Working notes behind Ownership resolution. Every query with its population and denominator, the report script and its unedited output, what was folded and what the fold could not reach, the quotes checked against the source, the external sources verified and rejected, and the judgement calls. Corpus-wide selection and extraction caveats are on Corpus and are not restated here.

This page is a carve-out, and most of its audit trail lives elsewhere on purpose. The content page was split out of webXray on 2026-09-11. The scripts that produce its live-database figures, their unedited output, the 175-row adjudication table and the inter-rater material were published in 2026-08 and 2026-09 under the webXray ids and keep those ids, because they are cited from the published record and moving them would break links for no gain:

No ~~DISCUSSION~~ block: comments belong on the content page. Citations use the same {[citekey]} keys and the same shared Bibliography; this page adds no bibliography entries of its own.

This reverses a recorded decision, so the reversal is recorded too. webxray §“Scope decision” says, on 2026-08-17:

A separate design:ownership_resolution page with webXray as a stub. Rejected: it would leave the wiki's existing red link pointing at a stub, and the material does not split cleanly — webXray's frozen list is the best available illustration of what goes wrong.

Both halves of that reasoning have since stopped holding:

• “webXray as a stub” was the wrong alternative to compare against. There is enough webXray-specific material for a real page — the three-licence history across four states, the availability census, the domain_owners.json schema, the frozen 2016 PSL, the Wayback timeline, and 15 corpus papers with a hand role map. What is left after the carve-out is 27 KB, not a stub.
• The red-link problem inverted. On 2026-09-07 the Roadmap queued design:ownership_resolution as a promised page and scripts/sitemap.mjs began gating on it, so from that date not writing it was the dangling promise. The id was fixed by that decision and is used unchanged.
• The discoverability argument is the item's whole point and was never answered. A reader asking “how do I attribute a third-party domain to a company?” does not search for a dead tool. Both earlier sittings recorded this as the right thing to do; neither did it.

Alternatives considered this sitting and rejected:

What moved, what stayed, and what had to be repointed. Nothing was re-derived in the move.

Day-one drift, found and fixed rather than left:

A site-wide inbound-link sweep found four more, on pages nobody would have thought to check. Grepping every cached page for a link to programming:crawler:webxray rather than only the pages this sitting edited:

Two of those four used the link text “webXray and domain-to-company ownership”, i.e. they were already treating the webXray page as the ownership page — which is the item's complaint, restated by the wiki itself. A carve-out's drift is not confined to the pages you edited; sweep for inbound links before calling it done. One inbound reference was deliberately left: provenance:design's generated page inventory records a size and a description as of an earlier date, and it is a dated snapshot rather than a live pointer.

All over data/extract/run1/extractions.jsonl (5,859 papers) and data/fulltext/<year>/<venue>/<slug>/paper.cols.txt, whitespace collapsed before matching so a term broken across a PDF column boundary still matches. Every count is a paper count. The script is scripts/report_ownership_resolution.mjs, published in full in section J with its unedited output in section K.

A narrowing probe that returns more hits than the loose one, or hits the loose one does not contain, is a broken probe. The script asserts both and throws rather than printing. Both widths are on the record because the gap is itself the claim:

Resource                 Tight probe  Loose probe  Status
-----------------------  -----------  -----------  ----------------------------------------------------------------
webXray                  15           233          hand-verified: all 15, in report_webxray.mjs
Tracker Radar            32           108          hand-verified: all 32, in report_webxray.mjs
WhoTracks.me             23           124          upper bound
Crunchbase               23           23           upper bound
Disconnect (list sense)  74           700          upper bound; the loose width is the reason this one is tightened
Public Suffix List       101          157          upper bound

The patterns themselves are in a code block and not in a table, because a DokuWiki table cell cannot hold a pipe and every one of these regexes is an alternation. \| is not an escape and wrapping the pattern in DokuWiki's nowiki delimiters does not help either: the cell splits regardless.

webXray             tight  /webx[\s-]?ray/i
                    loose  /webx[\s-]?ray|libert/i
                    15 -> 233. "Libert" is a common surname and a French word.
                    The loose width is a sanity bound, not a candidate set.

Tracker Radar       tight  /tracker[\s.-]?radar/i
                    loose  /tracker[\s.-]?radar|duckduckgo/i
                    32 -> 108. Most DuckDuckGo mentions are the search engine
                    or the browser, not the dataset.

WhoTracks.me        tight  /whotracks/i
                    loose  /whotracks|who ?tracks ?\.? ?me|ghostery/i
                    23 -> 124. Ghostery is mostly the extension.

Crunchbase          tight  /crunchbase/i
                    loose  /crunchbase|crunch base/i
                    23 -> 23. NO GAP: nobody spells it with a space. This
                    row's count is not an artefact of the pattern.

Disconnect          tight  /disconnect'?s? (entit|list|block|black|tracking)|disconnect\.me|entities\.json/i
                    loose  /disconnect/i
                    74 -> 700. "Disconnect" is an ordinary English word and a
                    CSP-literature technical term. This is the probe that
                    needed tightening, and the published row is the tight one.

Public Suffix List  tight  /public suffix/i
                    loose  /public suffix|etld\+?1|effective top-?level domain/i
                    101 -> 157. The extra 56 are papers that use the concept
                    without naming the list.

What this does not establish. A tight probe with no gap is not a probe with full recall — it is a probe whose count is stable under widening in the one direction tried. A paper that resolves domain ownership from a source none of these five names, or names none of them at all, is invisible to every width. See section F.

Seven papers whose tools[] / classification[] / otherToolsMentioned[] name one of the five resources but whose full text does not match the tight sweep. Printed rather than dropped, because an invisible residue is a residue nobody looks at:

  IEEE-SP/2024/to-auth-or-not-to-auth-a-comparative-analysis-of-the-pre-and-post-login-security
      named: Disconnect Tracker Protection List
  PETS/2022/on-dark-patterns-and-manipulation-of-website-publishers-by-cmps
      named: Disconnect | Disconnect tracking filter list
  PETS/2024/generalizable-active-privacy-choice-designing-a-graphical-user-interface-for-glo
      named: Disconnect Tracker Protection lists
  PETS/2024/support-personas-a-concept-for-tailored-support-of-users-of-privacy-enhancing-te
      named: Disconnect
  USENIX/2019/canvas-fast-and-inexpensive-automotive-network-mapping
      named: physical ECU access and disconnection
  WWW/2018/the-cost-of-digital-advertisement-comparing-user-and-advertiser-views
      named: Disconnect | Disconnect blacklist
  WWW/2023/who-funds-misinformation-a-systematic-analysis-of-the-ad-related-profit-routines
      named: Disconnect

Six of the seven are Disconnect, and they are the tightened pattern's misses: the extractor wrote “Disconnect Tracker Protection List” or bare “Disconnect” into a field, where the paper's own prose says something the list-sense pattern does not match. The seventh is a homonym the extractor invented — “physical ECU access and disconnection” in an automotive CAN-bus paper is not the Disconnect list. The tight pattern's cost is therefore about five papers, all Disconnect, all in the direction of undercounting, and the page's Disconnect row should be read as 74 rather than 74-to-79 only because the union is an upper bound on use anyway. The alternative — publishing the loose 700 — is not a trade worth making.

Every script whose figures the new page inherits was re-run before the move, against the same pinned inputs, and diffed against its committed output. A page whose report script no longer runs is a page whose numbers cannot be refreshed.

