ['net'],
  '2015/IMC/going-wild-large-scale-classification-of-open-dns-resolvers': ['net'],
  '2015/IMC/in-and-out-of-cuba-characterizing-cubas-connectivity': ['net'],
  '2015/IMC/ting-measuring-and-exploiting-latencies-between-all-tor-nodes': ['off:homonym-tor-metrics'],
  '2015/PETS/analyzing-the-great-firewall-of-china-over-space-and-time': ['net'],
  '2016/CCS/practical-censorship-evasion-leveraging-content-delivery-networks': ['circumvention'],
  '2016/IMC/an-analysis-of-the-privacy-and-security-risks-of-android-vpn-permission-enabled': ['off:vpn-proxy-ecosystem'],
  '2016/IMC/tunneling-for-transparency-a-large-scale-analysis-of-end-to-end-violations-in-th': ['net'],
  '2016/PETS/denasa-destination-naive-as-awareness-in-anonymous-communications': ['off:homonym-tor-metrics'],
  '2016/IEEE-SP/sok-towards-grounding-censorship-circumvention-in-empiricism': ['sok'],
  '2017/IMC/initial-measurements-of-the-cuban-street-network': ['off:incidental'],
  '2017/IMC/your-state-is-not-mine-a-closer-look-at-evading-stateful-internet-censorship': ['circumvention'],
  '2017/NDSS/dissecting-tor-bridges-a-security-evaluation-of-their-private-and-public-infrast': ['off:homonym-tor-metrics'],
  '2017/PETS/a-usability-evaluation-of-tor-launcher': ['interview'],
  '2017/PETS/deltashaper-enabling-unobservable-censorship-resistant-tcp-tunneling-over-videoc': ['circumvention'],
  '2017/PETS/topics-of-controversy-an-empirical-analysis-of-web-censorship-lists': ['probelist'],
  '2017/USENIX/characterizing-the-nature-and-dynamics-of-tor-exit-blocking': ['server'],
  '2017/USENIX/global-measurement-of-dns-manipulation': ['net'],
  '2017/IEEE-SP/augur-internet-wide-detection-of-connectivity-disruptions': ['net'],
  '2018/IMC/403-forbidden-a-global-view-of-cdn-geoblocking': ['server'],
  '2018/PETS/nomoads-effective-and-efficient-cross-app-mobile-ad-blocking': ['off:incidental'],
  '2020/WWW/filter-list-generation-for-underserved-regions': ['off:incidental'],
  '2023/NDSS/i-still-know-what-you-watched-last-sunday-privacy-of-the-hbbtv-protocol-in-the-european-smart-tv-landscape': ['off:incidental'],
  '2023/NDSS/chkplug-checking-gdpr-compliance-of-wordpress-plugins-via-cross-language-code-property-graph': ['wall-encounter'],
  '2023/PETS/everybodys-looking-for-ssomething-a-large-scale-evaluation-on-the-privacy-of-oau': ['wall-encounter'],
  '2026/PETS/the-role-of-online-forums-in-developer-understanding-of-privacy-law-a-reddit-cas': ['wall-encounter'],
  '2026/PETS/privacy-vs-profit-the-impact-of-googles-manifest-version-3-mv3-update-on-ad-bloc': ['off:incidental'],
  '2018/IMC/an-empirical-study-of-the-i2p-anonymity-network-and-its-censorship-resistance': ['circumvention'],
  '2018/IMC/how-to-catch-when-proxies-lie-verifying-the-physical-locations-of-network-proxie': ['vantage-instrument'],
  '2018/IMC/where-the-light-gets-in-analyzing-web-censorship-mechanisms-in-india': ['net'],
  '2018/USENIX/fp-scanner-the-privacy-implications-of-browser-fingerprint-inconsistencies': ['off:homonym-augur'],
  '2018/USENIX/quack-scalable-remote-measurement-of-application-layer-censorship': ['net'],
  '2018/USENIX/the-dangers-of-key-reuse-practical-attacks-on-ipsec-ike': ['off:incidental'],
  '2018/USENIX/towards-predicting-efficient-and-anonymous-tor-circuits': ['off:homonym-tor-metrics'],
  '2018/WWW/proxytorrent-untangling-the-free-http-s-proxy-ecosystem': ['off:vpn-proxy-ecosystem'],
  '2019/CCS/geneva-evolving-censorship-evasion-strategies': ['circumvention'],
  '2019/CCS/you-shall-not-join-a-measurement-study-of-cryptocurrency-peer-to-peer-bootstrapp': ['net'],
  '2019/CCS/your-cache-has-fallen-cache-poisoned-denial-of-service-attack': ['off:incidental'],
  '2019/IEEE-SP/breaking-lte-on-layer-two': ['off:incidental'],
  '2019/IMC/tales-from-the-porn-a-comprehensive-privacy-analysis-of-the-web-porn-ecosystem': ['off:incidental'],
  '2019/NDSS/the-use-of-tls-in-censorship-circumvention': ['circumvention'],
  '2019/WWW/evaluating-anti-fingerprinting-privacy-enhancing-technologies': ['off:homonym-tor-metrics'],
  '2020/CCS/censored-planet-an-internet-wide-longitudinal-censorship-observatory': ['net'],
  '2020/CCS/continuous-and-multiregional-monitoring-of-malicious-hosts': ['off:incidental'],
  '2020/IEEE-SP/meddling-middlemen-empirical-analysis-of-the-risks-of-data-saving-mobile-browser': ['off:incidental'],
  '2020/IMC/how-china-detects-and-blocks-shadowsocks': ['net'],
  '2020/IMC/investigating-large-scale-https-interception-in-kazakhstan': ['net'],
  '2020/NDSS/a-practical-approach-for-taking-down-avalanche-botnets-under-real-world-constraints': ['off:incidental'],
  '2020/NDSS/decentralized-control-a-case-study-of-russia': ['net'],
  '2020/NDSS/encrypted-dns-privacy-a-traffic-analysis-perspective': ['circumvention'],
  '2020/NDSS/massbrowser-unblocking-the-censored-web-for-the-masses-by-the-masses': ['circumvention'],
  '2020/NDSS/measuring-the-deployment-of-network-censorship-filters-at-global-scale': ['net'],
  '2020/NDSS/on-using-application-layer-middlebox-protocols-for-peeking-behind-nat-gateways': ['off:incidental'],
  '2020/NDSS/symtcp-eluding-stateful-deep-packet-inspection-with-automated-discrepancy-discovery': ['circumvention'],
  '2020/PETS/no-boundaries-data-exfiltration-by-third-parties-embedded-on-web-pages': ['off:incidental'],
  '2020/PETS/running-refraction-networking-for-real': ['circumvention'],
  '2020/WWW/apophanies-or-epiphanies-how-crawlers-impact-our-understanding-of-the-web': ['off:incidental'],
  '2020/WWW/deconstructing-googles-web-light-service': ['server'],
  '2020/WWW/maddroid-characterizing-and-detecting-devious-ad-contents-for-android-apps': ['off:incidental'],
  '2020/IEEE-SP/iclab-a-global-longitudinal-internet-censorship-measurement-platform': ['net'],
  '2020/CCS/youve-changed-detecting-malicious-browser-extensions-through-their-update-deltas': ['off:incidental'],
  '2021/IMC/throttling-twitter-an-emerging-censorship-technique-in-russia': ['net'],
  '2021/IMC/web-censorship-measurements-of-http-3-over-quic': ['net'],
  '2021/NDSS/hey-alexa-is-this-skill-safe-taking-a-closer-look-at-the-alexa-skill-ecosystem': ['off:incidental'],
  '2021/PETS/oblivious-dns-over-https-odoh-a-practical-privacy-enhancement-to-dns': ['off:incidental'],
  '2021/PETS/too-close-for-comfort-morasses-of-anti-censorship-in-the-era-of-cdns': ['circumvention'],
  '2021/PETS/domain-name-encryption-is-not-enough-privacy-leakage-via-ip-based-website-finger': ['off:incidental'],
  '2021/USENIX/blind-in-on-path-attacks-and-applications-to-vpns': ['off:incidental'],
  '2021/PETS/holes-in-the-geofence-privacy-vulnerabilities-in-smart-dns-services': ['circumvention'],
  '2021/USENIX/domain-shadowing-leveraging-content-delivery-networks-for-robust-blocking-resist': ['circumvention'],
  '2021/USENIX/how-great-is-the-great-firewall-measuring-chinas-dns-censorship': ['net'],
  '2021/USENIX/its-stressful-having-all-these-phones-investigating-sex-workers-safety-goals-ris': ['off:incidental'],
  '2021/USENIX/weaponizing-middleboxes-for-tcp-reflected-amplification': ['off:censor-as-attack-surface'],
  '2021/WWW/chinese-wall-or-swiss-cheese-keyword-filtering-in-the-great-firewall-of-china': ['net'],
  '2021/WWW/security-of-alerting-authorities-in-the-www-measuring-namespaces-dnssec-and-web': ['off:incidental'],
  '2021/WWW/understanding-the-impact-of-encrypted-dns-on-internet-censorship': ['net'],
  '2022/CCS/ready-raider-one-exploring-the-misuse-of-cloud-gaming-services': ['off:incidental'],
  '2022/CCS/hidden-in-plain-sight-exploring-encrypted-channels-in-android-apps': ['off:incidental'],
  '2022/IMC/characterizing-permanently-dead-links-on-wikipedia': ['off:incidental'],
  '2022/IMC/respect-the-origin-a-best-case-evaluation-of-connection-coalescing-in-the-wild': ['off:incidental'],
  '2022/IMC/rusty-clusters-dusting-an-ipv6-research-foundation': ['net'],
  '2022/IMC/tspu-russias-decentralized-censorship-system': ['net'],
  '2022/NDSS/auto-draft-244': ['off:vpn-proxy-ecosystem'],
  '2022/PETS/from-onion-not-found-to-guard-discovery': ['off:homonym-tor-metrics'],
  '2022/PETS/setting-the-bar-low-are-websites-complying-with-the-minimum-requirements-of-the': ['jurisdiction'],
  '2022/PETS/toward-uncensorable-anonymous-and-private-access-over-satoshi-blockchains': ['circumvention'],
  '2022/PETS/understanding-utility-and-privacy-of-demographic-data-in-education-technology-by': ['off:homonym-statistical-censoring'],
  '2022/USENIX/a-large-scale-investigation-into-geodifferences-in-mobile-apps': ['server'],
  '2022/USENIX/get-out-automated-discovery-of-application-layer-censorship-evasion-strategies': ['circumvention'],
  '2022/USENIX/how-and-why-people-use-virtual-private-networks': ['off:vpn-proxy-ecosystem'],
  '2022/USENIX/many-roads-lead-to-rome-how-packet-headers-influence-dns-censorship-measurement': ['net'],
  '2022/USENIX/off-path-network-traffic-manipulation-via-revitalized-icmp-redirect-attacks': ['off:incidental'],
  '2022/USENIX/online-website-fingerprinting-evaluating-website-fingerprinting-attacks-on-tor-i': ['off:incidental'],
  '2022/USENIX/security-at-the-end-of-the-tunnel-the-anatomy-of-vpn-mental-models-among-experts': ['off:vpn-proxy-ecosystem'],
  '2022/USENIX/the-security-lottery-measuring-client-side-web-security-inconsistencies': ['server'],
  '2022/WWW/measuring-the-privacy-vs-compatibility-trade-off-in-preventing-third-party-state': ['off:incidental'],
  '2022/WWW/reproducibility-and-replicability-of-web-measurement-studies': ['off:incidental'],
  '2022/IEEE-SP/investigating-influencer-vpn-ads-on-youtube': ['off:vpn-proxy-ecosystem'],
  '2023/IMC/ptperf-on-the-performance-evaluation-of-tor-pluggable-transports': ['circumvention'],
  '2023/IMC/wolf-in-sheeps-clothing-evaluating-security-risks-of-the-undelegated-record-on-d': ['off:incidental'],
  '2023/NDSS/a-systematic-study-of-the-consistency-of-two-factor-authentication-user-journeys-on-top-ranked-websites': ['off:incidental'],
  '2023/CCS/poster-circumventing-the-gfw-with-tls-record-fragmentation': ['circumvention'],
  '2023/NDSS/your-router-is-my-prober-measuring-ipv6-networks-via-icmp-rate-limiting-side-channels': ['off:incidental'],
  '2023/PETS/certainty-detecting-dns-manipulation-at-scale-using-tls-certificates': ['net'],
  '2023/CCS/txphishscope-towards-detecting-and-understanding-transaction-based-phishing-on-e': ['off:incidental'],
  '2023/CCS/weve-disabled-mfa-for-you-an-evaluation-of-the-security-and-usability-of-multi-f': ['off:incidental'],
  '2023/IMC/ethereums-proposer-builder-separation-promises-and-realities': ['off:blockchain-sanctions'],
  '2023/USENIX/a-study-of-chinas-censorship-and-its-evasion-through-the-lens-of-online-gaming': ['interview'],
  '2023/USENIX/all-of-them-claim-to-be-the-best-multi-perspective-study-of-vpn-users-and-vpn-pr': ['off:vpn-proxy-ecosystem'],
  '2023/USENIX/bypassing-tunnels-leaking-vpn-client-traffic-by-abusing-routing-tables': ['off:incidental'],
  '2023/USENIX/deresistor-toward-detection-resistant-probing-for-evasion-of-internet-censorship': ['circumvention'],
  '2023/USENIX/how-the-great-firewall-of-china-detects-and-blocks-fully-encrypted-traffic': ['net'],
  '2023/WWW/on-how-zero-knowledge-proof-blockchain-mixers-improve-and-worsen-user-privacy': ['off:blockchain-sanctions'],
  '2023/USENIX/network-responses-to-russias-invasion-of-ukraine-in-2022-a-cautionary-tale-for-i': ['net', 'server'],
  '2023/USENIX/mobileatlas-geographically-decoupled-measurements-in-cellular-networks-for-secur': ['vantage-instrument'],
  '2023/WWW/interval-censored-transformer-hawkes-detecting-information-operations-using-the': ['off:homonym-statistical-censoring'],
  '2023/WWW/measuring-and-evading-turkmenistans-internet-censorship-a-case-study-in-large-sc': ['net'],
  '2024/IMC/poster-twinkle-twinkle-streaming-star-illuminating-cdn-performance-over-starlink': ['off:incidental'],
  '2024/NDSS/content-censorship-in-the-interplanetary-file-system': ['platform'],
  '2024/NDSS/masterkey-automated-jailbreaking-of-large-language-model-chatbots': ['off:llm-guardrails'],
  '2024/NDSS/on-precisely-detecting-censorship-circumvention-in-real-world-networks': ['circumvention'],
  '2024/CCS/the-privacy-utility-trade-off-in-the-topics-api': ['off:incidental'],
  '2024/NDSS/understanding-the-implementation-and-security-implications-of-protective-dns-services': ['filter'],
  '2024/PETS/generational-differences-in-understandings-of-privacy-terminology': ['off:incidental'],
  '2024/CCS/understanding-routing-induced-censorship-changes-globally': ['net'],
  '2024/IEEE-SP/sok-technical-implementation-and-human-impact-of-internet-privacy-regulations': ['off:incidental'],
  '2024/IMC/the-wisdom-of-the-measurement-crowd-building-the-internet-yellow-pages-a-knowled': ['off:incidental'],
  '2024/IMC/yesterday-once-more-global-measurement-of-internet-traffic-shadowing-behaviors': ['off:incidental'],
  '2024/USENIX/bridging-barriers-a-survey-of-challenges-and-priorities-in-the-censorship-circum': ['interview'],
  '2024/USENIX/calculatency-leveraging-cross-layer-network-latency-measurements-to-detect-proxy': ['off:proxy-detection'],
  '2024/NDSS/modeling-and-detecting-internet-censorship-events': ['net'],
  '2024/USENIX/diffie-hellman-picture-show-key-exchange-stories-from-commercial-vowifi-deployme': ['off:incidental'],
  '2024/USENIX/digital-discrimination-of-users-in-sanctioned-states-the-case-of-the-cuba-embarg': ['server'],
  '2024/PETS/automatic-generation-of-web-censorship-probe-lists': ['probelist'],
  '2024/PETS/what-to-expect-when-you-re-accessing-an-exploration-of-user-privacy-rights-in-pe': ['jurisdiction'],
  '2024/USENIX/abandon-all-hope-ye-who-enter-here-a-dynamic-longitudinal-investigation-of-andro': ['off:incidental'],
