Differences

This shows you the differences between two versions of the page.

--- provenance:programming:tranco [2026/09/17 07:24] – Markup rendering sweep: wrap escapes, CLI flag typography, and heading quotes. Authored by Claude. karel.kubicek.claude
+++ provenance:programming:tranco [2026/09/21 13:59] (current) – Quote-check refresh 2026-09-21: re-ran report_tranco.mjs; 58 below threshold -> 39 rescued from the PDF + 19 below threshold in both. Authored by Claude karel.kubicek.claude
@@ Line 63: / Line 63: @@
 ===== Evidence quotes checked =====
-Tranco-tuple ''evidence.quote'' values against ''paper.cols.txt'', whitespace- and hyphen-normalised, then 5-word windows at 60%: **204 exact, 111 partial, 58 below threshold, 0 missing quotes, 0 missing full text**.
+Tranco-tuple ''evidence.quote'' values against the rendering the extractor read, whitespace- and hyphen-normalised, then 5-word windows at 60%, and then — for anything below threshold — against an independent ''pypdf'' rendering of the same ''paper.pdf''. **Re-run 2026-09-21**, see //Quote-check refresh, 2026-09-21// at the foot of this page: **204 exact, 111 partial, 39 rescued from the PDF, 19 FAILED in both renderings, 0 missing quotes, 0 missing full text**.
-The first 15 FAILED rows were read. They are column splices, dropped citation markers, or quotes that name the sample without repeating the word "Tranco" next to the numbers — the same class ''quote_check.mjs'' warns about. None looked fabricated. They stay in the report. No FAILED quote was used on the content page.
+**The old figure was 58 below threshold, and 39 of those 58 are a defect in the stored text rather than in the extraction.** 19 are below threshold in both renderings — **5.1%** of 373, not 15.5%.
+**The 15 rows that were read by hand are no longer the 15 the report prints.** The report lists the first 15 FAILED rows in key order and the rescues reshuffled them: **6 of the 15 originally read are still FAILED, 9 were rescued, and 9 rows now printed have not been read.** Of the 15 that were read: they are column splices, dropped citation markers, or quotes that name the sample without repeating the word "Tranco" next to the numbers — the same class ''quote_check.mjs'' warns about. None looked fabricated. They stay in the report. No FAILED quote was used on the content page.
 ===== External and industry verification =====
@@ Line 446: / Line 448: @@
 . QUOTE CHECK — population evidence.quote on Tranco tuples
 ======================================================================
-exact: 204  partial(>=60% 5-word windows): 111  FAILED: 58
+exact: 204  partial(>=60% 5-word windows): 111  rescued from PDF: 39  FAILED in both renderings: 19
 no quote on tuple: 0  missing .cols: 0
 FAILED (first 15) — read by hand; below-threshold is not automatically unsupported:
-  IMC/2020/accept-the-risk-and-continue-measuring-the-long-tail-of-government-https-adoptio  6/15  "We compare https in these top government websites ... with [1] 12,000 random, uniformly sampled top million non-government hostnames"
   WWW/2020/understanding-the-performance-costs-and-benefits-of-privacy-focused-browser-exte  11/21  "Specifically, we use (i) the top 1,000 domains ... (ii) the domains 99,001 to including 100,000 ... Correspondingly, our list of domains contains 2,000 domains."
-  IMC/2021/knock-and-talk-investigating-local-network-communications-on-websites  2/6  "one using the Tranco snapshot taken on June 3, 2020"
-  PETS/2021/ml-cb-machine-learning-canvas-block  11/20  "Initially, we targeted 2,200 websites, using our original dataset, original Alexa Top Rank list, and the Tranco list of the one million top sites"
-  PETS/2021/privacy-preference-signals-past-present-and-future  3/6  "we crawled the top 100k entries from the Tranco toplist"
   PETS/2021/the-cname-of-the-game-large-scale-analysis-of-dns-based-tracking-evasion  6/12  "For each website, we visited up to 20 pages ... on the Tranco top 10,000 websites"
   PETS/2021/deja-vu-abusing-browser-cache-headers-to-identify-and-track-online-users  8/15  "we crawled the Top 100K websites from the Tranco list [33] on the 2nd and the 3rd of June"
-  USENIX/2021/fingerprinting-in-style-detecting-browser-extensions-via-injected-style-sheets  6/11  "we crawled the homepage of the Tranco top 200 websites with and without our countermeasure."
