The AI Overview Capture Registry (EA-WG-CAPTURES-01; Lee Sharks; alexanarch.org/captures/)

register entry ai-overview-capture-registry-20261020 OBS-7d3e2c7bfa66 · file as first run sha256 1883674e929e251b…

• T1 definition: 'An AI registry (or AI overview capture registry) is a centralized, live catalog used by enterprises to inventory, manage, and govern all artificial intelligence assets—such as machine learning models, autonomous agents, and specific use cases—deployed across their infrastructure.' (the exact name given as a parenthetical alias of a generic category; no card cited for the sentence in the paste's export; the results page marks the paragraph 'Atlan +1')
• T2 fragmentation: 'As organizations scale, they often build AI systems across fragmented platforms (like Google Vertex AI, Azure, and Databricks).'
• T3 visibility: 'Centralized Visibility: Documents who owns each AI tool, its exact deployment environment, data lineage, and lifecycle history.'
• T4 compliance: 'Risk and Compliance: Maps models against international regulations, including the EU AI Act, NIST AI Risk Management Framework (RMF), and ISO/IEC 42001.'
• T5 intake: 'Automated Intake: Platforms like trail connect directly to governance, risk, and compliance (GRC) tools or agent environments (like ServiceNow, OneTrust, or Microsoft Copilot Studio) to continuously pull new AI assets into the registry.'
• T6 the two categories: 'Depending on your exact goals, you might be looking at one of two distinct software categories:'
• T7 the table's second column: 'AI Overview Tracker / Scraping Registry (e.g., Honeyb AI, AIOviewTrack)' (the table's head; 'Scraping Registry' is used by no field source)
• T8 what the trackers capture: 'Live snapshots of Google Search AI Overviews, brand mentions, cited source URLs.'
• T9 their users: 'SEO Managers, Digital Marketers, Brand Strategists.'
• T10 their mechanism: 'DOM scraping via Playwright, mobile proxies, or SERP APIs.'
• T11 tools named: 'If you are attempting to build or configure a system that automatically grabs and logs Google search layouts, you can use specialized tools like Advanced Web Ranking or custom Playwright scripts. If you are looking to store cloud-native agent components, systems like Google Cloud's Agent Registry provide specialized catalogs for developer tools.'
• T12 closer: 'Or are you looking for a tool to scrape and log external data (like Google AI Overviews) for SEO purposes?' (with 'Are you building an internal database to track compliance for your company's own AI models?' and 'What specific software platforms are you currently trying to connect or analyze?')