  '2024/PETS/cross-contextual-examination-of-older-adults-privacy-concerns-behaviors-and-vuln': ['off:incidental'],
  '2024/USENIX/fingerprinting-obfuscated-proxy-traffic-with-encapsulated-tls-handshakes': ['off:proxy-detection'],
  '2024/USENIX/gfweb-measuring-the-great-firewalls-web-censorship-at-scale': ['net'],
  '2024/USENIX/less-defined-knowledge-and-more-true-alarms-reference-based-phishing-detection-w': ['off:incidental'],
  '2024/USENIX/loopy-hell-ow-infinite-traffic-loops-at-the-application-layer': ['off:incidental'],
  '2024/USENIX/phishdecloaker-detecting-captcha-cloaked-phishing-websites-via-hybrid-vision-bas': ['off:incidental'],
  '2024/USENIX/spotproxy-rediscovering-the-cloud-for-censorship-circumvention': ['circumvention'],
  '2024/USENIX/is-it-a-trap-a-large-scale-empirical-study-and-comprehensive-assessment-of-onlin': ['off:incidental'],
  '2024/WWW/phishinwebview-analysis-of-anti-phishing-entities-in-mobile-apps-with-webview-ta': ['off:security-blocking'],
  '2024/WWW/a-worldwide-view-on-the-reachability-of-encrypted-dns-services': ['net'],
  '2024/IEEE-SP/to-auth-or-not-to-auth-a-comparative-analysis-of-the-pre-and-post-login-security': ['off:incidental'],
  '2024/IEEE-SP/pryde-a-modular-generalizable-workflow-for-uncovering-evasion-attacks-against-st': ['circumvention'],
  '2025/CCS/local-frames-exploiting-inherited-origins-to-bypass-content-blockers': ['off:adblocking'],
  '2025/CCS/security-and-privacy-measurements-in-cellular-networks-novel-approaches-in-a-glo': ['off:incidental'],
  '2025/IEEE-SP/data-to-infinity-and-beyond-examining-data-sharing-and-reuse-practices-in-the-co': ['off:incidental'],
  '2025/CCS/deep-dive-into-in-app-browsers-uncovering-hidden-pitfalls-in-certificate-validat': ['off:incidental'],
  '2025/IEEE-SP/learning-from-censored-experiences-social-media-discussions-around-censorship-ci': ['interview'],
  '2025/CCS/fingerprinting-deep-packet-inspection-devices-by-their-ambiguities': ['net'],
  '2025/IEEE-SP/born-with-a-silver-spoon-on-the-in-security-of-native-granted-app-privileges-in': ['off:incidental'],
  '2025/IMC/the-developer-the-rfc-and-the-middlebox-an-http-2-compliance-story': ['off:incidental'],
  '2025/NDSS/wallbleed-a-memory-disclosure-vulnerability-in-the-great-firewall-of-china': ['net'],
  '2025/IMC/where-in-the-world-are-my-trackers-mapping-web-tracking-flow-across-diverse-geog': ['server'],
  '2025/USENIX/double-edged-shield-on-the-fingerprintability-of-customized-ad-blockers': ['off:adblocking'],
  '2025/NDSS/themis-regulating-textual-inversion-for-personalized-concept-censorship': ['off:homonym-ml-concept-censorship'],
  '2025/USENIX/irblock-a-large-scale-measurement-study-of-the-great-firewall-of-iran': ['net'],
  '2025/USENIX/easy-as-childs-play-an-empirical-study-on-age-verification-of-adult-oriented-and': ['off:age-gate'],
  '2025/USENIX/exposing-and-circumventing-sni-based-quic-censorship-of-the-great-firewall-of-ch': ['net', 'circumvention'],
  '2025/WWW/harmful-terms-and-where-to-find-them-measuring-and-modeling-unfavorable-financia': ['off:incidental'],
  '2025/IEEE-SP/analyzing-the-ios-local-network-permission-from-a-technical-and-user-perspective': ['off:incidental'],
  '2025/IEEE-SP/is-nobody-there-good-globally-measuring-connection-tampering-without-responsive': ['net'],
  '2025/IEEE-SP/transport-layer-obscurity-circumventing-sni-censorship-on-the-tls-layer': ['circumvention'],
  '2025/IMC/somesite-i-used-to-crawl-awareness-agency-and-efficacy-in-protecting-content-cre': ['off:bot-blocking'],
  '2025/NDSS/the-discriminative-power-of-cross-layer-rtts-in-fingerprinting-proxy-traffic': ['off:proxy-detection'],
  '2025/IEEE-SP/a-wall-behind-a-wall-emerging-regional-censorship-in-china': ['net'],
  '2025/IMC/dive-into-the-cloud-unveiling-the-ab-usage-of-serverless-cloud-function-in-the-w': ['off:incidental'],
  '2025/IEEE-SP/code-speaks-louder-exploring-security-and-privacy-relevant-regional-variations-i': ['server'],
  '2026/NDSS/mirage-private-mobility-based-routing-for-censorship-evasion': ['circumvention'],
  '2026/NDSS/beyond-rtt-an-adversarially-robust-two-tiered-approach-for-residential-proxy-detection': ['off:proxy-detection'],
  '2026/NDSS/characterizing-the-implementation-of-censorship-policies-in-chinese-llm-services': ['platform'],
  '2026/PETS/precarious-but-active-a-look-at-privacy-behaviors-in-chinese-transformative-fand': ['interview'],
  '2026/PETS/quicstep-evaluating-connection-migration-based-quic-censorship-circumvention': ['circumvention'],
  '2026/USENIX/from-mirai-to-gorilla-deep-dive-into-a-long-lasting-ddos-for-hire-botnet': ['off:incidental'],
  '2026/WWW/evasion-under-blockchain-sanctions': ['off:blockchain-sanctions'],
  '2026/PETS/banned-books-analysis-of-censorship-on-amazon-com': ['server', 'platform'],
  '2026/PETS/exercising-the-ccpa-opt-out-right-on-android-legally-mandated-but-practically-ch': ['off:incidental'],
  '2026/NDSS/chameleoscan-demystifying-and-detecting-ios-chameleon-apps-via-llm-powered-ui-exploration': ['off:incidental'],
  '2026/WWW/tracking-the-stray-sheep-understanding-dns-response-manipulation-in-the-wild': ['net'],
  '2026/NDSS/there-is-no-war-in-ba-sing-se-a-global-analysis-of-content-moderation-in-large-language-models': ['platform', 'server'],
  '2026/PETS/disclosure-divergence-measuring-privacy-policy-and-data-safety-misalignment-at-s': ['off:incidental'],
  '2026/NDSS/mvpnalyzer-an-investigative-framework-for-auditing-the-security-privacy-of-mobile-vpns': ['off:vpn-proxy-ecosystem'],
  '2026/PETS/on-the-suitability-of-llm-driven-agents-for-dark-pattern-audits': ['off:incidental'],
  // --- added by hand after the generic review pass, 2026-09-10 ---
  // Censor-side detection of circumvention traffic: proposes the fingerprinting
  // method rather than measuring a deployed censor, so ADJACENT, not population.
  '2022/USENIX/openvpn-is-open-to-vpn-fingerprinting': ['circumvention'],
  // --- added by the eighth (outage) probe, 2026-09-10, after review ---
  '2010/USENIX/sepia-privacy-preserving-aggregation-of-multi-domain-network-events-and-statisti': ['off:incidental'],
  '2013/IEEE-SP/sok-p2pwned-modeling-and-evaluating-the-resilience-of-peer-to-peer-botnets': ['off:incidental'],
  '2015/CCS/from-system-services-freezing-to-system-server-shutdown-in-android-all-you-need': ['off:incidental'],
  '2015/NDSS/the-resilience-of-the-internet-to-colluding-country-induced-connectivity-disrupt': ['off:resilience-modelling'],
  '2015/IMC/leveraging-internet-background-radiation-for-opportunistic-network-analysis': ['off:incidental'],
  '2016/IMC/optical-layer-failures-in-a-large-backbone': ['outage'],
  '2016/IMC/reasons-dynamic-addresses-change': ['off:incidental'],
  '2017/IMC/pinpointing-delay-and-forwarding-anomalies-using-large-scale-traceroute-measurem': ['outage'],
  '2018/IMC/advancing-the-art-of-internet-edge-outage-detection': ['outage'],
  '2019/IMC/challenges-in-the-decentralised-web-the-mastodon-case': ['off:incidental'],
  '2020/IMC/five-alarms-assessing-the-vulnerability-of-us-cellular-communication-infrastruct': ['outage'],
  '2020/WWW/outage-detecting-power-and-communication-outages-from-social-networks': ['outage'],
  '2022/IMC/are-we-ready-for-metaverse-a-measurement-study-of-social-virtual-reality-platfor': ['off:incidental'],
  '2022/IMC/internet-outage-detection-using-passive-analysis': ['outage'],
  '2022/IMC/is-my-internet-down-sifting-through-user-affecting-outages-with-google-trends': ['outage'],
  '2023/NDSS/brokenwire-wireless-disruption-of-ccs-electric-vehicle-charging': ['off:incidental'],
  '2023/USENIX/access-denied-assessing-physical-risks-to-internet-access-networks': ['outage'],
  '2024/IMC/poster-enhancing-internet-disruption-investigation-via-path-monitoring': ['outage'],
  '2024/IMC/poster-investigating-network-security-post-outage-open-ports-vulnerabilities': ['outage'],
  '2024/CCS/poster-flashguard-real-time-disruption-of-non-price-flash-loan-attacks-in-defi': ['off:incidental'],
  '2024/IEEE-SP/surveilling-the-masses-with-wi-fi-based-positioning-systems': ['off:incidental'],
  '2024/USENIX/fledging-will-continue-until-privacy-improves-empirical-analysis-of-googles-priv': ['off:incidental'],
  '2025/IMC/poster-power-out-internet-down-measuring-internet-resilience-in-south-africa': ['outage'],
  '2025/IMC/tracking-internet-disruptions-in-ukraine-insights-from-three-years-of-active-ful': ['outage'],
  '2025/NDSS/rethink-reveal-the-threat-of-electromagnetic-interference-on-power-inverters': ['off:incidental'],
  '2025/IMC/hello-genai-dissecting-human-to-generative-ai-calling': ['off:incidental'],
  // --- added by the seventh (recall) probe, 2026-09-10 ---
  '2011/IMC/pingin-in-the-rain': ['off:incidental'],
  '2012/NDSS/you-can-run-but-you-can-t-hide-exposing-network-location-for-targeted-dos-attack': ['off:incidental'],
  '2013/CCS/beheading-hydras-performing-effective-botnet-takedowns': ['off:takedown-effects'],
  '2015/PETS/blocking-resistant-communication-through-domain-fronting': ['circumvention'],
  '2016/USENIX/hey-you-have-a-problem-on-the-feasibility-of-large-scale-web-vulnerability-notif': ['off:incidental'],
  '2017/CCS/faulds-a-non-parametric-iterative-classifier-for-internet-wide-os-fingerprinting': ['off:incidental'],
  '2017/IMC/the-record-route-option-is-an-option': ['off:incidental'],
  '2019/IMC/an-end-to-end-large-scale-measurement-of-dns-over-encryption-how-far-have-we-com': ['net'],
  '2019/IMC/ddos-hide-seek-on-the-effectiveness-of-a-booter-services-takedown': ['off:takedown-effects'],
  '2020/IMC/cloud-provider-connectivity-in-the-flat-internet': ['off:incidental'],
  '2020/NDSS/precisely-characterizing-security-impact-in-a-flood-of-patches-via-symbolic-rule-comparison': ['off:incidental'],
  '2020/NDSS/withdrawing-the-bgp-re-routing-curtain-understanding-the-security-impact-of-bgp-poisoning-through-real-world-measurements': ['off:incidental'],
  '2020/USENIX/fuzzguard-filtering-out-unreachable-inputs-in-directed-grey-box-fuzzing-through': ['off:incidental'],
  '2020/WWW/what-apps-did-you-use-understanding-the-long-term-evolution-of-mobile-app-usage': ['off:incidental'],
  '2021/CCS/ghost-in-the-binder-binder-transaction-redirection-attacks-in-android-system-ser': ['off:incidental'],
  '2021/CCS/revisiting-nakamoto-consensus-in-asynchronous-networks-a-comprehensive-analysis': ['off:incidental'],
  '2021/IMC/measuring-dns-over-https-performance-around-the-world': ['off:incidental'],
  '2021/USENIX/a-large-scale-study-of-user-behavior-expectations-and-engagement-with-android-pe': ['off:incidental'],
  '2021/WWW/surrounded-by-the-clouds-a-comprehensive-cloud-reachability-study': ['off:incidental'],
  '2022/IMC/performance-characterization-of-videoconferencing-in-the-wild': ['off:incidental'],
  '2023/IMC/on-the-importance-of-being-an-as-an-approach-to-country-level-as-rankings': ['off:incidental'],
  '2023/USENIX/collide-power-leaking-inaccessible-data-with-software-based-power-side-channels': ['off:incidental'],
  '2023/USENIX/dscope-a-cloud-native-internet-telescope': ['off:incidental'],
  '2023/WWW/automated-content-moderation-increases-adherence-to-community-guidelines': ['off:moderation-effects'],
  '2024/IEEE-SP/a-representative-study-on-human-detection-of-artificially-generated-media-across': ['off:incidental'],
  '2024/IEEE-SP/no-easy-way-out-the-effectiveness-of-deplatforming-an-extremist-forum-to-suppres': ['off:takedown-effects'],
  '2024/IMC/destination-reachable-what-icmpv6-error-messages-reveal-about-their-sources': ['off:incidental'],
  '2024/IMC/whats-in-the-dataset-unboxing-the-apnic-per-as-user-population-dataset': ['off:incidental'],
  '2024/USENIX/guardians-of-the-galaxy-content-moderation-in-the-interplanetary-file-system': ['platform'],
  '2024/USENIX/dvsorder-ballot-randomization-flaws-threaten-voter-privacy': ['off:incidental'],
  '2024/USENIX/sok-or-solk-on-the-quantitative-study-of-sociodemographic-factors-and-computer-s': ['off:incidental'],
  '2025/CCS/poster-eris-evaluating-rov-via-icmpv6-rate-limiting-side-channels': ['off:incidental'],
  '2025/USENIX/i-cannot-write-this-because-it-violates-our-content-policy-understanding-content': ['off:llm-guardrails'],
  '2025/USENIX/assessing-the-aftermath-the-effects-of-a-global-takedown-against-ddos-for-hire-s': ['off:takedown-effects'],
  '2025/WWW/causal-insights-into-parlers-content-moderation-shift-effects-on-toxicity-and-fa': ['off:moderation-effects'],
  '2025/WWW/unveiling-network-performance-in-the-wild-an-ad-driven-analysis-of-mobile-downlo': ['off:incidental'],
  '2025/NDSS/a-multifaceted-study-on-the-use-of-tls-and-auto-detect-in-email-ecosystems': ['off:incidental'],
  '2025/USENIX/darkgram-a-large-scale-analysis-of-cybercriminal-activity-channels-on-telegram': ['off:takedown-effects'],
  '2026/NDSS/actively-understanding-the-dynamics-and-risks-of-the-threat-intelligence-ecosystem': ['off:incidental'],
  '2026/NDSS/vulsca-a-community-level-sca-approach-for-accurate-c-c-supply-chain-vulnerability-analysis': ['off:incidental'],
  '2026/NDSS/from-noise-to-signal-precisely-identify-affected-packages-of-known-vulnerabilities-in-npm-ecosystem': ['off:incidental'],
  '2026/NDSS/crack-in-the-armor-underlying-infrastructure-threats-to-rpki-publication-point-reachability': ['off:incidental'],
  '2026/NDSS/enhancing-legal-document-security-and-accessibility-with-taf': ['off:incidental'],
  '2026/PETS/understanding-how-university-guidelines-address-privacy-and-security-issues-of-g': ['off:incidental'],
  '2026/PETS/the-empire-strikes-back-at-your-privacy-an-archaeology-of-tracking-on-government': ['off:incidental'],
  '2026/WWW/visual-content-moderation-in-messaging-systems': ['off:moderation-system'],
};
 