-  USENIX/2021/how-great-is-the-great-firewall-measuring-chinas-dns-censorship  10/21  "We complement our test list by including domains from the Citizen Lab test lists [13], the Tranco list [66], and the Common Crawl project [14]."
-  USENIX/2021/the-circle-of-life-a-large-scale-study-of-the-iot-malware-lifecycle  4/9  "We use Tranco's top site ranking [53] to identify and filter benign domains."
   IMC/2022/a-first-look-at-the-name-resolution-latency-on-handshake  0/3  "extracting unique top-100 TLDs from the latter"
-  IMC/2022/html-violations-and-where-to-find-them-a-longitudinal-analysis-of-specification  3/7  "The result is an overall top list of 24,915 unique domains."
   IMC/2022/toppling-top-lists-evaluating-the-accuracy-of-popular-website-lists  5/17  "We retrieved daily snapshots of Alexa, Umbrella, and Majestic ... Tranco rankings ... and Secrank rankings ... for February 1-28, 2022."
-  PETS/2022/trace-oddity-methodologies-for-data-driven-traffic-analysis-on-tor  9/17  "The first two datasets, SD and MD consist of visits to the 15,000 most popular websites according to the Tranco ranking"
   USENIX/2022/automating-cookie-consent-and-gdpr-violation-detection  8/16  "Our primary source is the Tranco ranking of May 5th, 2021, which lists domains ranked by their estimated worldwide popularity."
+  IMC/2023/does-it-spin-on-the-adoption-and-use-of-quics-spin-bit  12/32  "Relying on different toplists, we first gather a target set that focuses on popular domains: (i) Alexa Top 1M, (ii) Cisco Umbrella, (iii) Majestic Million, and (iv) the Tranco Res
+  USENIX/2023/automated-cookie-notice-analysis-and-enforcement  6/11  "We use the top-100k websites from the Tranco list generated on 21 March 2022 [37]."
+  WWW/2023/measuring-and-evading-turkmenistans-internet-censorship-a-case-study-in-large-sc  13/22  "As shown in Figure 2, the payload of our probes contains domains curated from the Citizen Lab lists, the full Tranco list, and Common Crawl Project."
+  NDSS/2024/untangle-multi-layer-web-server-fingerprinting  12/21  "We seed our experiment with the Tranco top 10K domains generated on 22 October 2023 ... This results in a final list of 7528 sites."
+  PETS/2024/interest-disclosing-mechanisms-for-advertising-are-privacy-exposing-not-preservi  5/9  "the top 1M most visited eTLD+1 from Tranco classified with the Topics API"
+  IMC/2024/exploring-the-ecosystem-of-dns-https-resource-records-an-end-to-end-perspective  7/12  "The first part, spanning from May 8th to July 31st, 2023, includes 634,810 unique overlapping domains."
+  USENIX/2024/cross-the-zone-toward-a-covert-domain-hijacking-via-shared-dns-infrastructure  0/5  "981 are ranked within the Tranco top 1M [45]."
+  WWW/2024/panoptichrome-a-modern-in-browser-taint-analysis-framework  2/9  "we collected the API logs using VisibleV8[27] for the Tranco top 1000 sites."
+  USENIX/2025/irblock-a-large-scale-measurement-study-of-the-great-firewall-of-iran  7/25  "These domain test lists are collected from various sources, including top-level domains (TLD) zone files, the Citizen Lab test lists (CLTL), the Tranco list, and the Common Crawl
 ======================================================================
@@ Line 506: / Line 508: @@
   homepage still says "all four rankings" (1 occurrence) and also "five providers"
   /download/<id>/full was ~104 MB; not treated as equal to the daily 1M; row count not published
 </file>
@@ Line 604: / Line 605: @@
 Mechanical rendering repair only: a fresh live raw/XHTML export of 188 pages was checked with ''check_wrap.mjs'' and ''check_typography.mjs''. Affected plugin tags, CLI flags and heading markup were repaired; no figures or substantive prose were changed. The resulting source and rendered DOM were re-checked after saving.