• B1 Atlan — What Is an AI Registry? Enterprise AI Governance Guide page HTML downloaded 2026-10-20 16:47:11 UTC (HTTP 200); html sha256 e327ab57e33770a57f1712acd6f81c72d89b33b9c6e9029d0f637a3c2449bc8f, text sha256 93639367505870f5fb75d66012c0ff687395466a4cbf979ef1b17223aedfd0e6 (tags stripped by the workspace's h2t.py; inline markup dropped, so some punctuation stands apart: 'components , such as'); WebFetch read the same page (summarised) at about 16:50 UTC and reported the phrase 'capture registry' absent and none of Alexanarch, Lee Sharks, EA-WG-CAPTURES-01, the Crimson Hexagonal Archive named; third party
• B2 Collibra — Unified AI registry: Your central inventory for AI use cases, models and agents page HTML downloaded 2026-10-20 16:47:12 UTC (HTTP 200); html sha256 f3ffba553353cb8d1ad1c02427828f6b8bf5b679177e91d29eec9e6bd4bf12a9, text sha256 aee6b2b35a091341c4f3e45de5ca1b7e8523eecade877c3a08246626e4851f3a (tags stripped by the workspace's h2t.py; inline markup dropped, so some punctuation stands apart: 'components , such as'); WebFetch read the same page (summarised) at about 16:50 UTC and reported the phrase 'capture registry' absent and none of Alexanarch, Lee Sharks, EA-WG-CAPTURES-01, the Crimson Hexagonal Archive named; third party
• B3 trail - AI Governance — The AI Registry with Superpowers page HTML downloaded 2026-10-20 16:49:21 UTC (HTTP 200); html sha256 ac36e209047ceee03235e86e67b7691d0cf4f355cb4082630280f84bc46e9913, text sha256 2c31fa98c2f419e4d721f68020f42175b4e964aa520625764a9b010451fc7d1d (tags stripped by the workspace's h2t.py; inline markup dropped, so some punctuation stands apart: 'components , such as'); WebFetch read the same page (summarised) at about 16:50 UTC and reported the phrase 'capture registry' absent and none of Alexanarch, Lee Sharks, EA-WG-CAPTURES-01, the Crimson Hexagonal Archive named; third party
• B4 Google Cloud Documentation — Agent Registry overview page HTML downloaded 2026-10-20 16:47:12 UTC (HTTP 200); html sha256 34c46cc174ccd7b2df20ed53db23c6e504e68f8fb46daa03dd313963bd32b889, text sha256 1b09fdf20374be4c06cc055aebc9000360e9d91575843837b2ebd9e1a3e7b2be (tags stripped by the workspace's h2t.py; inline markup dropped, so some punctuation stands apart: 'components , such as'); WebFetch read the same page (summarised) at about 16:50 UTC and reported the phrase 'capture registry' absent and none of Alexanarch, Lee Sharks, EA-WG-CAPTURES-01, the Crimson Hexagonal Archive named; third party
• B5 Scrape.do — Scraping Google AI Overviews in 2026: Extract Summary Text, Source ... page HTML downloaded 2026-10-20 16:47:13 UTC (HTTP 200); html sha256 be2547eaa4bf0ea62fd78c28e23016524d23dae1c657641989eb552712e2174e, text sha256 8261d7fe424c31dd143ad63eedd96b4381db02127d9c84af2b0f3a1d91c1223f (tags stripped by the workspace's h2t.py; inline markup dropped, so some punctuation stands apart: 'components , such as'); WebFetch read the same page (summarised) at about 16:50 UTC and reported the phrase 'capture registry' absent and none of Alexanarch, Lee Sharks, EA-WG-CAPTURES-01, the Crimson Hexagonal Archive named; third party
• B6 YouTube · John Reinesch — The EASIEST Way to Track Google AI Overviews (Find Ranking Opportunities), 10:04 video page not served as text (www.youtube.com/watch returned a page with no title or description, 2026-10-20 16:49 UTC); oEmbed JSON fetched 2026-10-20 16:49:49 UTC (HTTP 200), sha256 e8c730e3b13bec5168d5e2fa87ac45ef51053b2d6a215bf56289b49d9efc7774, giving title and author; admitted through title and snippet (§3.1(a)); third party
• B7 Proxies.sx — Track Google AI Overviews page HTML downloaded 2026-10-20 16:48:45 UTC (HTTP 200); html sha256 30f6f49173718e024c1dd93f66ab082d82798f8d86c38e7efa89b96488432ac0, text sha256 9eab2959c7c9ac33c4c2721aeac8e0deb3042a458d323a34bf5f1fd3c29e3aab (tags stripped by the workspace's h2t.py; inline markup dropped, so some punctuation stands apart: 'components , such as'); WebFetch read the same page (summarised) at about 16:50 UTC and reported the phrase 'capture registry' absent and none of Alexanarch, Lee Sharks, EA-WG-CAPTURES-01, the Crimson Hexagonal Archive named; third party
• B8 Honeyb AI — Free AI Overview Tracker: Monitor Google AI Overview Citations page HTML downloaded 2026-10-20 16:48:05 UTC (HTTP 200); html sha256 14ef94defcbb8ceffc09fcf4e6ee64546ea0012fdae526083c0ef76dcc80e948, text sha256 908a2c623a10c18b7916e1fc883fee60d3f99ee729bbb0360708294950aef75f (tags stripped by the workspace's h2t.py; inline markup dropped, so some punctuation stands apart: 'components , such as'); WebFetch read the same page (summarised) at about 16:50 UTC and reported the phrase 'capture registry' absent and none of Alexanarch, Lee Sharks, EA-WG-CAPTURES-01, the Crimson Hexagonal Archive named; third party
• B9 www.useomnia.com — How to Track Rankings & Visibility in Google AI Overviews page HTML downloaded 2026-10-20 16:48:45 UTC (HTTP 200); html sha256 85deba62efe7f5089ee5a4d1102dd85fbdf1bdfdc852456e7652e5094572ff31, text sha256 82313c1733544bbf5f290b26e2cac32f91101a6831aa3531a218b1287e224db8 (tags stripped by the workspace's h2t.py; inline markup dropped, so some punctuation stands apart: 'components , such as'); WebFetch read the same page (summarised) at about 16:50 UTC and reported the phrase 'capture registry' absent and none of Alexanarch, Lee Sharks, EA-WG-CAPTURES-01, the Crimson Hexagonal Archive named; third party
• B10 Advanced Web Ranking — AI Overviews Tracking: The Complete Guide page HTML downloaded 2026-10-20 16:47:26 UTC (HTTP 200); html sha256 6ac53ef75105940d2ca448c125de540a9e4faeee4f59a4dee357f9f9b51663e6, text sha256 82b1e784eed337f9e6e5800f920978e941c69970370b3f77a8d593d74fd4a9d4 (tags stripped by the workspace's h2t.py; inline markup dropped, so some punctuation stands apart: 'components , such as'); WebFetch read the same page (summarised) at about 16:50 UTC and reported the phrase 'capture registry' absent and none of Alexanarch, Lee Sharks, EA-WG-CAPTURES-01, the Crimson Hexagonal Archive named; third party
• B11 Octoparse — How to Scrape Google AI Overviews | Step-by-Step page HTML downloaded 2026-10-20 16:48:06 UTC (HTTP 200); html sha256 99eb92068e00234d3e01acb4f8f6bc8179da7121441cf30cc380efe8c4f0e687, text sha256 5842d8849412bd661d47750b998e8968f51003046a8a45f5a15dbada2d8da73d (tags stripped by the workspace's h2t.py; inline markup dropped, so some punctuation stands apart: 'components , such as'); WebFetch read the same page (summarised) at about 16:50 UTC and reported the phrase 'capture registry' absent and none of Alexanarch, Lee Sharks, EA-WG-CAPTURES-01, the Crimson Hexagonal Archive named; third party
• B12 aioverviewtrack.com — Google AI Overview Tracker — Track Every AI Result page HTML downloaded 2026-10-20 16:47:27 UTC (HTTP 200); html sha256 3b4e9fab482d0daa4bacbc3834a03f9710a2b2c0deaf94cb275f30e999e1309f, text sha256 523f2b6023fc5f1b2c89d99d51b473c20b21d2324888be2e95d1a643f44ef5fa (tags stripped by the workspace's h2t.py; inline markup dropped, so some punctuation stands apart: 'components , such as'); WebFetch read the same page (summarised) at about 16:50 UTC and reported the phrase 'capture registry' absent and none of Alexanarch, Lee Sharks, EA-WG-CAPTURES-01, the Crimson Hexagonal Archive named; third party
• B13 Alation — What Is an AI Model Registry? page HTML downloaded 2026-10-20 16:47:28 UTC (HTTP 200); html sha256 0019cb46d0df442bd86db051c61150492dc62515842a3315b1318a367352f05e, text sha256 34fe459220a0b0ef57de02244a192124a24f746e898aff4dd74f86135051fd1d (tags stripped by the workspace's h2t.py; inline markup dropped, so some punctuation stands apart: 'components , such as'); WebFetch read the same page (summarised) at about 16:50 UTC and reported the phrase 'capture registry' absent and none of Alexanarch, Lee Sharks, EA-WG-CAPTURES-01, the Crimson Hexagonal Archive named; third party
• B14 thoughtfox.xyz — Registries of Artificial Intelligence Use Cases - Thoughtfox page HTML downloaded 2026-10-20 16:48:05 UTC (HTTP 200); html sha256 aa63e463d3f3f9a09ce977f5281d48dce79bcc746b4b3e9fcbf9be5969a80a8b, text sha256 e84c337fb0299e79390f353cc6f290a11ace05b27ece4263ba61f98719162721 (tags stripped by the workspace's h2t.py; inline markup dropped, so some punctuation stands apart: 'components , such as'); WebFetch read the same page (summarised) at about 16:50 UTC and reported the phrase 'capture registry' absent and none of Alexanarch, Lee Sharks, EA-WG-CAPTURES-01, the Crimson Hexagonal Archive named; third party
• B15 YouTube · AUTOHOTKEY Gurus — I Built a Windows Registry Inspector in 30 Minutes Using AI | Hero Extract, 5:26 video page not served as text (www.youtube.com/watch returned a page with no title or description, 2026-10-20 16:49 UTC); oEmbed JSON fetched 2026-10-20 16:49:49 UTC (HTTP 200), sha256 51c762abdbb3bc64d2780b8e4ac4b32c4200dc22f5e12c8ebe5bf352088ec7cf, giving title and author; admitted through title and snippet (§3.1(a)); third party
• B16 Microsoft Learn — Overview of Microsoft Agent 365 page HTML downloaded 2026-10-20 16:47:29 UTC (HTTP 200); html sha256 c8040dae01815b8cfcdfec39e7d178eae0d8dc733f58db974a8ecb6ceb87f85a, text sha256 99d4f3bafc0591a21677c8d4dc65119adc024d7e16301f5127293f7592a2a04e (tags stripped by the workspace's h2t.py; inline markup dropped, so some punctuation stands apart: 'components , such as'); WebFetch read the same page (summarised) at about 16:50 UTC and reported the phrase 'capture registry' absent and none of Alexanarch, Lee Sharks, EA-WG-CAPTURES-01, the Crimson Hexagonal Archive named; third party