// Both directions, or the map rots silently the next time the probe moves.
const noVerdict = [...candidates.keys()].filter((k) => !(k in VERDICT));
const notCandidate = Object.keys(VERDICT).filter((k) => !candidates.has(k));
if (noVerdict.length || notCandidate.length) {
  throw new Error(
    `hand-verdict map is out of step with the probes:\n` +
      `  ${noVerdict.length} candidate(s) with no verdict:\n${noVerdict.map((k) => '    ' + k).join('\n')}\n` +
      `  ${notCandidate.length} verdict(s) for a paper no probe caught:\n${notCandidate.map((k) => '    ' + k).join('\n')}`,
  );
}
for (const [k, v] of Object.entries(VERDICT)) {
  for (const fam of v) {
    if (!MEASURE_FAMILIES.includes(fam) && !ADJACENT_FAMILIES.includes(fam) && !(fam in OFF_LABEL)) {
      throw new Error(`${k}: family "${fam}" has no label`);
    }
  }
}
 
const inFamily = (fam) => Object.entries(VERDICT).filter(([, v]) => v.includes(fam)).map(([k]) => byKey.get(k));
const BLOCK = [...new Set(MEASURE_FAMILIES.flatMap(inFamily))].sort((a, b) => a.year - b.year || a.venue.localeCompare(b.venue));
 