A drift check the carve-out made cheap. The new script re-derives the sweep counts from patterns written out independently and then cross-checks nine figures against report_webxray-output.txt. All nine agree. That cross-check is inside the script and throws, so a future run cannot publish two pages that disagree with each other:

Figure                   this script  report_webxray.mjs  verdict
-----------------------  -----------  ------------------  -------
webXray                  15           15                  agree
Tracker Radar            32           32                  agree
WhoTracks.me             23           23                  agree
Crunchbase               23           23                  agree
Disconnect (list sense)  74           74                  agree
Public Suffix List       101          101                 agree
union of the five        136          136                 agree
union ∩ crawled          107          107                 agree
crawled denominator      1120         1120                agree

This is the one substantive correction of the sitting, and it was published for three and a half weeks.

report_webxray.mjs printed, at two places:

  ${pct(ftPapers.length, crawled.length)} of the ${crawled.length} that ran a crawl
  ${pct(anyOwner.length, crawled.length)} of the ${crawled.length} that crawled

ftPapers and anyOwner are sweeps over the whole corpus. Dividing either by crawled.length pairs a corpus-wide numerator with the crawling denominator: the share is of a population the numerator was not drawn from.

Both figures were on the live webXray page, and the 12.1% was one of the figures moving to the new page — which is how it was caught: the new script computed the intersection because it had no reason not to, and the cross-check refused to pass. No guard on the old page could have found it. The number-guard family checks that a figure on a page matches the script's output; the script's output was the wrong number, and the page matched it faithfully.

The fix adds a named helper so the mistake cannot recur silently:

const crawledKeys = new Set(crawled.map(key));
// A share of `crawled` needs a numerator drawn from `crawled`.
const inCrawled = (ks) => ks.filter((k) => crawledKeys.has(k));

report_webxray-output.txt was regenerated and both pages corrected in the same sitting. The webXray page carries a footnote at the changed figure rather than silently restating it.

This page's corpus figures need no name fold: every count is a regex over full text or a set membership, and there are no free-text names being aggregated. That is stated rather than assumed, because it is unusual for a page on this wiki.

The live-list figures do fold, and the fold and its residue are published on webxray §B.11. Restated here only in summary, because the content page quotes both halves:

• Legal-form suffixes only (Inc, LLC, GmbH, S.A.S, …), plus punctuation normalisation. Deliberately no synonyms, so the “agree” columns are a lower bound on real agreement and the “neither” column an upper bound on real disagreement.
• Residue: the fold leaves 782 of webXray's 827 owner names untouched (94.6%). Of the 45 it changes, only 16 lose a legal-form suffix; the other 29 change on punctuation alone — 56.com, AT&T, Ask.com, Bootstrap_China, Clearstream.TV, Cm_browser, Dictionary.com, Dun & Bradstreet, and the rest are in the published output.
• What no fold reaches: “Google” against “Alphabet”, “Xandr” against “Microsoft Corporation”. Those are the granularity and vintage axes, and they are the page's subject rather than a normalisation failure. There is no principled synonym table for them, which is why the page reports the agreement figures as bounds and adjudicates a sample instead.

Checked on 2026-09-11 against data/fulltext/<year>/<venue>/<slug>/paper.cols.txt, whitespace collapsed and quotation marks and dashes normalised, with paper.norm.txt and paper.txt as fallbacks.

The Libert coverage quote is real but is not one contiguous string in any rendering of the PDF. It is the page's longest quotation and it is load-bearing for the coverage section, so it was chased rather than dropped. In paper.cols.txt, paper.norm.txt and paper.txt alike, the two-column layout interleaves a table caption through the sentence:

...because webxray's database of Table 1: Third-Party Prevalence, SSL Use, and First-Party
domain ownership primarily contains major ad networks rather Disclosure than small clients,
and policyxray only searches for identified †Denotes Company has Consumer Services parties,
variability in the long-tail of trackers may not have an outsized effect on overall findings
related to disclosure. Nonetheless, Company % Tracked % SSL % Disclosed it is important to
point out that the number of parties being searched Google † 82.81 80.35 38.29 for is fewer
than the total number of parties present.

Every clause is verbatim and in this order; the interleaved strings are “Table 1: Third-Party Prevalence, SSL Use, and First-Party Disclosure”, “†Denotes Company has Consumer Services” and the table's own header and first row. The page quotes the sentence as the author wrote it and carries a footnote saying exactly this, so a reader running the same grep does not conclude the quote is invented. This is the class of defect a plain substring check reports as a miss and a careless run reports as a fabrication.

One claim was corrected by reading the source. The old page said [5Yang, Zhiju; Yue, Chuan (2020): "A Comparative Measurement Study of Web Tracking on Mobile and Desktop Environments", in: Proceedings on Privacy Enhancing Technologies. (DOI)] “resolved organisations for 411 of 762 mobile-specific trackers using CrunchBase, webXray's list, TLS certificates and WHOIS in that order”. The paper's own sentence is narrower:

we successfully retrieved the organization information of 411 trackers from either the CrunchBase database, Tim Libert's library, or TLS certificate, and retrieved the organization […]

The 411 comes from the first three sources; WHOIS contributed a further 251 (“among those 251 trackers whose organization information was retrived from the WHOIS record”). The four sources were tried in that order, but the 411 is not a figure for all four. The new page states it as 411 from the first three and 251 from WHOIS. The consequence for the page's actual point is unchanged: the five registrar-privacy strings total 88, against the largest real company in the top-ten table at 48 (Adobe), and the paper says so itself — “these five organizations cannot represent the real organizations of those trackers”.

Repository and vendor state was re-fetched on 2026-09-11, not recalled. Training data is stale by construction for this kind of claim.

Sources rejected, recorded so the next run does not re-add them:

• Ghostery trackerdb is unmeasured. It is a live fourth option and the page gives numbers for two of the three live lists. Closing this needs a parser for the .eno + patterns model whose choices are checkable against the other three — a sitting's work, not a footnote. Filed as an open question on the content page.
• Recall of the candidate set is unknown. The 136 is a union of six patterns over full text. A paper that resolves domain ownership from a source none of the five names — a bespoke WHOIS pipeline, a commercial feed, an internal list — matches nothing at any width. The page reports 136 as an upper bound on use and says nothing about recall, because nothing here measures it.
• The share of the 136 that actually resolved an owner is not measured. Only two rows (webXray's 15 and Tracker Radar's 32) carry a hand verdict per paper. Doing the same for Disconnect's 74 is the obvious next increment and was not attempted this sitting.
• ~~No accuracy figure exists for the tail.~~ Partly closed on 2026-09-11, and the answer mattered. This read: “Every rate is conditional on the entry being adjudicable, and 18–30% of each list's draws were not. Whether unadjudicable domains differ systematically between lists is exactly what would decide whether Disconnect's lead survives, and a random sample cannot answer it.” A second pass with harder sourcing settled 28 of the 40, leaving 6.7% / 6.7% / 10.0%. The formal test found no significant difference between recovered and already-eligible rows (p = 0.64 / 1.00 / 0.40, on 7, 9 and 12 rows — almost no power), but Disconnect's estimate fell 94.4% → 89.2% and its lead over webXray stopped surviving Bonferroni. So the answer is: the tail was somewhat worse for the list the worry named, and the worry was worth acting on. See section L. What remains open is the residue's own residue — 12 domains that no register, no registry record and no archived legal page can reach, and nothing bounds those from inside the sample either.
• The coverage census has no neutral denominator. Tracker Radar's own crawl output is the frame, which flatters Tracker Radar. There is no independent census of third-party domains to use instead, and the page says so at every coverage figure rather than pretending otherwise.
• Whether a hierarchy built from a current list would beat both. webXray's parent_id tree is the only hierarchy any of these files carries and it is frozen; nobody has tried grafting a current corporate-structure source onto Tracker Radar's entities. Filed as an open question.
• Two adjudication rows from 2026-08-17 remain unresolved (the acquisition date behind 360yield.com, and fwmrm.net's post-2026-spinoff status). Recorded as unresolved rather than guessed; they are excluded from the 28.

The task brief for this sitting said Crawler's comparison row “still says webXray's crawler is PhantomJS (historically)”. It does not, and has not since 2026-08-17. The brief's account of its own state was stale; recorded here because a later sitting reading the brief would look for a correction that is already made.