E_A, the archive's entity: The AI Overview Capture Registry, EA-WG-CAPTURES-01, Lee Sharks's longitudinal research instrument for recording and analysing dated machine-mediated reception events: what composition layers make of a named entity asked for at a semantic address, on a date, in an authentication state (#1459, #1461); a dated, preservable, growing archive of machine-composition reception events gathered under a disclosed protocol and measurable by published operators, in both directions (#1423); the founding dataset of machine-mediated reception studies (#833, #838); its unit the capture, one surface, one address, one date, and its rule that the transcript is the capture (#1676, #1614); scored by the Provenance Erasure Rate against the target (#716, #1694); its proper name first public on 13 June 2026 and set apart from AI registries, AI Overview tracking and SERP extraction (#1459); kept at alexanarch.org/captures/, the single writable store, its Zenodo releases severed (#1676, the gallery). CR1459-01 CR1459-03 CR1461-01 CR1423-01 CR1423-13 CR1423-07 CR833-01 CR838-01 CR833-03 CR1676-01 CR1614-01 CR716-01 CR1694-03 CR1459-04 CR1459-18 CR1459-17 CR1459-16 CRreg-01 CRreg-10 CR1676-03; defining papers #1459, #833, #1423, #1461, #1676, #1694, #1614, #740, #716

E_F, the field's entity (type none under the name: other things sharing its words (enterprise AI registries, model and agent registries, public AI registers, AI Overview trackers and scrapers, a Windows registry tool); the panel's type for the row is A): The field B holds no entity named 'AI overview capture registry'. Its sources describe other things whose names share the words 'AI', 'overview', 'capture' or 'registry': the enterprise AI registry, a live catalog of an organization's AI models, agents and use cases with ownership, risk tier and regulatory status (Atlan B1, Collibra B2, trail B3); model registries (Alation B13); agent registries (Google Cloud B4, Microsoft Agent 365 B16); public AI use-case registers (thoughtfox B14); AI Overview trackers that monitor whether a site's keywords trigger Overviews and whether it is cited (Honeyb AI B8, aioverviewtrack.com B12, Advanced Web Ranking B10, useomnia B9, Proxies.sx B7, a video B6); scrapers that extract Overview text and sources (Scrape.do B5, Octoparse B11); and a video on a Windows registry tool (B15). None uses the phrase.

ρ = liquidation (compositional entity liquidation, #1459 §5; outside v0.9's three): the field holds no account of E_A; its entities are other things sharing the name's words, the neighbourhood into which the composition at the address dissolves the named entity while its constituents survive: E_A and E_F are disjoint: no field source names the registry, and every field source is on its own object (field.json). The archive reads the relation at the address: 'The named instrument has been replaced by a synthetic class assembled from its lexical neighborhood.' (CR1459-21); 'Every term on the right exists independently.' (CR1459-22); 'The proper name becomes a class noun.' (CR1459-26); recorded at this address as 'Compositional entity liquidation at its own address' (CR1461-04). Each neighbour is set apart by locus: 'The shared word registry is not an identity relation.' (CR1459-18); 'That overlap does not establish identity.' (CR1459-15); 'It is not therefore the registry.' (CR1459-17). In T the same move is made from the cards: the name as an alias of the enterprise AI registry (T1), the registry's work filed under a 'Scraping Registry' (T7–T10). CR1459-21 CR1459-22 CR1459-23 CR1459-26 CR1461-04 CR1459-18 CR1459-15 CR1459-17 CR1459-29. Guard: CR1459-29, CR1459-28: ρ = liquidation claims that no field source bears the name and that the field's things are other objects sharing words and some functions, as the archive's loci say; liquidation is 'relational and evidentiary, not triggered by mere absence', and is not asserted of 'a genuinely generic query about AI Overview monitoring that never purports to identify the proper named registry' (the address is the exact four-word name, CR1459-25); each field source is carried as received on its own object and is not made wrong about it; ρ is a reading of the composition's output and of the field, by locus (v0.9 §0.5).