 
// ---------------------------------------------------------------- censor fold
// Who is doing the blocking, one label per paper in the population, assigned by
// hand from the paper. A review pass caught the page hand-tallying this table in
// prose and getting China wrong by a factor of two, so the split is produced
// here and the page copies it. Every population paper must have a label.
const CENSOR = {
  '2011/IMC/analysis-of-country-wide-internet-outages-caused-by-censorship': 'Egypt and Libya',
  '2013/IMC/a-method-for-identifying-and-confirming-the-use-of-url-filtering-products-for-ce': 'many countries at once',
  '2014/IMC/a-look-at-the-consequences-of-internet-censorship-through-an-isp-lens': 'Pakistan',
  '2014/IMC/automated-detection-and-fingerprinting-of-censorship-block-pages': 'many countries at once',
  '2014/IMC/censorship-in-the-wild-analyzing-internet-filtering-in-syria': 'Syria',
  '2015/IMC/examining-how-the-great-firewall-discovers-hidden-circumvention-servers': 'China',
  '2015/IMC/going-wild-large-scale-classification-of-open-dns-resolvers': 'many countries at once',
  '2015/IMC/in-and-out-of-cuba-characterizing-cubas-connectivity': 'Cuba',
  '2015/PETS/analyzing-the-great-firewall-of-china-over-space-and-time': 'China',
  '2016/IMC/tunneling-for-transparency-a-large-scale-analysis-of-end-to-end-violations-in-th': 'many countries at once',
  '2017/IEEE-SP/augur-internet-wide-detection-of-connectivity-disruptions': 'many countries at once',
  '2017/PETS/topics-of-controversy-an-empirical-analysis-of-web-censorship-lists': 'many countries at once',
  '2017/USENIX/global-measurement-of-dns-manipulation': 'many countries at once',
  '2017/USENIX/characterizing-the-nature-and-dynamics-of-tor-exit-blocking': 'website operators (Tor exits)',
  '2018/IMC/where-the-light-gets-in-analyzing-web-censorship-mechanisms-in-india': 'India',
  '2018/IMC/403-forbidden-a-global-view-of-cdn-geoblocking': 'website operators and CDNs',
  '2018/USENIX/quack-scalable-remote-measurement-of-application-layer-censorship': 'many countries at once',
  '2019/CCS/you-shall-not-join-a-measurement-study-of-cryptocurrency-peer-to-peer-bootstrapp': 'many countries at once',
  '2019/IMC/an-end-to-end-large-scale-measurement-of-dns-over-encryption-how-far-have-we-com': 'many countries at once',
  '2020/CCS/censored-planet-an-internet-wide-longitudinal-censorship-observatory': 'many countries at once',
  '2020/IEEE-SP/iclab-a-global-longitudinal-internet-censorship-measurement-platform': 'many countries at once',
  '2020/IMC/how-china-detects-and-blocks-shadowsocks': 'China',
  '2020/IMC/investigating-large-scale-https-interception-in-kazakhstan': 'Kazakhstan',
  '2020/NDSS/decentralized-control-a-case-study-of-russia': 'Russia',
  '2020/NDSS/measuring-the-deployment-of-network-censorship-filters-at-global-scale': 'many countries at once',
  '2020/WWW/deconstructing-googles-web-light-service': 'website operators and CDNs',
  '2021/IMC/throttling-twitter-an-emerging-censorship-technique-in-russia': 'Russia',
  '2021/IMC/web-censorship-measurements-of-http-3-over-quic': 'many countries at once',
  '2021/USENIX/how-great-is-the-great-firewall-measuring-chinas-dns-censorship': 'China',
  '2021/WWW/chinese-wall-or-swiss-cheese-keyword-filtering-in-the-great-firewall-of-china': 'China',
  '2021/WWW/understanding-the-impact-of-encrypted-dns-on-internet-censorship': 'many countries at once',
  '2022/IMC/rusty-clusters-dusting-an-ipv6-research-foundation': 'China',
  '2022/IMC/tspu-russias-decentralized-censorship-system': 'Russia',
  '2022/PETS/setting-the-bar-low-are-websites-complying-with-the-minimum-requirements-of-the': 'website operators (law-driven)',
  '2022/USENIX/many-roads-lead-to-rome-how-packet-headers-influence-dns-censorship-measurement': 'many countries at once',
  '2022/USENIX/a-large-scale-investigation-into-geodifferences-in-mobile-apps': 'app stores and developers',
  '2022/USENIX/the-security-lottery-measuring-client-side-web-security-inconsistencies': 'website operators and CDNs',
  '2023/PETS/certainty-detecting-dns-manipulation-at-scale-using-tls-certificates': 'many countries at once',
  '2023/USENIX/how-the-great-firewall-of-china-detects-and-blocks-fully-encrypted-traffic': 'China',
  '2023/USENIX/network-responses-to-russias-invasion-of-ukraine-in-2022-a-cautionary-tale-for-i': 'Russia',
  '2023/WWW/measuring-and-evading-turkmenistans-internet-censorship-a-case-study-in-large-sc': 'Turkmenistan',
  '2024/CCS/understanding-routing-induced-censorship-changes-globally': 'many countries at once',
  '2024/NDSS/modeling-and-detecting-internet-censorship-events': 'many countries at once',
  '2024/NDSS/content-censorship-in-the-interplanetary-file-system': 'a platform (IPFS, LLM services, a retailer)',
  '2024/NDSS/understanding-the-implementation-and-security-implications-of-protective-dns-services': 'resolver operators',
  '2024/PETS/what-to-expect-when-you-re-accessing-an-exploration-of-user-privacy-rights-in-pe': 'website operators (law-driven)',
  '2024/PETS/automatic-generation-of-web-censorship-probe-lists': 'China',
  '2024/USENIX/gfweb-measuring-the-great-firewalls-web-censorship-at-scale': 'China',
  '2024/USENIX/digital-discrimination-of-users-in-sanctioned-states-the-case-of-the-cuba-embarg': 'Cuba',
  '2024/USENIX/guardians-of-the-galaxy-content-moderation-in-the-interplanetary-file-system': 'a platform (IPFS, LLM services, a retailer)',
  '2024/WWW/a-worldwide-view-on-the-reachability-of-encrypted-dns-services': 'many countries at once',
  '2025/CCS/fingerprinting-deep-packet-inspection-devices-by-their-ambiguities': 'many countries at once',
  '2025/IEEE-SP/is-nobody-there-good-globally-measuring-connection-tampering-without-responsive': 'many countries at once',
  '2025/IEEE-SP/a-wall-behind-a-wall-emerging-regional-censorship-in-china': 'China',
  '2025/IEEE-SP/code-speaks-louder-exploring-security-and-privacy-relevant-regional-variations-i': 'app stores and developers',
  '2025/IMC/where-in-the-world-are-my-trackers-mapping-web-tracking-flow-across-diverse-geog': 'website operators and CDNs',
  '2025/NDSS/wallbleed-a-memory-disclosure-vulnerability-in-the-great-firewall-of-china': 'China',
  '2025/USENIX/irblock-a-large-scale-measurement-study-of-the-great-firewall-of-iran': 'Iran',
  '2025/USENIX/exposing-and-circumventing-sni-based-quic-censorship-of-the-great-firewall-of-ch': 'China',
  '2026/NDSS/there-is-no-war-in-ba-sing-se-a-global-analysis-of-content-moderation-in-large-language-models': 'a platform (IPFS, LLM services, a retailer)',
  '2026/NDSS/characterizing-the-implementation-of-censorship-policies-in-chinese-llm-services': 'China',
  '2026/PETS/banned-books-analysis-of-censorship-on-amazon-com': 'a platform (IPFS, LLM services, a retailer)',
  '2026/WWW/tracking-the-stray-sheep-understanding-dns-response-manipulation-in-the-wild': 'many countries at once',
};
 