The one correction that was owed, and that this sitting created, is the carve-out's own drift: that bullet ended “see webXray for what survives of it: the ownership database, and how it compares to Tracker Radar and Disconnect today”, and the comparison had just left that page. Repointed to Ownership resolution in revision listed below. Nothing else on Crawler was touched: its webXray figure of 7 is the tools[] schema count for a table whose whole column is schema counts, and changing it to the sweep's 15 would make that one cell incomparable with its neighbours.

All four were told explicitly that the author's context may not be exhaustive, and all four were handed the page text, the report script, its unedited output and these notes. The three focused passes ran in parallel; the generic pass ran afterwards, against the pages as the first three had left them.

All four passes returned and every finding was applied. The generic pass — the one with no checklist — found more, and worse, than the three focused passes combined. Each was handed the page text, scripts/report_ownership_resolution.mjs, its unedited output, scripts/report_webxray-output.txt and these notes, against the bundle in review_own/. Every finding below was re-verified by the author before being accepted — a reviewer's report is a lead, not a result.

The Choosing a resolution source now table read “emerging, and not yet at the domain layer”. The reviewer found arXiv:2606.20868, Can LLMs Reason About Brand Ownership?, submitted 2026-06-18 — four models evaluated on domain-to-brand attribution over 36 heavily-phished brands.

Verified independently before accepting: the abstract was fetched (HTTP 200) and read. The paper is real, the date is right, and the task genuinely is domain-layer ownership attribution. Two qualifications the reviewer did not make, and which changed how it went on the page:

• Its question is phishing and squatting defence — “is this domain the brand's own?” — not third-party tracker attribution. 36 brands is not a coverage claim.
• Its result cuts the other way. Models enumerate a brand's domains at up to 82% precision from memory, but on ownership verification without external tools macro F1 is at most 0.37, rising by up to 0.65 with WHOIS. So the honest update is not “LLMs are arriving, consider them” but “the one published attempt at this task fails without retrieval” — which strengthens the row's existing advice rather than softening it.

The corpus-scoped half of the sentence (“no corpus paper applies this to third-party domain ownership”) was not changed: it was and is true, and an arXiv preprint is outside the seven venues. Cited as a footnote with its arXiv id rather than a {[key]}, so no bibliography entry and no cache purge.

The page footnotes the [2Libert, Timothy (2018): "An Automated Approach to Auditing Disclosure of Third-Party Data Collection in Website Privacy Policies", in: Proceedings of the ACM Web Conference. (DOI)] coverage sentence as non-contiguous in every rendering. The reviewer found that [6Selmo, Carlos; Carisimo, Esteban; Bustamante, Fabián E.; Alvarez-Hamelin, J. Ignacio (2025): "Learning AS-to-Organization Mappings with Borges", in: Proceedings of the 2025 ACM Internet Measurement Conference, pp. 120-133. (DOI)]'s “an Internet shaped by constant mergers, rebrandings, and regional variation” has the inverse problem, unflagged.

Verified independently: the string constant mergers does not appear in paper.cols.txt at all — nor in paper.norm.txt or paper.txt — while pypdf over paper.pdf returns the sentence fully contiguous. So the quote is correct and the corpus's own rendering is the thing that is wrong. Footnoted to say so, because a reader checking the quote against the corpus would otherwise conclude it was invented.

This is the two-renderings rule in both directions on one page: Libert fails in every rendering and is real; Borges fails in the column-repaired rendering and is real. A quote-check against a single rendering would have reported one false FAIL and, had the Libert footnote not already existed, one apparent fabrication.

This is the most valuable finding of the sitting, and section B of this page had just asserted the opposite.

report_webxray.mjs ends with a section Z, Every number on the page that does NOT come from this corpus, which exists so check_page_numbers.mjs can tell a non-corpus figure from a corpus one. Five of its lines were hardcoded constants, and they were the ICANN+PRIVATE fold with exact-key lookup — the variant owner_dbs.py's own output labels <- THE TRAP and both content pages warn against by name, because it manufactures a coverage hole out of googleapis.com:

Verified independently before accepting: each figure was matched against the live owner_dbs.py run in section B, and grepped for on both content pages. None appears as prose on a live page, so nothing a reader saw was wrong — but the ALLOW list of a number guard is exactly where a wrong figure does its damage silently, by blessing the wrong value if it ever reaches a page. Corrected, and the corrected lines carry a comment naming the trap.

What this says about section B of this page. B calls report_webxray.mjs's output “byte-identical” to its committed copy, and it was — before and after. Byte-identity proves a script matches its own last run. It proves nothing about whether the content is still true, and here the script had been faithfully reproducing a stale constant. A reproduction record is not a correctness record, and this page should not have implied otherwise.

report_webxray.mjs buffered every line and flushed once at the bottom. Its section-A guard —

if (roleResidue.length || roleExtra.length) {
  P('FAILURE: the hand map and the sweep disagree. Re-read the new papers before publishing any figure below.');
}

— called P() and let the script continue. Section B then dereferences ROLE.get(k).role on the now-missing key and throws, before the single console.log at the bottom has run.

Reproduced by deleting one ROLE entry: stdout 0 bytes, a bare TypeError stack, and the FAILURE line that exists for precisely that case nowhere at all. The guard fires and is then thrown away. Fixed with a fail() helper that flushes the buffer, repeats the message on stderr and exits 1; re-running the same mutation now prints 1,283 bytes ending in the named paper and the FAILURE line, and exits 1. On a healthy run the output is byte-identical to before, so the fix is behaviour-neutral where it should be.

• All 22 distinct citekeys across the three pages resolve to exactly one bibliography entry; no collisions anywhere in the 993-entry file.
• All 11 quoted strings and 13 attributed non-quote claims verified verbatim or in substance, including the self-contradictory [3Steffens, Marius; Musch, Marius; Johns, Martin; Stock, Ben (2021): "Who’s Hosting the Block Party? Studying Third-Party Blockage of CSP and SRI", in: Proceedings of the Network and Distributed System Security Symposium. (Link)] sentence this page declines to derive a percentage from, and the [5Yang, Zhiju; Yue, Chuan (2020): "A Comparative Measurement Study of Web Tracking on Mobile and Desktop Environments", in: Proceedings on Privacy Enhancing Technologies. (DOI)] 411/251 split this sitting corrected.
• All 15 rows of the webXray publication table verified against source text — the reviewer was asked for five and did fifteen.
• Every figure on both content pages independently re-derived from a live re-run of all six scripts, including the inter-rater kappa, whose invocation the reviewer had to reconstruct because it was not in the brief. Both published code snippets (owner_lookup.py and resolve()) executed against the cached JSON and their output matched the page character for character.
• Three further mutations of report_ownership_resolution.mjs by the reviewer, including a re-creation of this sitting's own denominator bug — all caught, none vacuous.
• Every repository, licence, URL, version, dependency and acquisition claim on all three pages re-fetched and holding to the day, including the 2020-06-24 Disconnect relicensing commits, PolyForm Strict still having no SPDX id, and lxml/psycopg2-binary wheels still stopping at CPython 3.9 against a 2025-10-31 Python 3.9 EOL.

webxray's 2026-09-05 reviewer log records, verbatim: “The pages never say the adjudicators were language models … ACCEPTED, blocking. The content page now says 'by Sonnet sub-agents working to a published brief, single-rated'.” The carve-out dropped that sentence, and the new page said only “adjudicated one by one against primary sources — company newsrooms, SEC filings, or the domain's own legal documents”.

Verified before accepting: the clause is in the pre-split page text (live_programming_crawler_webxray.txt), and a grep of the new page for Sonnet, sub-agent and language model returned zero hits. A reader deciding whether to cite 94.4% was being told, by omission, that a person adjudicated it.

Restored in two places — beside the 28-row table and, for the first time, beside the random-sample table, which never carried it even before the split. No reviewer in three sittings had noticed the second gap. A finding accepted as blocking can be undone by a later structural edit, and no guard on this wiki looks for that.