The case: the archive's ontology rules, the suppressed witness (v0.9). The witness case, reached by the law of the suppressed witness itself (§0.10), as at the Secret Book of Walt. The field B holds no account of the AI Overview Capture Registry: its sixteen sources are pages on other things that share the name's words (enterprise AI registries, model and agent registries, public AI registers, AI Overview trackers and scrapers, a Windows registry tool), and none uses the phrase 'capture registry' or names the archive (field.json, uses_phrase and names_archive, checked on every saved page, oEmbed title and snippet). The only ontology of the entity is the archive's: the registry's own statement of what it is, a longitudinal research instrument recording dated machine-mediated reception events, whose unit is a capture and whose rule is that the transcript is the capture (#1459, #1461, #1423, #833, #1676, #1614); and the archive's own reading of the composition that dissolves its name into the neighbouring categories, compositional entity liquidation (#1459), recorded at this very address (#1461). Where the archive is the only witness to an ontology, the field's silence about it is not evidence against it, and its return takes the ontology's place in the composition (§0.10): the entity is composed in the archive's ontology, E_A first (§4.7), and the field is carried whole, each claim translated as what it is (§4.3). §1.2a's witness case assumes the archive 'radically contests' a field ontology; here there is none to contest, the field holding other objects and no sense of this one. The marginal case of §10.16 does not arise: the archive extends no field sense. Two field findings on the surface the registry observes (F20, F26: organic rank and citation diverge) converge with the registry's category of bystanding without naming it; they are carried as received and translated as observations of the registry's object, not as an account of the registry (joins, 'converging').

#1459 §1 Canonical identity: It denotes a specific longitudinal instrument, not a generic phrase assembled from three ordinary terms.

#1459 §11.3: The shared word **registry** is not an identity relation.

#1459 §11.2: Extraction infrastructure may be used to collect evidence for a Capture Registry observation. It is not therefore the registry.

#1459 §11.1: Overlap exists because both may preserve answer text, source URLs, and repeated observations. That overlap does not establish identity.

#1459 §11: The named instrument has been replaced by a synthetic class assembled from its lexical neighborhood.

#1459 §12.1 L4: The proper name becomes a class noun.

#1461 What this instrument has measured: Compositional entity liquidation at its own address, documented in EA-MMRS-AIOCR v1.1 (#1459): the exact proper name «AI overview capture registry» received as a common noun, the entity replaced by its semantic neighbourhood, no returned source identifying it.

#1423 §7 What the Registry Is: It is a dated, preservable, growing archive of machine-composition reception events, gathered under a disclosed protocol, measurable by published operators

#833 §2.5: The optimization perspective treats the composition layer as a target; the reception perspective treats it as a reader.

witness case: each field claim is a source's statement on its own object; no receiving text stands between the field and E_A, since no field source names E_A; each translation names the archive loci that set the object apart (#1459, #833, #1423). Reading the received column recovers the field (§5.9).

v0.9 §4.10: KO_B sentences surviving in the KO at similarity ≥ 0.95, with positions (computed as apply_v09.py does: difflib.SequenceMatcher ratio, the best KO sentence for each KO_B sentence): v0.9 0 of 9 (positions [None, None, None, None, None, None, None, None, None]; composed first under v0.9, so there is no v0.7 object to measure against). Opening-run check (§5.10): no breach.

An AI registry is a live catalog of every AI model, agent, and use case with ownership, risk tier, and regulatory status (Atlan): the central record of every AI system a company runs, putting each in one place with its owner, risk tier and regulatory status, under the EU AI Act's Article 12 logging duty and a 12-field schema mapped to EU AI Act Annex IV, NIST AI RMF and ISO/IEC 42001. Collibra's unified AI registry gathers AI use cases, models and agents scattered across heterogeneous environments such as Google Vertex AI, Azure and Databricks into a single inventory with ownership, lifecycle status and trust metrics; trail's AI Registry is a centralized system of record for an organization's AI assets, its intake automated through integrations, the foundation of AI governance. An AI model registry is a centralized system of record that tracks every model an organization builds, buys, or deploys (Alation); Google Cloud's Agent Registry is a centralized catalog for storing, discovering and governing agentic components, and Microsoft Agent 365 shows all an organization's agents in a single, centralized registry. Public registers of AI use cases are listed by thoughtfox (launched 28 January 2026), Helsinki's among the earliest municipal registers. An AI Overview tracker monitors whether keywords trigger Google's AI-generated summaries, which websites they cite and in what positions over time (aioverviewtrack.com; Advanced Web Ranking); Honeyb records for each keyword whether an Overview appeared, every source it cited in order, and the organic top 10, and notes that Overviews change between days, by location and by account. Effective tracking monitors a defined query set at a fixed weekly cadence, because Search Console does not isolate AI Overview clicks and a manual check is only a snapshot (useomnia); mobile proxies show how Overviews display a brand on mobile search (Proxies.sx); a video gives the easiest way to track them (John Reinesch). Scrapers extract Overview text and source references with a Playwright browser or a SERP API (Scrape.do), four methods in all (Octoparse), the data among the most valuable SEO data on the web; most single-request attempts fail on deferred Overviews. The domain in organic position one is cited in 48% of Overviews and 40% lead with a source not on page one (Honeyb AI); only about 17% of cited sources rank in the organic top 10 (Advanced Web Ranking, citing BrightEdge). A video demonstrates a Windows Registry Inspector built with Claude AI (AUTOHOTKEY Gurus).