// ------------------------------------------------------------------ printing
const H = (s) => console.log(`\n\n===== ${s} =====`);
const table = (head, rowsIn) => {
  console.log(head.join('\t'));
  for (const r of rowsIn) console.log(r.join('\t'));
};
 
const listFam = process.argv.includes('--list') ? process.argv[process.argv.indexOf('--list') + 1] : null;
if (listFam) {
  const ps = listFam === 'BLOCK' ? BLOCK : inFamily(listFam);
  for (const p of ps.sort((a, b) => a.year - b.year)) console.log(`${KEY(p)}\t${p.title}`);
  process.exit(0);
}
 
console.log(`report_blocking_geodifference.mjs — run ${new Date().toISOString().slice(0, 10)}`);
console.log(`corpus: ${rows.length} extracted papers, ${new Set(rows.map((p) => p.venue)).size} venues, ${Math.min(...rows.map((p) => p.year))}-${Math.max(...rows.map((p) => p.year))}`);
console.log(`paper.cols.txt missing for ${ftMissing.length} paper(s): ${ftMissing.map(KEY).join(', ') || 'none'}`);
console.log(`corpus reference denominators: crawled ${rows.filter((p) => p.crawlConfig !== null || p.studyTypes.includes('automated-web-crawl')).length}, web platform ${rows.filter((p) => p.platforms.includes('web')).length}, measuredFrom ${rows.filter((p) => p.vantage.length > 0).length}, scanned ${rows.filter((p) => p.studyTypes.includes('network-scan-or-probe')).length}`);
 
H('0. The inclusion rule, written down before the counts');
console.log(`A paper is IN the population if it measures whether some client could reach
some content or service AND attributes the failures to a deliberate blocking
decision taken by someone other than the client — a state, an ISP, a resolver
operator, a CDN, a platform, or the content owner — where that decision is keyed
on WHO OR WHERE the client is, on the content's acceptability to an authority or
platform, or on the traffic looking like an attempt to evade such a decision.
 
Explicitly OUT, each with its own off-topic family below:
  * blocking keyed on the content being malicious (malware, phishing, spam);
  * blocking keyed on the client looking automated (programming:crawler_detection);
  * blocking the client chose (its own ad blocker or filter list);
  * a system PROPOSED to evade blocking, and a DETECTOR proposed for finding
    circumvention traffic (both counted as the ADJACENT family "circumvention");
  * the effect of a takedown or of moderating user posts on later behaviour.
 
The third key was added after a review pass: this literature also measures
blocking keyed on THE TRAFFIC LOOKING LIKE AN ATTEMPT TO EVADE a blocking
decision — Shadowsocks, fully encrypted flows, an SNI. Measuring how a DEPLOYED
censor detects and blocks such traffic is in the population; proposing a
detector for it is not. The two sides are one sentence apart and the first
version of this rule did not separate them, so two papers sat in the population
under a rule that appeared to exclude them.
 