The page said: “Disconnect's lead over webXray is a difference this sample can establish; its lead over Tracker Radar is not, and neither is Tracker Radar's over webXray.” Two p-values were published (0.014, 0.82). The Disconnect-versus-Tracker-Radar pair was never computed — the webXray provenance log shows the original reviewer supplied only the other two, and the sentence generalised from their absence.

Verified: Fisher's exact re-implemented from scratch over the eligible raw counts reproduces the two published values to three decimals (0.0137, 0.820) and gives p = 0.0248 for the untested pair. The sentence was false by the page's own test, directly under the short answer's headline comparison.

Rewritten as a three-row table with all three p-values and a Bonferroni threshold of 0.0167 for three tests on one sample — under which 0.014 survives and 0.025 does not, so the recommendation stands but for a stated reason rather than an invented one. The page now says in its own text that an earlier version asserted this without computing it.

This is the failure mode to remember: an absence of evidence in a reviewer's report was read as evidence of absence. Nothing in the three focused briefs would have caught it — the figures pass checks the page against the script, and this number was in neither.

“domain-level — … Every entry weighs the same.” It does not: owner_random_sample.py computes R = Σ_h (N_h/N)·p_h over the 24/12/12/12 quartile allocation. That is why 94.4% ≠ 39/42 = 92.9% and 69.9% ≠ 35/49 = 71.4% — and why the Fisher paragraph's raw counts did not match the table above it, which a reader dividing them would have read as an error on the page. Corrected, with the raw counts now stated beside the weighted ones.

The author's audit table below claims “all consistent; one error found”. That was wrong, and it is left standing above with this correction beside it rather than quietly rewritten — an audit that reports a clean result it did not earn is worse than no audit.

What the author checked without them, so the gap is bounded rather than open:

The 2026-09-05 sample drew 180 times over 175 distinct domains and could not adjudicate 40 of them (22.9%), 42 draws. Every accuracy rate it produced was therefore conditional on the entry being adjudicable, the exclusion rate rose toward the prevalence tail, and it was worst — 30% — for Disconnect, the list the content page recommends. Nothing inside that sample could test whether the excluded rows were systematically worse. This pass re-adjudicated exactly those 40 rows, changing nothing else, against three routes the first pass did not take.

The 2026-09-05 output file is kept, not overwritten. out/owner_random_sample-output.txt is the before; out/owner_random_sample-tail-output.txt is what the page's figures are now checked against, and out/owner_random_sample-tail.diff is the 338-line diff between them.

These counts are after the hand corrections. They differ from the owner_tail_merge.py output published in the appendix, which ran before them: that run reports live-source 12 and rdap-history 10, because the i.ua correction moved one row from rdap-history to live-source. Nothing else moved. The largest single route is live-source – a live document the first pass did not reach, usually because it looked for a page naming the domain where this pass looked for the entity.

The registrar hop is nonetheless the cheapest thing to add to any future pass: it is one extra socket per domain, and it settled rows RDAP alone could not. It produced Applied Technologies Internet SAS for at-o.net, Conversant, Inc. for the three ValueClick-style sync domains, Acoustic, L.P. for pages02.net, Leven Labs, Inc. for both Admiral domains and Google LLC for blogblog.com — none of which RDAP shows. This container has no whois(1) and no apt, so scripts/whois43.py speaks the protocol directly: IANA referral, TLD server, then the registrar's server when the registry is thin.

Three source kinds are new in this pass and every row records which it used:

So 17 of the 28 recovered rows clear the original bar and 11 rest on a kind the first pass did not use. tls-san — another organisation's certificate on this host — was defined in the brief as corroboration only and is the source of record for 0 rows.

The one non-OK row is adgrx.com: samsungads.ca serves an expired certificate, so curl refuses it and the checker never sees a body. Fetched by hand with –insecure: HTTP 200, 46,160 bytes, and the quoted row is byte-verbatim in it. The checker was deliberately not given an –insecure retry — an ownership citation whose host identity cannot be validated should surface, not be swallowed.

owner_verify_sources.py was written for the 2026-09-05 rows and this pass cited kinds of evidence it had never seen. Each defect would have reported a correct citation as a fabrication, and each fix carries a control:

Final tally over the 40 tail rows: FETCHFAIL=1, NOSOURCE=12, OK=27. The 12 NOSOURCE are the rows still unresolved, which have no citation by definition.

Six rows a machine flagged and a human then re-checked. keep means the citation holds and the checker or the transport was at fault; resource means the claim holds and the row cited the wrong document for it; rescore means the verdict was wrong on the brief's own rules.

Only one verdict changed, i.ua, and it moves Disconnect up, which is the self-serving direction. The alternative reading is that “I.UA” names no company at all and the row is an error; under it Disconnect's domain-level current would be lower still than the 89.2% now published, so the published figure is the more favourable of the two readings and not the less.

12 distinct domains of the 175, down from 40. These are not rows where the register was not tried. Each note below is the adjudicator's own account of what each of the three routes returned, in full and not truncated – an earlier draft cut this column at 420 characters mid-word, which made “every route tried” a claim a reader could not check.

Two of the twelve are adjudicator limits rather than evidence limits, and saying so matters because they are the two a person could still close. hqseek.com has a live site: it redirects to an adult domain and the adjudicator declined to fetch further, so the row is unresolved because a model stopped, not because the evidence is gone – and its note records that Tracker Radar's claimed owner matches no name the domain ever surfaced, which would be an error rather than an unknown if anyone read the pages. stripst.com is unresolved because Stripchat's legal pages return HTTP 406 to a non-browser client, which is a transport failure, not an absence of documents. The other ten are genuinely evidence-exhausted.

Two obstacles are worth naming for anyone attempting this again: mmstat.com would be settled by China's ICP registry at beian.miit.gov.cn, which serves a JavaScript challenge no fetch here could pass; agkn.com by TransUnion's own pages, which return 403 to everything that is not a browser session. The remaining residue is a lower bound on what harder sourcing can reach, not a claim that these domains have no owner.

• Whether the 12 still-unresolved rows are worse than the rest is still unbounded, and this pass makes the bound narrower rather than tighter: the rows it recovered are by construction the easier end of the residue, so the test in section B of report_tail_pass.py is a test of the recovered rows, not of the residue. It can show the recovered rows are worse, or fail to; it cannot say anything about what is left. report_tail_pass.py says so in its own docstring and the content page says so in prose.
• The recovered/existing comparison has almost no power. 7, 9 and 12 recovered rows. A difference would have to be enormous to register, and none of the three p-values is interpretable as evidence of no difference.
• The adjudicators were again language models, one rater per domain, working to out/tail_adj/INSTRUCTIONS_TAIL.md. No inter-rater agreement was measured for this pass — the +0.400 kappa on the content page is the 2026-09-05 rows. Measuring it would need a second model over the same 40 and was not attempted this sitting.
• Three sub-agents could not re-issue a port-43 query themselves (this sandbox's curl has no whois:// support and bash has no /dev/tcp) and relied on the probe's already-fetched WHOIS text, disclosing it per row. Every such citation was independently re-fetched by owner_verify_sources.py afterwards, which is what closes that gap.
• cratecamera.com and sa-as.com rest on an uncorroborated registrant organisation and nothing further could be found for either: Admiral publishes no allowlist of its own domains, and no Foundry page names sa-as.com. They are in the sensitivity run for that reason.

All four told explicitly that the author's context may not be exhaustive, all handed the page text, the scripts, their unedited output and these notes. The three focused passes ran in parallel against a bundle cut at 19:21; the generic pass ran afterwards against the pages as the first three had left them.

It ran after the three focused ones, against the pages as they had left them, and it returned 15 findings. Fourteen were accepted. Two of them changed what the page claims, and both were things all three focused passes and the author's own site-wide sweep had read past.