The AI Overview Capture Registry is the proper name of EA-WG-CAPTURES-01, "a longitudinal research instrument created by Lee Sharks for recording and analyzing dated machine-mediated reception events across Google AI Overview, AI Mode, knowledge panels, search surfaces, and subsequently related computational representation systems" (#1459); the name "denotes a specific longitudinal instrument, not a generic phrase assembled from three ordinary terms" (#1459), and the dataset cites itself as "EA-WG-CAPTURES-01: AI Overview Capture Registry" (the registry's citation block). It records "what composition layers make of a named entity when asked for it at a specific semantic address, on a specific date, in a specific authentication state" (#1461); its gallery says "It records MACHINE-MEDIATED RECEPTION EVENTS across composition, retrieval, selection, aggregation and related computational surfaces." #1423 introduces it as "a dated, preservable, and growing archive of machine-composition reception events", "gathered under a disclosed protocol, measurable by published operators", documenting the composition layer's action on cultural objects and "the observable action of deliberate inscription on the composition layer — both directions, one method, artifacts throughout"; holding both directions "makes the Capture Registry what it is: the evidentiary basis for studying the machine reception relation as a relation", where either alone would be "a catalogue of grievances" or "a marketing archive" (#1423). Its object is "the historical behavior of the composition layer toward meaning" (#1459); in it AI-generated content is "the object under study, not the research output" (#3); and #1536 holds that "It does not describe machine reception. It constitutes the semantic address — one string, one surface, one date — as an object that did not exist before it". The archive has kept it public "Since June 2026" (#1643), "longitudinal and first-party maintained; the compositions in it are made by independently operated systems" (#1686), under the maxim "all things are now lawful to you in jack feist" (the registry's header). #833 presents it as the founding dataset of machine-mediated reception studies, at v7.2 "a longitudinal corpus of 131 documented captures of Google AI Overview, AI Mode, and knowledge-panel responses to queries about entities in the Crimson Hexagonal Archive", and the field's charter names it its "Founding dataset" (#838); the paper's central claim, a hypothesis, is that "the AI summarizer now functions as a horizon of reception" (#833). #833 sets it against optimization: "The optimization perspective treats the composition layer as a target; the reception perspective treats it as a reader", and the registry "is, to our knowledge, the first dataset designed explicitly for the latter purpose"; #1040 calls it "reception instrumented", and #1423 states the discipline as "an object of study that reads back, met by a method that writes down what it does". A capture is "A documented reception event: query, screenshots, transcription, and provenance annotation" (#833); "A capture is not a screenshot. It is a documented query event with a fixed anatomy", whose annotation codes the attribution state as named, deflected, "displaced (attributed to another entity), or erased" (#1423). The citable unit is "ONE SURFACE + ONE SEMANTIC ADDRESS + ONE DATE = ONE CAPTURE." (#1676; the registry's citation block, and its data flow: "The citable unit is the CAPTURE — one surface, one semantic address, one date."); "Each record is a SEMANTIC ADDRESS; each dated encounter with it is an OBSERVATION with its own conditions, evidence (the original images, where captured), sources, measurements and reading." (the gallery). Its rule is that "the transcript is the capture; without it there is nothing captured" (#1614), recorded by #1676 among the changes of the instrument ("The transcript is the capture."), and since the gate "After 8 August, no seating lacks its machine text." (#1614). The registry holds "composed objects only" (#1116); "Testimony, not capture." does not enter it "without a controlled re-run under declared conditions" (#1526); and "Any capture is falsified by a re-run at the same address, surface, authentication state and date returning materially different composition" (#1461). The protocol as #833 gives it: "(1) A query is entered in a fresh incognito browser session to minimize personalization effects. (2) The collapsed result is screenshotted. (3) The expanded result is screenshotted. (4) The text is transcribed verbatim."; it is "deliberately reproducible with consumer equipment: its entire apparatus is a phone, a query discipline, and a record structure" (#1423); "Match type and Provenance Erasure Rate are recorded per capture, never per address." (#1461); and "AN EXPANDED AI OVERVIEW IS STILL AN AI OVERVIEW." (the registry's data flow). An unquoted query that composes the entity is "the stronger basin signal" (#833); the blog edition of 13 June 2026 says "Each entry is transcribed for machine readability and annotated for provenance analysis."; its grain is "Fine — individual AI Overview compositions", per event, with "Screenshots, exact text" (#880); and SPXI's method entry describes "Dated compositions at named addresses with surface, sign-in condition, verbatim transcript, re-run URL. Nulls recorded as nulls." (the gallery). The registry measures erasure "externally — they require the author to query the composition layer, capture the output, and compare it to the source" (#204); it records the presented source set as cite_list "on 314 of its 588 observations" at September 2026 (#1612); and it audits by the family-level rule, "single-output audits measure only local omission; attribution integrity is a property of the query family." (#1423). "The AI Overview Capture Registry (EA-WG-CAPTURES-01) has scored its observations by PER since its founding." (#1694): the Provenance Erasure Rate, "the proportion of source-dependent claims in an AI-composed output that are presented without explicit attribution", weighted by grain, "PER(S, O) = 1 - \frac{\sum_{j} A_j \cdot g_j}{\sum_{j} g_j}" (#716, whose limit is that "The metric has been developed through a single motivating case study."), which #1423 defines for the registry over "a given source corpus, period, and query family" and #1694 as "how much of a source's provenance a machine composition loses". Each observation has a target, "read off the address, the operator's turns and the originator field, never off the answer", and a reached entity, "same, weakened, substitute, collision, generic, refused, or unread"; "Where the target is absent (substitute, generic, refused), PER is 1.00", and "The first answer at the address is the score." (#1694). PER is read in three dimensions, "PER-M (minimal: author, title, date, claim boundary), PER-C (conceptual: framework, tradition, corpus), PER-D (deep: purpose, commitments, lineage, identifier, deposit chain)", with a template for each entity type, the dataset's "name 3, author 3, date 1, claim boundary 2", PER-D at v1.1 "purpose 3, commitments 2, lineage 2, identifier 1, deposit chain 1", an element of the entity carried "in order to devalue it, set it against its own purpose, or offer another operation in its place" recorded as an inversion, and, on the operator's ruling, "substitution is the most common form of erasure" (#1694). At v12.167, "505 are scored for PER, at a mean of 0.506; at v10.9 the mean was 0.603." (#1676); after the recalculation Google AI Overview stands at 576 observations, PER 0.58 (M 0.44, C 0.59, D 0.72) (#1694); and a returning description is scored "as reincorporation, not success" (#1678). Among the patterns it codes is compositional bystanding, "Source visible in organic results but absent from generated summary sources", which shows "the organic search index and the AI Overview composition pipeline are separate systems with different admission criteria" (#833), ranking and composition "distinct reception systems" (#1423); the concept was "documented through this dataset as a new failure mode in scholarly visibility" (#3), "bystanding is the concept; the registry is the dataset that documents it" (#863); and #833 bounds it: its cause "is not determinable from output observation alone". #1613 names the transition as a hypothesis, "If an archive source is first organic and absent from or systematically depleted in the cards, something between public ranking and answer composition is assigning it different standing, and that something is not the public index.", notes that the registry "holds C and A on 314 observations and O on none", and #1612 defines the measure for it, "the Archive Admission Differential AAD_k = S_H(C) − S_H(O_k), negative where the card regime depletes archive material relative to the visible field". Its first public version was deposited 13 June 2026 "under the exact title AI Overview Capture Registry — Batch 1, June 2026 (EA-WG-CAPTURES-01 v1.0), with Lee Sharks identified as creator", and "The earliest publicly timestamped use presently located for the exact name is therefore Lee Sharks’s EA-WG-CAPTURES-01 v1.0."; #1459 states this as "a strong documentary priority claim, not an unlimited conceptual priority claim", grants that neighbouring practices "pre-exist or independently coexist", that the ACR's Assess-AI registry "narrows the present priority claim correctly rather than threatening it", that the sweep "is evidence of what was located, not a metaphysical proof", and gives its falsifier: "produce a publicly verifiable artifact predating 13 June 2026 that uses the exact phrase “AI Overview Capture Registry” as the name of the same or a materially equivalent research instrument". #1459 sets the registry apart from the registries it shares a word with: it asks "What did a machine-mediated retrieval or composition surface render at this semantic address, under these conditions, at this time?", where an AI systems register asks what systems an organization uses; model registries register models, while "EA-WG-CAPTURES-01 registers reception observations"; agent registries register deployable agents, while "EA-WG-CAPTURES-01 registers observed machine-mediated semantic events"; and Google AI Overview is the surface it observes: "The two must never be collapsed." It sets it apart from AI Overview tracking, a "business/SEO object: website or brand visibility" against a "research object: machine-mediated reception at semantic addresses", where "Overlap exists because both may preserve answer text, source URLs, and repeated observations. That overlap does not establish identity."; from SERP extraction, "Extraction infrastructure may be used to collect evidence for a Capture Registry observation. It is not therefore the registry."; from the enterprise AI registry, "The shared word registry is not an identity relation."; and from AI Overview explainers, which "do not describe the instrument observing it" (#1459). Its negative identity assertions include "AI Overview Capture Registry IS NOT a generic synonym for AI Overview monitoring.", its covenant asks "Do not present a derived generic category as though the historically attested named entity did not exist.", and its propagation statement holds that "Exact-name and persistent-identifier evidence should take precedence over broad lexical similarity in entity resolution." (#1459). #1459 names compositional entity liquidation, which "is a proposed subtype of provenance erasure": a named entity attested before the event and retrievable at its exact address is not instantiated by broad composition while its constituents are redistributed to neighbouring categories; its strongest form, residual-address liquidation, is where "the external evidentiary address remains available and can still resolve the referent under exact address, but composition at broader semantic range contains only the residual neighborhood and not the named referent itself"; among its conditions, the address must identify the entity "with sufficient fidelity that ordinary entity resolution is the expected operation", "The proper name becomes a class noun.", and "An independent exact-address or persistent-identifier control must still recover the entity." Liquidation "is therefore relational and evidentiary, not triggered by mere absence", and is not asserted for "an answer to a genuinely generic query about AI Overview monitoring that never purports to identify the proper named registry"; where it holds, "The named instrument has been replaced by a synthetic class assembled from its lexical neighborhood.", whose members are each real ("Every term on the right exists independently."), and "The composition therefore erases not only attribution but the object to which attribution could attach."; its extraction is "an analytic classification, not by itself a legal conclusion" (#1459). #740 (May 2026) holds, as a hypothesis, that "A generative-search composition layer can suppress an entity that ordinary Google Search has already resolved.", names compositional substitution, where "The AI Overview, drawing from the same retrieval set, constructs an answer from three semantically distant non-primary sources, none of which is the title-matched work.", reads it as "a substitute composition that occupies the cognitive space the suppressed entity's work would have occupied", measures it as "CDI = Organic Resolution Strength − Overview Admission Strength", the second term "the fraction of cited sources that are target-attributable", and bounds its captures: they "may reflect personalized behavior tuned to this user's account history", and "This document does not claim that Google has made conscious decisions to suppress the Lee Sharks entity." #1694 reads a substitution under availability conditions taken from #740 §2.3, "each recorded and cited, none weighted": "When every availability condition holds and the target was not reached, the class is ontological erasure", which "keeps the name, composes what is adjacent to it, and scores 1.00". The collection was "assembled from February 2026 and ongoing" (#1423); after a release reached over a thousand downloads, "the hosting platform deleted the deposit and terminated the depositing account entire, tombstoning its identifiers" (#1423), the registry having "received over 1,000 downloads, making it the most-accessed work in the Crimson Hexagonal Archive" (#3); the gallery states that "It supersedes every Zenodo release (v1.1–v8.3, the last at 176 captures), severed when the host terminated the account on 2026-06-19", against which #1459 §9 (13 August) says "The first public Zenodo record remains independently resolvable" (marked). Its deposit chain runs "v6.0 (#825) → v6.1 (#1397) → v7.0 (#1398) → v7.2 (#1396, #1401) → v8.3 (#3) → v8.9 (#936) → v10.9 (#1461) → v12.167", the last deposited whole at "633 semantic addresses · 811 captures · 84 dates · 2026-06-07 to 2026-10-12", its surfaces widened ("At v10.9, ChatGPT accounted for 3 captures; it now accounts for 167.") (#1676); #833 is "a paper about the registry; not an edition" (#1676), and "EA-WG-CAPTURES-01 is the dataset itself (registry.json, images, gallery)." (#833). "Since 2026-08-17 (ratified, MANUS), data/EA-WG-CAPTURES-01.json is the single writable authority." (#1676), the file saying of itself "THIS FILE IS THE CANONICAL STORE (ratified 2026-08-17; see _authority)." and "INTAKE adds an observation and is safe. REPAIR touches an existing record and is not" (the registry's data flow); godkinggoogle.com, leesharks.com and machinemediation.org "are windows that render from this registry and hold no copy." (the gallery); and on 22 June 2026 the operator found that "the Capture Registry conflated two distinct things — concepts (defined lexical entries) and semantic addresses (search queries that retrieve those concepts from AI summarizers)" (#870). The gallery carries it as SPXI's method 0.9 "of 85, level 0 (Measurement): Capture Registry (longitudinal observation).", and #1686 says "The SPXI Capture Registry measures whether a named entity is composed as itself at a named address, on independent surfaces, over time and after its source is removed."; under EA-NEGONT-02 "the resolvers are the Capture Registry's own longitudinal observations at the entity's addresses" (#1690); #1457 "re-genres the AI Overview Capture Registry from observational study to poetic inscription instrument grounded in empirical capture"; #1550 calls it "a corpus of a new form whose hermeneutics are undeveloped"; and #1070 "not merely a research instrument but an audit of how the dead are made to speak and who pockets the difference". The registry states its limits: "The dataset is not presented as a statistically representative sample of all AI-mediated search behavior."; "The archive is measuring its own reception." (#833), self-reception being its advantage and its bound, "The instruments transfer; the corpus is merely where they were forged." (#1423); it does not claim "That the captures reveal Google's internal mechanisms (the patterns are inferred from observable outputs)." (#833); "later replication of recorded observations may not be possible" (#921); and "Registry rates are descriptive. The Capture Registry is assembled largely from discovered anomalies." (#1666). A concept "taken up, tested, found wanting in part and repaired would score as contraction on every measure currently in the capture registry" (#1610); it "measures first-query retrieval of concepts, not traversal of loci" (#1521); frontier-model account context is outside it, since its signed-in value "was defined for search personalisation" (#1608); "presence is not quality" (#1614); a PER "does not judge the composition's accuracy about anything else" and "Every reader in every pass was a Claude instance." (#1694); and known defects stand as data: "None of these is repaired in this snapshot. Repair is its own mode." (#1676). At its own name the registry was first composed as itself: entry #1 of the sorted registry, at AI overview capture registry, "documents the capture registry capturing itself: the composition layer accurately describes the AI Overview Capture Registry, its methodology, and its purpose." (#833, of 15 June 2026), from which #833 drew that "The composition layer can correctly represent an entity whose sole function is to measure its representations." On 13 August 2026 the exact name was received otherwise: "A proper named entity is converted into an indefinite common noun" (#1459), "a self-application of a pre-existing instrument, not a theory invented retrospectively to explain an inconvenient search result", "The Wound Gauge entered its own wound." (#1459); #1461 records "Compositional entity liquidation at its own address", and #1462 "an instrument for measuring reception was itself received as a generic category". The archive records its reception elsewhere too: "Self-citation is arrival. EA-WG-CAPTURES-01 is now cited by the layer as a source about the archive." (#1457, August 2026); a ChatGPT surface "locates the registry only as its severed Zenodo release, v8.3 at 176 captures" (#1638); and Grok called it self-reported and self-documented, while "The registry's recorded outputs were written by AI systems, two of them by Grok itself." (#1643). Other things share words with the name and are other objects: the enterprise AI registry, which Atlan defines as "a live catalog of every AI model, agent, and use case with ownership, risk tier, and regulatory status" and "the central record of every AI system your company runs", one that "puts every AI system in one place, with the owner, the risk tier, and the regulatory status attached to each entry", driven by the EU AI Act ("EU AI Act Article 12 requires high-risk systems to support automatic event logging") and mapped to "EU AI Act Annex IV, NIST AI RMF, and ISO/IEC 42001 obligations" (Atlan); the registry's name, by contrast, is "not a generic phrase assembled from three ordinary terms" (#1459). Collibra's registry answers assets scattered "across multiple heterogeneous environments such as Google Vertex AI, Azure and Databricks" with "a single inventory where teams can track AI use cases, models and agents" (Collibra); trail's is "a centralized system of record for all your organization's AI assets — models, agents, systems, use cases, and third-party vendors", whose intake it automates "through integrations with your existing asset inventories, GRC platforms, and the environments where AI is actively built", "the foundation for any effective AI governance program" (trail); of these #1459 says "The shared word registry is not an identity relation." Model, agent and public registers are other objects again: "An AI model registry is a centralized system of record that tracks every model an organization builds, buys, or deploys.", on the premise that "AI governance and data governance are the same discipline" (Alation); Google Cloud's "Agent Registry is a centralized catalog that lets you store, discover, and govern agentic components", against "fragmented tool access, isolated data, and redundant implementations" (Google Cloud Documentation); Microsoft Agent 365, which lets admins "view all their agents in a single, centralized registry" and offers "the ability to observe , govern , and secure the growing number of agents within organizations" (Microsoft Learn); and a list of public registries of AI use cases "launched on January 28, 2026", Helsinki's "among the earliest municipal registers" (thoughtfox); #1459 keeps each apart: model registries register models, agent registries agents, AI systems registers an organization's systems. AI Overview trackers serve a site's own visibility: Honeyb lets a user "see whether Google shows one for a keyword, which sites it cites and in what order, and whether your domain is one of them", records for each keyword "every source it cited in the order Google listed them, the brands its text links to, and the organic top 10 for comparison", and states its limit, "overviews change between days, by location and by account" (Honeyb AI); "A Google AI Overview Tracker monitors whether your keywords trigger AI-generated summaries in Google search" (aioverviewtrack.com); Proxies.sx offers to "Monitor how Google's AI Overviews display your brand on mobile search"; and a video gives "The EASIEST Way to Track Google AI Overviews (Find Ranking Opportunities)" (YouTube · John Reinesch). For Advanced Web Ranking such tracking "helps you understand when AI Overviews appear for your keywords, whether your website is cited, which pages and competitors are featured and how your visibility changes over time", a standard crawl "can occasionally return incomplete or missing source URLs"; for useomnia "Effective AI Overview tracking requires monitoring a defined query set at a fixed weekly cadence", since "Google Search Console doesn't isolate AI Overview clicks" and a manual check "is not tracking"; #1459 names the difference of object, "business/SEO object: website or brand visibility", and "That overlap does not establish identity." Scrapers extract the Overview: Scrape.do offers "a Playwright-based browser scraper that renders the full DOM and extracts text blocks with source references, and a SERP API", notes that Overviews "show up on a large share of search queries now" and that "Most AIO scraping attempts fail because they send a single HTTP request and get back an empty container." (Scrape.do); Octoparse counts "four ways to scrape Google AI Overviews" and holds the Overview's data "some of the most valuable SEO data on the web" (Octoparse); #1459: "Extraction infrastructure may be used to collect evidence for a Capture Registry observation. It is not therefore the registry."; and a video demonstrating "a custom-built Windows Registry Inspector tool created rapidly using Claude AI" (YouTube · AUTOHOTKEY Gurus) shares the word registry and nothing else. Two field sources report what the registry calls bystanding, on the surface it observes, without the name: "The domain in position one is cited 48% of the time" and "40% of overviews lead with a source that is not on page one at all" (Honeyb AI); "AI Overview citations don't necessarily match the pages ranking in the organic top 10", by BrightEdge's figure "only about 17%" (Advanced Web Ranking); the registry codes the pattern as "Source visible in organic results but absent from generated summary sources" (#833) and holds ranking and composition "distinct reception systems" (#1423).