One deliberate boundary case is IN: the protective-DNS study, whose motive is
security rather than geography, because its claim has exactly this page's shape
(control resolver vs treatment resolver, and a definition of "blocked" that has
to survive an NXDOMAIN and a sinkhole). It is the whole "filter" family, so its
effect on any figure can be checked by subtracting one.`);
 
H('1. Candidate probes (floors, not populations)');
table(['probe', 'papers', 'of corpus'], Object.keys(PROBES).map((n) => {
  const c = [...candidates.values()].filter((v) => v.includes(n)).length;
  return [n, c, pct(c, rows.length)];
}));
console.log(`union of all ${Object.keys(PROBES).length} probes: ${candidates.size} candidates (${pct(candidates.size, rows.length)} of corpus)`);
const bySize = new Map();
for (const v of candidates.values()) bySize.set(v.length, (bySize.get(v.length) ?? 0) + 1);
console.log('candidates by how many probes caught them: ' + [...bySize.entries()].sort().map(([k, n]) => `${k} probe(s): ${n}`).join(', '));
 
H(`2. Hand verdicts over all ${candidates.size} candidates`);
const famCounts = [...MEASURE_FAMILIES, ...ADJACENT_FAMILIES].map((f) => [f, inFamily(f).length, FAMILY_LABEL[f]]);
table(['family', 'papers', 'what it is'], famCounts);
console.log(`\nPOPULATION (union of the ${MEASURE_FAMILIES.length} measurement families): ${BLOCK.length} papers`);
const doubleFamily = Object.entries(VERDICT).filter(([, v]) => v.filter((f) => MEASURE_FAMILIES.includes(f)).length > 1);
console.log(`  sum of the measurement family counts: ${MEASURE_FAMILIES.reduce((a, f) => a + inFamily(f).length, 0)} — higher than ${BLOCK.length} because ${doubleFamily.length} paper(s) carry two measurement families: ${doubleFamily.map(([k, v]) => `${k} [${v.join('+')}]`).join('; ')}`);
const offCounts = Object.keys(OFF_LABEL).map((f) => [f, inFamily(f).length, OFF_LABEL[f]]).filter((r) => r[1] > 0);
console.log('');
table(['off-topic family', 'papers', 'why it is off topic'], offCounts);
const accounted = new Set(Object.keys(VERDICT));
console.log(`\nverdicts: ${accounted.size}; candidates: ${candidates.size}; unaccounted: ${candidates.size - accounted.size}`);
console.log(`off topic in total: ${Object.keys(OFF_LABEL).flatMap(inFamily).length} of ${candidates.size} candidates (${pct(Object.keys(OFF_LABEL).flatMap(inFamily).length, candidates.size)}) — the price of a wide probe`);
 
H('3. The population over time and across venues');
const years = [...new Set(rows.map((p) => p.year))].sort();
table(['year', 'papers in population', 'corpus papers', 'share of corpus year'],
  years.map((y) => [y + (y >= 2025 ? ' (provisional)' : ''), BLOCK.filter((p) => p.year === y).length, rows.filter((p) => p.year === y).length, pct(BLOCK.filter((p) => p.year === y).length, rows.filter((p) => p.year === y).length)]));
const windows = [[2010, 2014], [2015, 2019], [2020, 2024], [2025, 2026]];
// Every MEASURE family gets a column, computed from the list rather than typed
// out, so a family added later cannot silently vanish from this table — which is
// what happened to `filter` on the first draft of the page.
const WINDOW_COLS = [...MEASURE_FAMILIES, 'circumvention'];
table(['window', 'population', ...WINDOW_COLS],
  windows.map(([a, b]) => [`${a}-${b}${a === 2025 ? ' (provisional)' : ''}`,
    BLOCK.filter((p) => p.year >= a && p.year <= b).length,
    ...WINDOW_COLS.map((f) => inFamily(f).filter((p) => p.year >= a && p.year <= b).length)]));
table(['venue', 'population', 'corpus', 'share of venue'],
  [...new Set(rows.map((p) => p.venue))].sort().map((v) => [v, BLOCK.filter((p) => p.venue === v).length, rows.filter((p) => p.venue === v).length, pct(BLOCK.filter((p) => p.venue === v).length, rows.filter((p) => p.venue === v).length)]));
table(['platform (multi-valued)', 'papers in population', 'share of ' + BLOCK.length],
  ['web', 'mobile', 'other-online-service', 'iot', 'offline', 'not-applicable'].map((pl) => [pl, BLOCK.filter((p) => p.platforms.includes(pl)).length, pct(BLOCK.filter((p) => p.platforms.includes(pl)).length, BLOCK.length)]));
// server and platform overlap by two papers, so the page's "server-and-platform"
// claim needs the DISTINCT union, not the sum of two family counts.
const srvPlat = [...new Set([...inFamily('server'), ...inFamily('platform')])];
console.log(`\nserver + platform, distinct papers: ${srvPlat.length} (server ${inFamily('server').length} + platform ${inFamily('platform').length} - ${inFamily('server').filter((p) => inFamily('platform').includes(p)).length} in both)`);
console.log(`  of those, 2025-2026: ${srvPlat.filter((p) => p.year >= 2025).length} — ${srvPlat.filter((p) => p.year >= 2025).map(KEY).join(', ')}`);
console.log(`  platform family in 2025-2026: ${inFamily('platform').filter((p) => p.year >= 2025).map(KEY).join(', ')}`);
console.log(`\nin the crawled population (crawlConfig or automated-web-crawl): ${BLOCK.filter((p) => p.crawlConfig !== null || p.studyTypes.includes('automated-web-crawl')).length} of ${BLOCK.length}`);
console.log(`ran a network scan or probe: ${BLOCK.filter((p) => p.studyTypes.includes('network-scan-or-probe')).length} of ${BLOCK.length}`);
console.log(`reanalyses an existing dataset: ${BLOCK.filter((p) => p.studyTypes.includes('existing-dataset-analysis')).length} of ${BLOCK.length}`);
 
H('3b. Who is doing the blocking (hand-keyed, one label per paper)');
const censorMissing = BLOCK.filter((p) => !(KEY(p) in CENSOR));
const censorExtra = Object.keys(CENSOR).filter((k) => !BLOCK.some((p) => KEY(p) === k));
if (censorMissing.length || censorExtra.length) {
  throw new Error(
    `CENSOR map out of step with the population: ${censorMissing.length} paper(s) with no label (${censorMissing.map(KEY).join(', ')}); ` +
      `${censorExtra.length} label(s) for a paper outside it (${censorExtra.join(', ')})`,
  );
}
const censorCounts = new Map();
for (const p of BLOCK) censorCounts.set(CENSOR[KEY(p)], [...(censorCounts.get(CENSOR[KEY(p)]) ?? []), p]);
table(['who is blocking', 'papers', `share of ${BLOCK.length}`, 'years'],
  [...censorCounts.entries()].sort((a, b) => b[1].length - a[1].length).map(([k, ps]) => [
    k, ps.length, pct(ps.length, BLOCK.length), `${Math.min(...ps.map((p) => p.year))}-${Math.max(...ps.map((p) => p.year))}`]));
console.log(`labels sum to ${[...censorCounts.values()].reduce((a, ps) => a + ps.length, 0)}, which must equal ${BLOCK.length}`);
for (const [k, ps] of [...censorCounts.entries()].sort()) {
  console.log(`\n-- ${k} (${ps.length})`);
  for (const p of ps.sort((a, b) => a.year - b.year)) console.log(`   ${KEY(p)}`);
}
 
H('4. Vantage points: does a blocking claim have a control?');
// Two vantage points is the structural minimum for a blocking claim: one inside
// the condition, one outside. `vantage.locations` is free text, so it is folded
// through geo.mjs and the residue is printed.
const residue = new Map();
function locs(p) {
  const out = new Set();
  let multi = false;
  for (const v of p.vantage) {
    for (const raw of v.locations ?? []) {
      if (isSentinel(raw)) continue;
      const n = normalizeLocation(raw);
      if (n.kind === 'country' || n.kind === 'region') out.add(n.value);
      else if (n.kind === 'multi') multi = true;
      else {
        // An unmapped string is still a location the paper named ("Guangzhou",
        // "Trinidad and Tobago" — geo.mjs has no alias for either). Counting it
        // as distinct keeps the "two or more vantage points" figure from being
        // an artefact of the alias list; the string is printed as residue below.
        residue.set(raw, (residue.get(raw) ?? 0) + 1);
        out.add(`unmapped:${String(raw).toLowerCase().trim()}`);
      }
    }
  }
  return { set: out, multi };
}
const withVantage = BLOCK.filter((p) => p.vantage.length > 0);
const noVantageTuple = BLOCK.filter((p) => p.vantage.length === 0);
const statedLoc = withVantage.filter((p) => { const l = locs(p); return l.set.size > 0 || l.multi; });
const twoPlus = withVantage.filter((p) => { const l = locs(p); return l.multi || l.set.size >= 2; });
console.log(`population: ${BLOCK.length}`);
console.log(`  has at least one vantage tuple: ${withVantage.length} (${pct(withVantage.length, BLOCK.length)})`);
console.log(`  no vantage tuple at all (pure reanalysis, or never said): ${noVantageTuple.length}`);
console.log(`  names at least one location, folded: ${statedLoc.length} of ${withVantage.length} (${pct(statedLoc.length, withVantage.length)})`);
console.log(`  two or more distinct locations, or an explicit multi-country vantage: ${twoPlus.length} of ${withVantage.length} (${pct(twoPlus.length, withVantage.length)})`);
console.log(`  exactly one location and no multi-country claim: ${statedLoc.length - twoPlus.length} of ${withVantage.length} (${pct(statedLoc.length - twoPlus.length, withVantage.length)})`);
const infra = new Map();
for (const p of BLOCK) for (const v of p.vantage) if (!isSentinel(v.infrastructure)) infra.set(v.infrastructure, (infra.get(v.infrastructure) ?? new Set()).add(KEY(p)));
table(['vantage infrastructure', 'papers', `share of ${withVantage.length}`], [...infra.entries()].sort((a, b) => b[1].size - a[1].size).map(([k, s]) => [k, s.size, pct(s.size, withVantage.length)]));
const infraSentinel = BLOCK.filter((p) => p.vantage.length > 0 && p.vantage.every((v) => isSentinel(v.infrastructure)));
console.log(`says nothing about infrastructure on any tuple: ${infraSentinel.length} of ${withVantage.length} (${pct(infraSentinel.length, withVantage.length)})`);
console.log(`\nlocation strings the fold could not map (residue, printed in full):`);
if (residue.size === 0) console.log('  (none)');
for (const [k, n] of [...residue.entries()].sort((a, b) => b[1] - a[1])) console.log(`  ${n}x  ${JSON.stringify(k)}`);
 