What this pass cost and what it bought. It was the only one of the four to question the *choice of metric* rather than the arithmetic within one, and that question changed the page's headline claim from “the second pass overturned Disconnect's lead” to “it overturned a legal-entity-level lead over a list this page calls historical, and cost it nothing at the company level”. The three focused passes checked every number against the script and found the script's own metric switch invisible, because each number was right for the metric its section used. A checklist cannot find a framing defect; this is the strongest evidence so far that the generic slot earns its place.

Three of the four passes independently reported that the content page's <WRAP important> “short answer” box still carried 94.4% / 73.7% / 93.9%. It did when their bundle was cut at 19:21. A site-wide sweep (scripts/sweep_moved_figures.py) had found and fixed it at 19:28, along with three more the first patch missed. The finding was right and the bundle was stale — the same trap this wiki has recorded before. Two of the four passes also named the <WRAP todo> that still asked for the thing this sitting did; that one was not yet fixed and has been.

Fisher's exact (independently reimplemented: 0.031 / 0.083 / 0.831, matching); the eligibility rule (read against the estimator's, equivalent, and asserted at runtime); the worst-case convention; the raw-count arithmetic; the source-kind split (9 / 2 / 17, recomputed from the rows); the reproducibility of every committed output; citekeys (22, all resolving once against a freshly fetched bibliography, none added); all 28 recovered citations re-fetched and verified verbatim by a second party; the three error verdicts, each re-derived from a primary source; the i.ua override, judged substantively right by an independent read of the surrounding agreement text.

report_ownership_resolution.mjs

// Every corpus figure on Design:Ownership resolution, with its denominator.
//
//   node scripts/report_ownership_resolution.mjs > scripts/report_ownership_resolution-output.txt
//   node scripts/report_ownership_resolution.mjs --wiki    # DokuWiki tables
//
// This page was carved out of Programming:Crawler:webXray on 2026-09-11. The
// corpus-side figures it inherited were computed by scripts/report_webxray.mjs,
// whose population is "papers that name webXray". That is the wrong population
// for a page about ownership resolution in general, so this script re-derives
// the landscape figures from the corpus independently and then CROSS-CHECKS the
// overlapping rows against report_webxray.mjs's committed output. Two
// implementations agreeing is a stronger audit trail than one copied number.
//
// Three things this script does that a naive sweep would not:
//
// 1. Every probe is run at TWO widths and the script asserts tight <= loose AND
//    tight subset-of loose. A narrowing probe that returns more hits, or hits
//    the loose one does not contain, is a broken probe, not a finding. The
//    Disconnect probe is the one that needs this: plain /disconnect/i is a
//    common English word and the CSP literature uses it as a technical term.
//
// 2. Every count is a PAPER count over data/fulltext/<year>/<venue>/<slug>/
//    paper.cols.txt with whitespace collapsed, because a PDF line break inside
//    "Public Suffix List" silently undercounts it.
//
// 3. Papers with no full text are counted and printed, so the sweep's own
//    denominator is visible rather than assumed to be 5,859.
//
// Every row here is an UPPER BOUND on use unless the output says hand-verified:
// a full-text match is a mention, not a use. The two rows that ARE hand
// verified (webXray, Tracker Radar) carry their verdicts in report_webxray.mjs.
 
import fs from 'node:fs';
import path from 'node:path';
import { dataRoot, loadExtractions, POPULATIONS, pct, table, wikiTable } from './lib.mjs';
 
const WIKI = process.argv.includes('--wiki');
const rows = loadExtractions();
const key = (p) => `${p.venue}/${p.year}/${p.slug}`;
const FT = path.join(dataRoot(), 'fulltext');
 
const out = [];
const P = (s = '') => out.push(s);
const heading = (s) => {
  P('');
  P(`=== ${s} ===`);
  P('');
};
const T = (headers, body) => P(WIKI ? wikiTable(headers, body) : table(headers, body));
 
// --- full text, whitespace collapsed -------------------------------------
// Streamed, not cached: the corpus is 5,859 papers and holding every
// paper.cols.txt in a Map exhausts the default V8 heap. One pass, every
// pattern tested against each paper's text, then the text is dropped.
function readText(k) {
  const [venue, year, slug] = k.split('/');
  const f = `${FT}/${year}/${venue}/${slug}/paper.cols.txt`;
  return fs.existsSync(f) ? fs.readFileSync(f, 'utf8').replace(/\s+/g, ' ') : null;
}
 
// Which papers have full text at all, and the hits for every pattern, in one
// pass. PATTERNS is filled below before this runs.
function sweepAll(patterns) {
  const hits = new Map([...patterns.keys()].map((n) => [n, []]));
  const haveText = new Set();
  for (const p of rows) {
    const k = key(p);
    const t = readText(k);
    if (t === null) continue;
    haveText.add(k);
    for (const [name, re] of patterns) if (re.test(t)) hits.get(name).push(k);
  }
  for (const v of hits.values()) v.sort();
  return { hits, haveText };
}
 
// --- the probes, each at two widths --------------------------------------
// `loose` must be a strict superset of `tight` by construction; the script
// checks that it is in fact one and throws if not.
const PROBES = [
  {
    name: 'webXray',
    tight: /webx[\s-]?ray/i,
    loose: /webx[\s-]?ray|libert/i,
    note: 'hand-verified: all 15, in report_webxray.mjs',
  },
  {
    name: 'Tracker Radar',
    tight: /tracker[\s.-]?radar/i,
    loose: /tracker[\s.-]?radar|duckduckgo/i,
    note: 'hand-verified: all 32, in report_webxray.mjs',
  },
  {
    name: 'WhoTracks.me',
    tight: /whotracks/i,
    loose: /whotracks|who ?tracks ?\.? ?me|ghostery/i,
    note: 'upper bound',
  },
  {
    name: 'Crunchbase',
    tight: /crunchbase/i,
    loose: /crunchbase|crunch base/i,
    note: 'upper bound',
  },
  {
    name: 'Disconnect (list sense)',
    tight: /disconnect'?s? (entit|list|block|black|tracking)|disconnect\.me|entities\.json/i,
    loose: /disconnect/i,
    note: 'upper bound; the loose width is the reason this one is tightened',
  },
  {
    name: 'Public Suffix List',
    tight: /public suffix/i,
    loose: /public suffix|etld\+?1|effective top-?level domain/i,
    note: 'upper bound',
  },
];
 
// One pass over the full text: every probe, both widths, plus the set of papers
// that have a readable rendering at all.
const PATTERNS = new Map();
for (const pr of PROBES) {
  PATTERNS.set(`${pr.name}|tight`, pr.tight);
  PATTERNS.set(`${pr.name}|loose`, pr.loose);
}
const { hits: SWEPT, haveText: HAVE_TEXT } = sweepAll(PATTERNS);
 
const withText = rows.filter((p) => HAVE_TEXT.has(key(p)));
const crawled = rows.filter(POPULATIONS.crawled);
 
P('==============================================================================');
P('Design:Ownership resolution — every corpus figure, with its denominator');
P('==============================================================================');
P('');
P(`Corpus: ${rows.length} extracted papers, 7 venues (CCS, IMC, NDSS, PETS, USENIX`);
P('Security, TheWebConf, IEEE S&P), 2010–2026. EuroS&P, ACSAC, RAID, AsiaCCS, CHI');
P('and SOUPS are absent, so every figure here is a claim about those seven venues.');
P('');
P(`Papers with a readable paper.cols.txt: ${withText.length} of ${rows.length} (${pct(withText.length, rows.length)}).`);
P(`Papers that ran a crawl (crawlConfig present or studyTypes includes automated-web-crawl): ${crawled.length}.`);
P('The sweep denominator is the full corpus; papers with no full text can only');
P('ever be misses, so every sweep count below is a floor on the true mention count.');
 