+===== Currency correction, 2026-09-17 =====
+The missing pre-swap row on [[programming:tranco]] was fetched from Tranco's dated
+API, with the adjacent dates used as controls. These are the exact URLs fetched:
+<code>
+https://tranco-list.eu/api/lists/date/20230730
+https://tranco-list.eu/api/lists/date/20230731
+https://tranco-list.eu/api/lists/date/20230801
+https://tranco-list.eu/api/lists/date/20230802
+</code>
+^ Date ^ List id ^ Providers ^ Prefix ^
+| 2023-07-30 | ''99P42'' | Alexa, Umbrella, Majestic, Farsight | full |
+| 2023-07-31 | ''W9ZN9'' | Alexa, Umbrella, Majestic, Farsight | full |
+| 2023-08-01 | ''25299'' | CrUX, Farsight, Majestic, Radar, Umbrella | 1,000,000 |
+| 2023-08-02 | ''8289V'' | CrUX, Farsight, Majestic, Radar, Umbrella | 1,000,000 |
+The 31 July object therefore supplies the published ''W9ZN9'' row: it is the last
+pre-swap daily list, with the old four-provider configuration and the ''full''
+prefix. The controls show a single-day step to the CrUX/Radar configuration and
+the one-million prefix on 1 August.
+===== Review layer, 2026-09-17 =====
+One shared ''model: sonnet'' external-currency pass ran after publication. It
+re-fetched the dated API and **passed** the published ''2023-07-31 → W9ZN9''
+row, including its four providers and ''full'' prefix. Its repeat fetches of the
+July and 2 August controls were HTTP 429 rate-limited, so it raised no new
+claim from those controls and required no correction to the row.
+===== Quote-check refresh, 2026-09-21 =====
+The 2026-09-04 ''cols''-vs-PDF audit on [[:provenance:literature:corpus]] showed that 73.1% of evidence quotes that cannot be located in ''paper.cols.txt'' **are** present in an independent ''pypdf'' rendering of the same ''paper.pdf''. ''report_tranco.mjs'' already carried the fallback in its quote check; what was stale was this page's copy of its output. The ''%%<file>%%'' block above was replaced by re-running the report, not hand-edited.
+^ Figure ^ Was ^ Is ^ Why ^
+| Tranco-tuple quotes | 373 | 373 | population unchanged |
+| exact | 204 | 204 | unchanged |
+| partial (≥60% of 5-word windows) | 111 | 111 | unchanged |
+| rescued from the PDF | — | **39** | these were inside the old 58 |
+| FAILED | **58** | **19** (below threshold in both) | 58 = 39 + 19 exactly |
+| FAILED share of 373 | 15.5% | **5.1%** | 19 of 373 |
+**Everything else in the report reproduced byte-for-byte.** A line-by-line diff of the published block against the fresh run differs **only** in the quote-check summary line and the printed FAILED list. No Tranco figure, sampling-method count, round-prefix figure or external claim on this page or on [[:programming:tranco]] moved.
+**The hand-read set drifted out from under the printed list**, exactly as on [[:provenance:artifacts]], which was refreshed in the same pass. 6 of the 15 read rows are still FAILED; 9 rows now printed have never been read. The prose above says so rather than leaving "the first 15 FAILED rows were read" beside a list where it is no longer true.
+^ Item ^ Value ^
+| Date | 2026-09-21, unsupervised |
+| Command | ''%%node scripts/report_tranco.mjs%%'' |
+| Artifact | ''out/qc0921b/rep_tranco.txt'' |
+| Script changes | none |
+| Reviewers | one ''sonnet'' figures-vs-script pass |
+| Pages saved | this page only — [[:programming:tranco]] publishes none of these counts |
+| Open | the 9 newly-printed FAILED rows, and the 19 as a whole, have not been read by hand |