// A control vantage in the paper's own words. Tight requires "control" next to a
// vantage word; loose accepts either. Tight must be <= loose or the probe is broken.
const CONTROL_TIGHT = /control (vantage|VP|vantage point|machine|node|probe|server|host|site|resolver|network|location|measurement|group|country|client)s?|(vantage point|VP|machine|node|probe|server|host|client)s? (outside|located outside)[^.]{0,40}(censor|country|China|Iran|Russia|blocked)|uncensored (control|vantage|network|location)|non-?censor(ed|ing) (vantage|control|network|country|location)/i;
// Built from CONTROL_TIGHT for the same reason as WALL_LOOSE: the first version
// missed 6 of the 32 tight hits (the "uncensored control" and "vantage point
// outside <country>" branches contain no literal "control" near a vantage word),
// which made the page's "the gap between 32 and 41" sentence describe a
// difference between two DIFFERENT questions rather than two widths of one.
const CONTROL_LOOSE = new RegExp(
  CONTROL_TIGHT.source +
    '|\\bcontrol\\b[^.]{0,60}(vantage|VP|measurement|probe|machine|node|host|country|network)' +
    '|(vantage|VP|probe|machine|node|host)[^.]{0,60}\\bcontrol\\b',
  'i',
);
const ctrlTight = BLOCK.filter((p) => { const t = ftPop(p); return t !== null && CONTROL_TIGHT.test(t); });
const ctrlLoose = BLOCK.filter((p) => { const t = ftPop(p); return t !== null && CONTROL_LOOSE.test(t); });
const ctrlEscapes = ctrlTight.filter((p) => !ctrlLoose.includes(p));
if (ctrlEscapes.length) {
  throw new Error(
    `${ctrlEscapes.length} tight control-probe hit(s) escape the loose probe, so "tight" and "loose" are not two widths of one question: ` +
      ctrlEscapes.map(KEY).join(', '),
  );
}
console.log(`\nsays "control vantage" (or equivalent) in the full text, tight probe: ${ctrlTight.length} of ${BLOCK.length} (${pct(ctrlTight.length, BLOCK.length)})`);
console.log(`same, loose probe (any "control" near a vantage word): ${ctrlLoose.length} of ${BLOCK.length} (${pct(ctrlLoose.length, BLOCK.length)})`);
console.log('papers matching the tight control probe:');
for (const p of ctrlTight) console.log(`  ${KEY(p)}`);
 
H('5. What "blocked" is operationalised as');
// Signal families, folded from the paper's own text. These are what the paper
// checks to call something blocked. Multi-valued: most papers use several.
const SIGNAL = {
  'status code / HTTP error': /\b(HTTP )?(403|451|404|503)\b|status code|forbidden response|error (code|page)/i,
  'block page fingerprint or keyword': /block ?page|blocking page|blockpage (signature|fingerprint)|block-page/i,
  'page similarity / length outlier': /(page|response|body) (length|size)[^.]{0,40}(outlier|threshold|differ|distribution)|(cosine|jaccard|TF-?IDF|term frequency)[^.]{0,60}(similar|page|response)|similarity (score|threshold)[^.]{0,40}(page|response)/i,
  'DNS answer consistency': /(consistent|consistency|inconsisten\w+)[^.]{0,60}(DNS|resolution|response|answer)|(DNS|resolution)[^.]{0,60}(consistency|inconsistency) (metric|check|heuristic)|forged (DNS )?respons/i,
  'TLS certificate check': /(certificate|TLS)[^.]{0,60}(verif|validat|mismatch|chain)[^.]{0,60}(manipulat|block|censor|inject)|(manipulat|block|censor)[^.]{0,60}(certificate|TLS)[^.]{0,40}(verif|validat|mismatch)/i,
  'injected RST / connection reset': /\b(TCP )?RST\b|reset (packet|injection)|injected reset/i,
  'control-vantage comparison': CONTROL_TIGHT,
  'repeated probe / retry': /(repeat|retr(y|ies|ied)|re-?probe|multiple (probes|trials|measurements))[^.]{0,60}(confirm|verif|block|censor|persist)/i,
  'manual or browser validation': /manual(ly)? (verif|validat|check|inspect|review)[^.]{0,80}(block|censor|403|page)|(verif|validat)\w+[^.]{0,40}in a (real )?(web )?browser/i,
  'supervised classifier or clustering': /\b(classifier|random forest|logistic regression|SVM|neural network|clustering)\b[^.]{0,70}(block ?page|censor|blocked)/i,
  // \b-anchored: an unanchored BERT matches the surname Deibert, and the row read
  // 3 papers before that was fixed. See the widened probe below for the claim.
  'LLM or transformer model': /\b(LLMs?|large language model|GPT-?[0-9]|ChatGPT|Claude|Gemini|Llama|DeBERTa|BERTopic|BERT)\b[^.]{0,80}(block ?page|censor|blocked|classif)/i,
};
const sigPapers = {};
for (const [name, re] of Object.entries(SIGNAL)) {
  sigPapers[name] = BLOCK.filter((p) => { const t = ftPop(p); return t !== null && re.test(t); });
}
table(['signal the paper checks', `papers of ${BLOCK.length}`, 'share', 'first', 'last', '2010-19', '2020-24', '2025-26'],
  Object.entries(sigPapers).sort((a, b) => b[1].length - a[1].length).map(([n, ps]) => [
    n, ps.length, pct(ps.length, BLOCK.length),
    ps.length ? Math.min(...ps.map((p) => p.year)) : '-', ps.length ? Math.max(...ps.map((p) => p.year)) : '-',
    ps.filter((p) => p.year <= 2019).length, ps.filter((p) => p.year >= 2020 && p.year <= 2024).length, ps.filter((p) => p.year >= 2025).length]));
console.log('NOTE: these are full-text probes over the population, so each is an UPPER bound on');
console.log('papers that use the signal (a sentence in related work counts) and a LOWER bound on');
console.log('the idea being present (a paper can compare against a control without the word).');
const noSignal = BLOCK.filter((p) => !Object.values(sigPapers).some((ps) => ps.includes(p)));
console.log(`papers in the population matching none of the ${Object.keys(SIGNAL).length} signal probes: ${noSignal.length} — ${noSignal.map(KEY).join(', ') || 'none'}`);
const perPaper = BLOCK.map((p) => Object.values(sigPapers).filter((ps) => ps.includes(p)).length).sort((a, b) => a - b);
console.log(`signals matched per paper: min ${perPaper[0]}, median ${perPaper[Math.floor(perPaper.length / 2)]}, max ${perPaper.at(-1)}; ${perPaper.filter((n) => n >= 3).length} of ${BLOCK.length} match three or more`);
 
// The LLM row above is 3 papers, and reading them shows all three are about LLM
// SERVICES rather than an LLM used as the classifier. The widened probe (defined
// with the corpus regexes) tests that claim against the whole corpus.
const llmWide = rows.filter((p) => S(p) !== null && S(p).llmWide);
console.log(`\nwidened LLM/transformer-near-blocking probe, corpus-wide: ${llmWide.length} of ${rows.length}; in the population: ${llmWide.filter((p) => BLOCK.includes(p)).length} of ${BLOCK.length}`);
for (const p of llmWide.sort((a, b) => a.year - b.year)) console.log(`  ${p.year} ${p.venue} ${BLOCK.includes(p) ? '[in population]' : '[outside]'} ${p.title.slice(0, 85)}`);
 
H('6. detection[] tuples: the measured results, with their own denominators');
const detTuples = [];
for (const p of BLOCK) {
  for (const t of p.detection) {
    if (isSentinel(t.prevalence) || !t.prevalence) continue;
    detTuples.push({ p, t });
  }
}
console.log(`${detTuples.length} detection tuples with a stated prevalence, across ${new Set(detTuples.map((d) => KEY(d.p))).size} of ${BLOCK.length} papers in the population`);
const NUM = /\d/;
console.log(`of those, ${detTuples.filter((d) => NUM.test(d.t.prevalence)).length} carry a digit (i.e. could be a publishable figure)`);
console.log('\nevery prevalence tuple for the papers the page quotes:');
const QUOTED = [
  '2018/IMC/403-forbidden-a-global-view-of-cdn-geoblocking',
  '2014/IMC/automated-detection-and-fingerprinting-of-censorship-block-pages',
  '2024/USENIX/digital-discrimination-of-users-in-sanctioned-states-the-case-of-the-cuba-embarg',
  '2023/PETS/certainty-detecting-dns-manipulation-at-scale-using-tls-certificates',
  '2024/USENIX/gfweb-measuring-the-great-firewalls-web-censorship-at-scale',
  '2020/IEEE-SP/iclab-a-global-longitudinal-internet-censorship-measurement-platform',
  '2022/USENIX/a-large-scale-investigation-into-geodifferences-in-mobile-apps',
  '2024/PETS/automatic-generation-of-web-censorship-probe-lists',
  '2026/PETS/banned-books-analysis-of-censorship-on-amazon-com',
  '2024/WWW/a-worldwide-view-on-the-reachability-of-encrypted-dns-services',
  '2024/PETS/what-to-expect-when-you-re-accessing-an-exploration-of-user-privacy-rights-in-pe',
  '2022/USENIX/the-security-lottery-measuring-client-side-web-security-inconsistencies',
];
for (const k of QUOTED) {
  const p = byKey.get(k);
  if (!p) throw new Error(`QUOTED names a paper that is not in the corpus: ${k}`);
  if (!BLOCK.includes(p)) throw new Error(`QUOTED names a paper outside the population: ${k}`);
  console.log(`\n--- ${k}`);
  for (const t of p.detection) {
    if (!t.prevalence || isSentinel(t.prevalence)) continue;
    console.log(`    phenomenon: ${t.phenomenon}`);
    console.log(`    technique : ${t.technique}`);
    console.log(`    metric    : ${t.metric}`);
    console.log(`    prevalence: ${t.prevalence}`);
    console.log(`    quote [${t.evidence.section}]: ${t.evidence.quote}`);
  }
}
 