// --- A. probe widths ------------------------------------------------------
heading('A. Probe widths: tight must be contained in loose');
 
const results = new Map();
const widthRows = [];
for (const pr of PROBES) {
  const tight = SWEPT.get(`${pr.name}|tight`);
  const loose = SWEPT.get(`${pr.name}|loose`);
  const tightSet = new Set(tight);
  const looseSet = new Set(loose);
  const escaped = tight.filter((k) => !looseSet.has(k));
  if (tight.length > loose.length) {
    throw new Error(`probe ${pr.name}: tight (${tight.length}) > loose (${loose.length}) — the narrowing probe is broken`);
  }
  if (escaped.length > 0) {
    throw new Error(`probe ${pr.name}: ${escaped.length} tight hits are not in the loose set: ${escaped.slice(0, 5).join(', ')}`);
  }
  results.set(pr.name, tightSet);
  widthRows.push([pr.name, String(tight.length), String(loose.length), pr.note]);
}
T(['Resource', 'Tight probe', 'Loose probe', 'Status'], widthRows);
P('');
P('Both widths are printed because the gap is the claim. Disconnect at the loose');
P('width matches 700 papers and almost none of them mean the list, which is why');
P('the published row is the tight one and why the page says so. A row whose two');
P('widths are close is a row whose count is not an artefact of the pattern.');
 
// --- B. the published landscape table ------------------------------------
heading('B. The ownership-resolution landscape (the page\'s main table)');
 
const landscape = PROBES.map((pr) => {
  const s = results.get(pr.name);
  const inCrawled = crawled.filter((p) => s.has(key(p))).length;
  return [pr.name, String(s.size), pct(s.size, rows.length), String(inCrawled), pct(inCrawled, crawled.length), pr.note.startsWith('hand-verified') ? 'yes' : 'no — upper bound'];
})
  .sort((a, b) => Number(b[1]) - Number(a[1]));
T(['Resource named in full text', 'Papers', `Share of ${rows.length}`, `of which crawled`, `Share of ${crawled.length}`, 'Hand-verified?'], landscape);
 
// --- C. the union ---------------------------------------------------------
heading('C. The union: how widespread is ownership resolution at all?');
 
// The union is over the five OWNERSHIP resources. The Public Suffix List is
// deliberately excluded: it answers "what is the registrable domain", not
// "whose is it", and including it would nearly double the union on a resource
// that every crawl paper touches for unrelated reasons.
const UNION_NAMES = ['webXray', 'Tracker Radar', 'WhoTracks.me', 'Crunchbase', 'Disconnect (list sense)'];
const union = new Set();
for (const n of UNION_NAMES) for (const k of results.get(n)) union.add(k);
P(`Union of ${UNION_NAMES.join(' / ')}:`);
P(`  ${union.size} papers — ${pct(union.size, rows.length)} of the ${rows.length}-paper corpus,`);
const unionCrawled = crawled.filter((p) => union.has(key(p)));
P(`  and ${pct(unionCrawled.length, crawled.length)} of the ${crawled.length} that ran a crawl (${unionCrawled.length} papers).`);
P('');
P('The Public Suffix List is NOT in the union: it answers "what is the registrable');
P('domain", not "whose is it". Adding it would take the union to ' +
  (() => { const u2 = new Set(union); for (const k of results.get('Public Suffix List')) u2.add(k); return u2.size; })() +
  ' on a resource');
P('every crawling paper touches for unrelated reasons.');
P('');
P('Each member is an upper bound on USE, so the union is an upper bound too:');
P('136 papers mention one of these; far fewer resolved an owner with one.');
 
// --- D. per-year adoption -------------------------------------------------
heading('D. Per-year: is ownership resolution growing?');
 
const years = [...new Set(rows.map((p) => p.year))].sort();
const yearRows = years.map((y) => {
  const inYear = rows.filter((p) => p.year === y);
  const hit = inYear.filter((p) => union.has(key(p)));
  const star = y >= 2025 ? '*' : '';
  return [`${y}${star}`, String(inYear.length), String(hit.length), pct(hit.length, inYear.length)];
});
T(['Year', 'Corpus papers', 'Naming an ownership resource', 'Share'], yearRows);
P('');
P('* 2025 and 2026 are the provisional corpus edge: CCS 2026 and IMC 2026 have not');
P('  been held, and IEEE S&P / WWW 2026 abstracts are not in OpenAlex, so selection');
P('  under-samples them BY CONSTRUCTION. Do not read 2026 as a complete year.');
P('  The share, not the count, is the readable column for those two rows.');
 
// --- E. venue shape -------------------------------------------------------
heading('E. Venue shape of the union');
 
const venues = [...new Set(rows.map((p) => p.venue))].sort();
const venueRows = venues.map((v) => {
  const inVenue = rows.filter((p) => p.venue === v);
  const hit = inVenue.filter((p) => union.has(key(p)));
  return [v, String(inVenue.length), String(hit.length), pct(hit.length, inVenue.length)];
}).sort((a, b) => Number(b[2]) - Number(a[2]));
T(['Venue', 'Corpus papers', 'Naming an ownership resource', 'Share of venue'], venueRows);
P('');
P('Read the SHARE column, not the count: the venues differ in size by a factor of');
P('three. IMC and PETS are where this work lands; the shares are small everywhere,');
P('which is the honest headline — ownership resolution is a step inside a paper,');
P('not a paper topic, and most papers that do it do not say how.');
 
// --- F. the schema's own view --------------------------------------------
heading('F. What the extraction schema sees, against the sweep');
 
// classification[].groundTruthSource / resourceName naming an ownership list.
const SCHEMA_RE = /webx[\s-]?ray|tracker[\s.-]?radar|whotracks|disconnect|crunchbase/i;
const schemaPapers = new Set();
for (const p of rows) {
  const names = [
    ...p.tools.map((t) => t.name),
    ...p.classification.map((c) => c.resourceName),
    ...p.classification.map((c) => c.groundTruthSource),
    ...p.otherToolsMentioned.map((t) => t.name),
  ].filter((s) => typeof s === 'string');
  if (names.some((n) => SCHEMA_RE.test(n))) schemaPapers.add(key(p));
}
const both = [...union].filter((k) => schemaPapers.has(k));
P(`Papers whose tools[] / classification[] / otherToolsMentioned[] name one of the`);
P(`five resources:            ${schemaPapers.size}`);
P(`Papers whose FULL TEXT names one:              ${union.size}`);
P(`In both:                                       ${both.length}`);
P(`Full text only (schema misses):                ${union.size - both.length}`);
P(`Schema only (no full-text match):              ${schemaPapers.size - both.length}`);
P('');
P('The schema-only residue is printed rather than dropped:');
const schemaOnly = [...schemaPapers].filter((k) => !union.has(k)).sort();
for (const k of schemaOnly) {
  const p = rows.find((r) => key(r) === k);
  const hits = [
    ...p.tools.map((t) => t.name),
    ...p.classification.map((c) => c.resourceName),
    ...p.classification.map((c) => c.groundTruthSource),
    ...p.otherToolsMentioned.map((t) => t.name),
  ].filter((s) => typeof s === 'string' && SCHEMA_RE.test(s));
  P(`  ${k}`);
  P(`      named: ${[...new Set(hits)].join(' | ')}`);
  P(`      full text present: ${HAVE_TEXT.has(k)}`);
}
P('');
P('This is why the page publishes the full-text sweep and not the schema field:');
P('the schema fires on a fraction of the papers that name these lists, because');
P('most mentions are in related work or a reference list rather than in a tools');
P('sentence the extractor reads as a tool. Neither signal is "the" population;');
P('both are reported.');
 