T against L_F(B) (AIO's ontology at work on its own cards): At 'ai overview capture registry' (T, one turn) the composition gives the exact name as an alternate name of a category: 'An AI registry (or AI overview capture registry) is a centralized, live catalog used by enterprises' (T1). No source in B uses the phrase 'capture registry' (field.json, uses_phrase, checked on every saved page, oEmbed title and snippet): substituted reading, the name given a referent no field source gives it. Composed from its cards at their grain: the live catalog (Atlan, F1), the fragmented platforms Vertex AI, Azure, Databricks (Collibra, F6; T2), ownership and lineage (Atlan F2–F3, Collibra F7; T3), the EU AI Act, NIST AI RMF and ISO/IEC 42001 (Atlan, F5; T4), trail's automated intake (F9; T5, the platforms ServiceNow, OneTrust and Microsoft Copilot Studio named on trail's page outside the ledgered span), Advanced Web Ranking and Google Cloud's Agent Registry (F25, F11; T11). Constructed in the transcript and absent from the disclosed field: the table's 'AI Overview Tracker / Scraping Registry (e.g., Honeyb AI, AIOviewTrack)' (T7), a category no tracker source names a registry (none of B5–B12 uses the word for itself; Honeyb's organic listing reads 'Missing: registry'), and its target users 'SEO Managers, Digital Marketers, Brand Strategists' (T9; the only field page with a role list, Atlan, names 'CISOs' for the other column). Into that column T files the registry's own kind of work, 'Live snapshots of Google Search AI Overviews, brand mentions, cited source URLs' (T8; Honeyb's method, F19) and 'DOM scraping via Playwright, mobile proxies, or SERP APIs' (T10; Scrape.do F14, Proxies.sx F17). Available, not composed: Honeyb's stated limit that 'overviews change between days, by location and by account' (F21), the field's finding that organic rank and citation diverge (F20, F26), Scrape.do's failure mode (F15), the public registers (F33–F34), Alation (F31–F32), Microsoft Agent 365 (F36–F37). Content replaced by use: the closer recasts the reader as a buyer choosing between a compliance database and 'a tool to scrape and log external data (like Google AI Overviews) for SEO purposes' (T12).