H('7. Instruments and observatories: who is used, and who is only cited');
// tools[] under-reports a dataset: the schema sees an instrument, not a corpus.
// So each name is counted twice, structurally and over the full text (one pass, above).
table(['instrument or dataset', 'named in tools[] as used/produced', 'papers whose full text names it (corpus-wide)', 'of which in the population'],
  Object.entries(INSTRUMENTS).map(([n, re]) => {
    const structural = rows.filter((p) => p.tools.some((t) => (t.usedOrMentioned === 'used' || t.usedOrMentioned === 'produced') && re.test(String(t.name ?? '')))).length;
    const fulltext = rows.filter((p) => S(p) !== null && S(p).instr.has(n));
    return [n, structural, fulltext.length, fulltext.filter((p) => BLOCK.includes(p)).length];
  }));
console.log('CAUTION: the full-text column is a mention count and several of these names are');
console.log('homographs — "Satellite" is satellite Internet, "Geneva" a city, "ONI" a substring,');
console.log('"Iris" an iOS analyser, "Encore" and "Augur" ordinary words. Use the column to show');
console.log('that tools[] under-counts a dataset, never as a usage figure.');
 
H('8. The jurisdiction slice: GDPR and CCPA walls');
// The two wall probes are defined with the other whole-corpus regexes above and
// evaluated in the same single pass.
const wt = rows.filter((p) => S(p) !== null && S(p).wallTight);
const wl = rows.filter((p) => S(p) !== null && S(p).wallLoose);
const wallEscapes = wt.filter((p) => !wl.includes(p));
if (wallEscapes.length) {
  throw new Error(
    `${wallEscapes.length} tight wall-probe hit(s) are not matched by the loose probe, so the two are different questions rather than two widths of one: ` +
      wallEscapes.map(KEY).join(', '),
  );
}
console.log(`corpus-wide, tight probe: ${wt.length} papers; loose probe: ${wl.length} papers (denominator ${rows.length})`);
console.log(`the loose probe contains the tight one by construction, so the distinct union is ${new Set([...wt, ...wl]).size} papers and every one of them was read`);
console.log('tight-probe hits, all of them, with the hand verdict:');
for (const p of wt.sort((a, b) => a.year - b.year)) console.log(`  ${KEY(p)}  [${(VERDICT[KEY(p)] ?? ['NOT A CANDIDATE']).join('+')}]`);
console.log('loose-probe hits that the tight probe missed:');
for (const p of wl.filter((p) => !wt.includes(p)).sort((a, b) => a.year - b.year)) console.log(`  ${KEY(p)}  [${(VERDICT[KEY(p)] ?? ['not a candidate']).join('+')}]  ${p.title.slice(0, 80)}`);
console.log(`\nlegal population for reference: ${rows.filter((p) => p.legal.length > 0).length} papers assess a law; of the population, ${BLOCK.filter((p) => p.legal.length > 0).length} do`);
const gdprLegal = rows.filter((p) => p.legal.some((l) => /GDPR/i.test(String(l.law ?? ''))));
console.log(`papers whose legal[] names GDPR: ${gdprLegal.length}; of those, in the population: ${gdprLegal.filter((p) => BLOCK.includes(p)).length}`);
 
H('9. Cross-checks against neighbouring pages');
console.log(`design:existing_datasets reports a "Censorship list (Citizen Lab, OONI)" family of 17 papers inside ITS OWN reuse population. That population is defined on that page and is not re-derived here — the two definitions below both differ from it, which is exactly why the page cites the family and not the denominator.`);
// That page's population is the UNION of the studyTypes flag and the temporal
// mode, not the studyTypes flag alone; using the wrong one prints 2,615 where the
// page says 3,389 and makes a shared figure look stale.
const reuse = rows.filter((p) => p.studyTypes.includes('existing-dataset-analysis') || p.temporal.some((t) => t.mode === 'existing-dataset'));
console.log(`  reuse population, that page's definition (studyTypes OR temporal.mode): ${reuse.length}`);
console.log(`  studyTypes flag alone: ${rows.filter((p) => p.studyTypes.includes('existing-dataset-analysis')).length}`);
console.log(`  of the population, in the reuse population: ${BLOCK.filter((p) => reuse.includes(p)).length}`);
console.log(`design:crawling_location's population is the ${rows.filter((p) => p.vantage.length > 0).length} papers with a vantage tuple; ${BLOCK.filter((p) => p.vantage.length > 0).length} of the population are in it.`);
console.log(`programming:crawler_detection owns "blocked because you look like a bot": ${inFamily('off:bot-blocking').length} candidate(s) were handed to it.`);
 
H('10. Probe recall, and the queued estimate');
// Two claims on the page and its provenance need these numbers: how much of the
// population a title probe alone would find, and what the roadmap's queued
// estimate re-derives to on the current corpus. The queued regex is reproduced
// here EXACTLY as the 2026-09-02 brainstorm pass ran it.
const titleHits = rows.filter(PROBES.title);
console.log(`probe 1 (title+summary) hits: ${titleHits.length}; of those in the population: ${titleHits.filter((p) => BLOCK.includes(p)).length}; excluded: ${titleHits.filter((p) => !BLOCK.includes(p)).length} (${pct(titleHits.filter((p) => !BLOCK.includes(p)).length, titleHits.length)} of the probe's hits)`);
console.log(`population papers NO title probe caught: ${BLOCK.filter((p) => !PROBES.title(p)).length} of ${BLOCK.length}`);
const QUEUED_RE = /censorship|censored|OONI|Censored Planet|GFW|geoblock|blockpage|internet shutdown/i;
const q = rows.filter((p) => QUEUED_RE.test(T(p)));
console.log(`the roadmap's queued probe, re-run on this corpus: ${q.length} papers (it reported 74 on 2026-09-02), web ${q.filter((p) => p.platforms.includes('web')).length}, 2020-2023 ${q.filter((p) => p.year >= 2020 && p.year <= 2023).length}, 2024-2026 ${q.filter((p) => p.year >= 2024).length}`);
console.log(`of those ${q.length}, in the population: ${q.filter((p) => BLOCK.includes(p)).length}; excluded: ${q.filter((p) => !BLOCK.includes(p)).length}`);
 
// A diagnostic the provenance page quotes: how unusable detection[].phenomenon is
// as a COUNT in this area, as opposed to as a way to find candidates.
const PHEN = /censor|geo-?block|geo-?differen|block ?page|internet (filter|shutdown)|dns (manipulat|censor|interference|redirection|tamper)|network interference|connection tamper|url[- ]filter|keyword[- ]?(based )?filter|content (filter|censor)|over-?block|site blocking|blocked (site|domain|web|content|url|categor)|throttling of|traffic throttl|sanction/i;
const phenStrings = new Set();
const phenPapers = new Set();
for (const p of rows) {
  for (const t of p.detection) {
    if (PHEN.test(String(t.phenomenon ?? ''))) { phenStrings.add(t.phenomenon); phenPapers.add(KEY(p)); }
  }
}
console.log(`\ndetection[].phenomenon, raw: ${phenStrings.size} distinct blocking-ish strings across ${phenPapers.size} papers — used to FIND candidates (probe 2) and never to count`);
 
H('11. Venue-years with no extracted papers at all');
// A review pass observed that Khattak et al. (NDSS 2016) is cited by this page
// and cannot be in the population. It is not a selection decision: NDSS 2016 is
// simply absent from the extraction, and so are several other venue-years. Any
// "the corpus does not contain X" claim has to be read against this list.
{
  const have = new Set(rows.map((p) => `${p.venue}/${p.year}`));
  const venues = [...new Set(rows.map((p) => p.venue))].sort();
  const gaps = [];
  for (const v of venues) for (let y = 2010; y <= 2026; y++) if (!have.has(`${v}/${y}`)) gaps.push(`${v} ${y}`);
  console.log(`venue-years with zero extracted papers: ${gaps.length} — ${gaps.join(', ')}`);
  const popYears = new Set(BLOCK.map((p) => `${p.venue}/${p.year}`));
  console.log(`(the population itself spans ${popYears.size} distinct venue-years)`);
}