// --- G. cross-check against report_webxray.mjs ---------------------------
heading('G. Cross-check against scripts/report_webxray-output.txt');
 
const committed = 'scripts/report_webxray-output.txt';
if (!fs.existsSync(committed)) {
  P(`${committed} not found — cross-check SKIPPED.`);
} else {
  const txt = fs.readFileSync(committed, 'utf8');
  const CHECKS = [
    ['webXray', /^webXray\s+(\d+)\s/m],
    ['Tracker Radar', /^Tracker Radar\s+(\d+)\s/m],
    ['WhoTracks.me', /^WhoTracks\.me\s+(\d+)\s/m],
    ['Crunchbase', /^Crunchbase\s+(\d+)\s/m],
    ['Disconnect (list sense)', /^Disconnect \(list sense\)\s+(\d+)\s/m],
    ['Public Suffix List', /^Public Suffix List\s+(\d+)\s/m],
  ];
  const checkRows = [];
  let bad = 0;
  for (const [name, re] of CHECKS) {
    const m = txt.match(re);
    if (m === null) throw new Error(`cross-check: could not find "${name}" in ${committed}`);
    const theirs = Number(m[1]);
    const mine = results.get(name).size;
    if (theirs !== mine) bad += 1;
    checkRows.push([name, String(mine), String(theirs), theirs === mine ? 'agree' : '*** DISAGREE ***']);
  }
  const um = txt.match(/^\s*(\d+) papers \((\d+\.\d)% of the corpus\)\. (\d+) of them are among the (\d+) that crawled -- (\d+\.\d)% of that population\./m);
  if (um === null) throw new Error(`cross-check: could not find the union line in ${committed}`);
  if (Number(um[1]) !== union.size) bad += 1;
  checkRows.push(['union of the five', String(union.size), um[1], Number(um[1]) === union.size ? 'agree' : '*** DISAGREE ***']);
  if (Number(um[3]) !== unionCrawled.length) bad += 1;
  checkRows.push(['union ∩ crawled', String(unionCrawled.length), um[3], Number(um[3]) === unionCrawled.length ? 'agree' : '*** DISAGREE ***']);
  if (Number(um[4]) !== crawled.length) bad += 1;
  checkRows.push(['crawled denominator', String(crawled.length), um[4], Number(um[4]) === crawled.length ? 'agree' : '*** DISAGREE ***']);
  T(['Figure', 'this script', 'report_webxray.mjs', 'verdict'], checkRows);
  P('');
  if (bad > 0) {
    throw new Error(`${bad} figures disagree between the two independent implementations — do not publish either.`);
  }
  P('All rows agree. The two scripts share no code beyond lib.mjs and the');
  P('whitespace-collapse convention; the patterns were written out separately.');
}
 
// --- H. the page's non-corpus figures, listed so they are not mistaken ----
heading('H. Every number on the page that does NOT come from this corpus');
 
P('Listed so a reader auditing the page knows which script to re-run, and so');
P('that a refresh of this script is never mistaken for a refresh of the page.');
P('');
T(['Figure family', 'Script', 'Snapshot it is pinned to'], [
  ['List shape (827 / 19,148 / 1,887 owners; 3,215 / 38,368 / 7,850 domains)', 'owner_dbs.py --cache out/webxray/cache', '2026-08-17'],
  ['Coverage census (669 / 5,581 / 2,268 of 32,369; the prevalence weights; the rank slices)', 'owner_dbs.py --cache out/webxray/cache', '2026-08-17'],
  ['Agreement between pairs (612 / 464 / 1,538 and the "neither" columns)', 'owner_dbs.py --cache out/webxray/cache', '2026-08-17'],
  ['The 28-row hand adjudication of high-prevalence disagreements', 'owner_adjudication.py --table', '2026-08-17'],
  ['Accuracy rates (69.2 / 73.9 / 89.2%, the encounter-weighted column, the CIs)', 'owner_random_sample.py --sample out/owner_sample.json --rows out/adj_rows_tail_final.json', '2026-09-05 draw, 2026-09-11 second pass'],
  ['What the second pass over the 40 unadjudicable rows changed', 'owner_tail_probe.sh | owner_tail_merge.py | owner_tail_corrections.py | report_tail_pass.py', '2026-09-11'],
  ['Coverage counts inside the random-sample section (664 / 5,566 / 2,264 of 32,337)', 'owner_sample.py --cache out/owner_cache', '2026-09-05'],
  ['Self-named entity shares (0.8% / 0.7% / 3.2%)', 'owner_selfname_census.py', '2026-09-05'],
  ['Inter-rater kappa (+0.400, +0.680)', 'owner_irr_kappa.py', '2026-09-11'],
  ['Repository state, licences, Wayback dates', 'hand checks + wayback_webxray.sh', '2026-08-17 / 2026-09-05'],
]);
P('');
P('Two of the three lists change weekly. Re-run before citing.');
 
console.log(out.join('\n'));

Run as node scripts/report_ownership_resolution.mjs.

report_ownership_resolution-output.txt

==============================================================================
Design:Ownership resolution — every corpus figure, with its denominator
==============================================================================
 
Corpus: 5859 extracted papers, 7 venues (CCS, IMC, NDSS, PETS, USENIX
Security, TheWebConf, IEEE S&P), 2010–2026. EuroS&P, ACSAC, RAID, AsiaCCS, CHI
and SOUPS are absent, so every figure here is a claim about those seven venues.
 
Papers with a readable paper.cols.txt: 5855 of 5859 (99.9%).
Papers that ran a crawl (crawlConfig present or studyTypes includes automated-web-crawl): 1120.
The sweep denominator is the full corpus; papers with no full text can only
ever be misses, so every sweep count below is a floor on the true mention count.
 
=== A. Probe widths: tight must be contained in loose ===
 
Resource                 Tight probe  Loose probe  Status
-----------------------  -----------  -----------  ----------------------------------------------------------------
webXray                  15           233          hand-verified: all 15, in report_webxray.mjs
Tracker Radar            32           108          hand-verified: all 32, in report_webxray.mjs
WhoTracks.me             23           124          upper bound
Crunchbase               23           23           upper bound
Disconnect (list sense)  74           700          upper bound; the loose width is the reason this one is tightened
Public Suffix List       101          157          upper bound
 
Both widths are printed because the gap is the claim. Disconnect at the loose
width matches 700 papers and almost none of them mean the list, which is why
the published row is the tight one and why the page says so. A row whose two
widths are close is a row whose count is not an artefact of the pattern.
 
=== B. The ownership-resolution landscape (the page's main table) ===
 
Resource named in full text  Papers  Share of 5859  of which crawled  Share of 1120  Hand-verified?
---------------------------  ------  -------------  ----------------  -------------  ----------------
Public Suffix List           101     1.7%           46                4.1%           no — upper bound
Disconnect (list sense)      74      1.3%           65                5.8%           no — upper bound
Tracker Radar                32      0.5%           28                2.5%           yes
WhoTracks.me                 23      0.4%           19                1.7%           no — upper bound
Crunchbase                   23      0.4%           11                1.0%           no — upper bound
webXray                      15      0.3%           13                1.2%           yes
 
=== C. The union: how widespread is ownership resolution at all? ===
 
Union of webXray / Tracker Radar / WhoTracks.me / Crunchbase / Disconnect (list sense):
  136 papers — 2.3% of the 5859-paper corpus,
  and 9.6% of the 1120 that ran a crawl (107 papers).
 
The Public Suffix List is NOT in the union: it answers "what is the registrable
domain", not "whose is it". Adding it would take the union to 221 on a resource
every crawling paper touches for unrelated reasons.
 
Each member is an upper bound on USE, so the union is an upper bound too:
136 papers mention one of these; far fewer resolved an owner with one.
 