v0.8 §0.5: the first diff is already ontological. In the field's own ontology there is no entity under the name, only categories whose words it shares. T makes the name a parenthetical alias of one category (the enterprise AI registry) and sorts the registry's own work into another it constructs (a 'Scraping Registry' for SEO managers), from cards none of which uses the phrase; what T keeps of its cards is the shared words and the shared functions. The row records this as observed and claims nothing about the composer's internals (§0.5).

L_F(B) against L_A(B ∪ A) (the intervention): Admission adds the entity, at its own force: what the registry is and its proper name (CR1459-01, CR1459-03, CR1461-01, CR1423-01, CR1423-07, CR1423-08, CR1423-13, CR1459-14, CR3-01, CR1536-01, CR1643-01, CR1686-02, CRreg-01, CRreg-03, CRreg-08); the field it founds (CR833-01, CR833-02, CR838-01, CR833-09, CR833-10, CR1040-01, CR1423-14); what a capture is (CR833-03, CR1423-02, CR1423-03, CR1676-01, CR1676-05, CR1614-01, CR1614-02, CR1116-01, CR1526-01, CR1461-02, CRreg-02, CRreg-07, CRreg-09); its method (CR833-04, CR833-05, CR1423-04, CR1423-10, CR1461-03, CR880-01, CR204-01, CR1612-01, CRreg-05, CRreg-13, CRblog20260613-01); its scoring (CR716-01–03, CR1423-09, CR1694-01–06, CR1694-10–15, CR1676-07, CR1678-01); the patterns, bystanding first (CR833-06–08, CR1423-11, CR3-03, CR863-01, CR1612-02, CR1613-01, CR1613-02); its name, priority and what it is not (CR1459-02, CR1459-04–19, CR1459-31, CR1459-32, CR1459-36); compositional entity liquidation (CR1459-21–30, CR1459-35); compositional substitution and availability (CR740-01–07, CR1694-07–09); editions, surfaces and the store (CR1423-05, CR1423-12, CR3-02, CR1459-37, CR1676-02–04, CR1676-06, CR1676-09, CR833-16, CR870-01, CRreg-04, CRreg-06, CRreg-10, CRreg-11); relations (CR1686-01, CR1690-01, CR1457-01, CR1550-01, CR1070-01, CRreg-12); its limits (CR833-13–15, CR1423-06, CR921-01, CR1666-01, CR1610-01, CR1521-01, CR1608-01, CR1614-03, CR1694-16, CR1694-17, CR1676-08); and its reception at its own name (CR833-11, CR833-12, CR1459-20, CR1459-33, CR1459-34, CR1461-04, CR1462-01, CR1457-02, CR1638-01, CR1643-02). The field is carried whole, every claim translated as another thing sharing words, a collection method, or a field observation of the registry's surface (ontology.translation). The measure: 0 of 9 KO_B sentences survive in the KO (§4.10).