=== D. Per-year: is ownership resolution growing? ===
 
Year   Corpus papers  Naming an ownership resource  Share
-----  -------------  ----------------------------  -----
2010   119            0                             0.0%
2011   116            1                             0.9%
2012   151            0                             0.0%
2013   125            0                             0.0%
2014   166            0                             0.0%
2015   190            0                             0.0%
2016   182            2                             1.1%
2017   231            4                             1.7%
2018   254            5                             2.0%
2019   402            9                             2.2%
2020   404            15                            3.7%
2021   379            13                            3.4%
2022   546            19                            3.5%
2023   719            19                            2.6%
2024   690            17                            2.5%
2025*  770            21                            2.7%
2026*  415            11                            2.7%
 
* 2025 and 2026 are the provisional corpus edge: CCS 2026 and IMC 2026 have not
  been held, and IEEE S&P / WWW 2026 abstracts are not in OpenAlex, so selection
  under-samples them BY CONSTRUCTION. Do not read 2026 as a complete year.
  The share, not the count, is the readable column for those two rows.
 
=== E. Venue shape of the union ===
 
Venue    Corpus papers  Naming an ownership resource  Share of venue
-------  -------------  ----------------------------  --------------
PETS     510            49                            9.6%
USENIX   1410           19                            1.3%
WWW      843            19                            2.3%
IMC      638            17                            2.7%
CCS      990            13                            1.3%
IEEE-SP  767            11                            1.4%
NDSS     701            8                             1.1%
 
Read the SHARE column, not the count: the venues differ in size by a factor of
three. IMC and PETS are where this work lands; the shares are small everywhere,
which is the honest headline — ownership resolution is a step inside a paper,
not a paper topic, and most papers that do it do not say how.
 
=== F. What the extraction schema sees, against the sweep ===
 
Papers whose tools[] / classification[] / otherToolsMentioned[] name one of the
five resources:            89
Papers whose FULL TEXT names one:              136
In both:                                       82
Full text only (schema misses):                54
Schema only (no full-text match):              7
 
The schema-only residue is printed rather than dropped:
  IEEE-SP/2024/to-auth-or-not-to-auth-a-comparative-analysis-of-the-pre-and-post-login-security
      named: Disconnect Tracker Protection List
      full text present: true
  PETS/2022/on-dark-patterns-and-manipulation-of-website-publishers-by-cmps
      named: Disconnect | Disconnect tracking filter list
      full text present: true
  PETS/2024/generalizable-active-privacy-choice-designing-a-graphical-user-interface-for-glo
      named: Disconnect Tracker Protection lists
      full text present: true
  PETS/2024/support-personas-a-concept-for-tailored-support-of-users-of-privacy-enhancing-te
      named: Disconnect
      full text present: true
  USENIX/2019/canvas-fast-and-inexpensive-automotive-network-mapping
      named: physical ECU access and disconnection
      full text present: true
  WWW/2018/the-cost-of-digital-advertisement-comparing-user-and-advertiser-views
      named: Disconnect | Disconnect blacklist
      full text present: true
  WWW/2023/who-funds-misinformation-a-systematic-analysis-of-the-ad-related-profit-routines
      named: Disconnect
      full text present: true
 
This is why the page publishes the full-text sweep and not the schema field:
the schema fires on a fraction of the papers that name these lists, because
most mentions are in related work or a reference list rather than in a tools
sentence the extractor reads as a tool. Neither signal is "the" population;
both are reported.
 
=== G. Cross-check against scripts/report_webxray-output.txt ===
 
Figure                   this script  report_webxray.mjs  verdict
-----------------------  -----------  ------------------  -------
webXray                  15           15                  agree
Tracker Radar            32           32                  agree
WhoTracks.me             23           23                  agree
Crunchbase               23           23                  agree
Disconnect (list sense)  74           74                  agree
Public Suffix List       101          101                 agree
union of the five        136          136                 agree
union ∩ crawled          107          107                 agree
crawled denominator      1120         1120                agree
 
All rows agree. The two scripts share no code beyond lib.mjs and the
whitespace-collapse convention; the patterns were written out separately.
 
=== H. Every number on the page that does NOT come from this corpus ===
 
Listed so a reader auditing the page knows which script to re-run, and so
that a refresh of this script is never mistaken for a refresh of the page.
 
Figure family                                                                             Script                                                                                       Snapshot it is pinned to
----------------------------------------------------------------------------------------  -------------------------------------------------------------------------------------------  ---------------------------------------
List shape (827 / 19,148 / 1,887 owners; 3,215 / 38,368 / 7,850 domains)                  owner_dbs.py --cache out/webxray/cache                                                       2026-08-17
Coverage census (669 / 5,581 / 2,268 of 32,369; the prevalence weights; the rank slices)  owner_dbs.py --cache out/webxray/cache                                                       2026-08-17
Agreement between pairs (612 / 464 / 1,538 and the "neither" columns)                     owner_dbs.py --cache out/webxray/cache                                                       2026-08-17
The 28-row hand adjudication of high-prevalence disagreements                             owner_adjudication.py --table                                                                2026-08-17
Accuracy rates (69.2 / 73.9 / 89.2%, the encounter-weighted column, the CIs)              owner_random_sample.py --sample out/owner_sample.json --rows out/adj_rows_tail_final.json    2026-09-05 draw, 2026-09-11 second pass
What the second pass over the 40 unadjudicable rows changed                               owner_tail_probe.sh | owner_tail_merge.py | owner_tail_corrections.py | report_tail_pass.py  2026-09-11
Coverage counts inside the random-sample section (664 / 5,566 / 2,264 of 32,337)          owner_sample.py --cache out/owner_cache                                                      2026-09-05
Self-named entity shares (0.8% / 0.7% / 3.2%)                                             owner_selfname_census.py                                                                     2026-09-05
Inter-rater kappa (+0.400, +0.680)                                                        owner_irr_kappa.py                                                                           2026-09-11
Repository state, licences, Wayback dates                                                 hand checks + wayback_webxray.sh                                                             2026-08-17 / 2026-09-05
 
Two of the three lists change weekly. Re-run before citing.

• Ownership resolution — the page these notes are for.
• webxray — the live-list scripts, the fold residues, the Wayback log, the 2026-08-17 and 2026-09-05 run tables.
• random_sample — the 175-row adjudication, the estimator, the inter-rater draw and kappa.
• residue_pass — the 2026-09-11 second pass over the residue in full: its probe, its brief, its merge gate, its corrections, its report, its guard, its mutation harness and every unedited output, including the 40-domain probe log and the diff between the two estimates.
• Corpus — corpus-level selection and extraction caveats.

[1]

Matte, Célestin; Bielova, Nataliia; Santos, Cristiana Teixeira (2020): "Do Cookie Banners Respect my Choice? Measuring Legal Compliance of Banners from IAB Europe's Transparency and Consent Framework", in: Proceedings of the IEEE Symposium on Security and Privacy. (DOI)

[2]

Libert, Timothy (2018): "An Automated Approach to Auditing Disclosure of Third-Party Data Collection in Website Privacy Policies", in: Proceedings of the ACM Web Conference. (DOI)

[3]

Steffens, Marius; Musch, Marius; Johns, Martin; Stock, Ben (2021): "Who’s Hosting the Block Party? Studying Third-Party Blockage of CSP and SRI", in: Proceedings of the Network and Distributed System Security Symposium. (Link)

[4]

Sánchez-Rola, Iskander; Dell'Amico, Matteo; Balzarotti, Davide; Vervier, Pierre-Antoine; Bilge, Leyla (2021): "Journey to the Center of the Cookie Ecosystem: Unraveling Actors' Roles and Relationships", in: Proceedings of the IEEE Symposium on Security and Privacy. (DOI)

[5]

Yang, Zhiju; Yue, Chuan (2020): "A Comparative Measurement Study of Web Tracking on Mobile and Desktop Environments", in: Proceedings on Privacy Enhancing Technologies. (DOI)

[6]

Selmo, Carlos; Carisimo, Esteban; Bustamante, Fabián E.; Alvarez-Hamelin, J. Ignacio (2025): "Learning AS-to-Organization Mappings with Borges", in: Proceedings of the 2025 ACM Internet Measurement Conference, pp. 120-133. (DOI)