Generative AI and Jobs: A Refined Global Index of Occupational Exposure

register entry ilo-generative-ai-and-jobs-index-aio-20261029 OBS-79a0cee7426a · file as first run sha256 d92a8850603766d9…

• T1-1 definition: 'The International Labour Organization (ILO) published its updated assessment in the report "Generative AI and Jobs: A Refined Global Index of Occupational Exposure" (ILO Working Paper 140).'
• T1-2 global exposure: 'About 24% (1 in 4) of global workers are in occupations with some level of exposure to generative AI (GenAI).'
• T1-3 highest category: 'Only 3.3% of global employment falls into the highest exposure category where a large majority of tasks can be heavily impacted.'
• T1-4 transformation: 'The research emphasizes job transformation rather than widespread job elimination or mass redundancy, as very few occupations consist entirely of automatable tasks.'
• T1-5 income: 'High-income countries (HICs) show a 34% total GenAI exposure rate, compared to just 11% in low-income countries (LICs), reflecting differences in economic structures and digital adoption.'
• T1-6 gender: 'Women are nearly twice as likely to hold jobs exposed to GenAI as men (roughly 29% vs. 16% across sampled countries), largely because women are more heavily represented in clerical, administrative, and routine information-processing roles.'
• T1-7 most exposed: 'Clerical support workers, administrative staff, secretaries, receptionists, data entry operators, and translators face the highest exposure.'
• T1-8 least exposed: 'Manual trades, construction, agriculture, and heavy manufacturing involve physical and non-routine tasks with very low GenAI exposure.'
• T1-9 method, task level: 'Evaluates nearly 30,000 to over 52,000 individual task data points across the International Standard Classification of Occupations (ISCO-08) framework.'
• T1-10 method, gradients: 'Combines worker surveys, expert Delphi panel inputs, and large-scale AI assessments to categorize jobs into four distinct exposure gradients based on mean exposure scores and task variability.'
• T1-11 closer: 'Would you like to explore: • Specific exposure scores for a particular occupation or industry? • How GenAI impacts administrative versus technical roles? • Policy recommendations from the ILO for workforce upskilling?'

• B1 International Labour Organization — Generative AI and Jobs: A Refined Global Index of Occupational Exposure | International Labour Organization card; B1.html sha256 f53847a18e44fe9c, HTTP 200 at 2026-10-29T22:30:26Z
• B2 International Labour Organization — Generative AI and jobs: A 2025 update | International Labour Organization card; B2.html sha256 d01982bba323de96, HTTP 200 at 2026-10-29T22:30:28Z; byte-identical to the census copy ilo-genai-exposure-index--brief
• B3 LinkedIn·Grace S. Thomson — A Refined Global Index of Occupational Exposure | Grace S. Thomson card; unreachable (unreachable: no URL found); its title and snippet its only text
• B4 NASK – Państwowy Instytut Badawczy — ILO-NASK Indeks card; B4.html sha256 54770c9c03391c09, HTTP 200 at 2026-10-29T22:31:06Z
• B5 The People Space — What ILO's Global Index tells us about GenAI and jobs | The People Space card; B5.html sha256 674448bff10dc4e5, HTTP 200 at 2026-10-29T22:30:31Z
• B6 YouTube·International Labour Organization — Generative AI and Jobs: A global analysis of potential effects on job quantity and quality. card; unreachable (unreachable: no URL found; a video card is admitted through title and snippet (§3.1(a)), and this card has no snippet); its title its only text
• B7 YouTube·International Labour Organization — How is generative AI reshaping gender inequalities at work? card; unreachable (unreachable: no URL found; admitted through its title); its title its only text
• B8 www.scribd.com — Generative AI and Jobs - A Refined Global Index of Occupational Exposure - Scribd card; unreachable (unreachable: no URL found); its title and snippet its only text
• B9 International Labour Organization — Generative AI and Jobs: A Refined Global Index of Occupational Exposure (ILO Working Paper 140, HTML edition) own_page; B9.html sha256 f33cc70351ee9068, HTTP 200 at 2026-10-29T21:49:13Z; copied from the census's saved copy
• B10 International Labour Organization — Observatory on AI and Work in the Digital Economy own_page; B10.html sha256 402a902dac92f13a, HTTP 200 at 2026-10-29T21:49:12Z; copied from the census's saved copy
• B11 Wikipedia — Generative AI and Jobs: A Refined Global Index of Occupational Exposure (no article; cited in four articles) wikipedia; no article (raw wikitext and rendered page HTTP 404 for the title, the Global Index's name and the capture's query); four citing articles fetched as raw wikitext, HTTP 200, 2026-10-29T22:32Z

E_A, the archive's entity: the same index, in the Labor Vector census record and the archive's deposits: an exposure estimate placed at the intersection of the distribution of gains and losses, under the question whose labor, directed by whom, to whose gain; it estimates where task labor could pass to a GenAI system on grounds of technical feasibility, meets task-origin only as a prior on which tasks are exposed, and observes no performed work, so cannot speak to a system's labor on a commissioned task or to user labor drag; who supplies, directs and gains it records as 'not recorded' as observed; what it settles (which tasks are scored feasible, what share of employment is exposed) is kept apart from what it does not (observed change, causation, direction, gains, use); of its class, the archive holds that AI integrates the means of semantic production and automates its labor under concentrated ownership, that displaced labor is relocated and enclosed, not vanished, and that an account of AI job loss narrated as mechanism is a scenario with a redistributive resolution LV-02 LV-05 LV-06 LV-07 LV-08 LV-10 LV-11 LV-12 LV-13 LV-28 LV-30 LV-34 LV-35 LV-41 LV-42 LV-49 O499-01 O533-01 O521-01 O520-01; defining papers #499, #520, #521, #533

E_F, the field's entity (type C — a conventional reading holds it (public entity: an index of occupational exposure published as a working paper, with its research brief)): the ILO's refined global index of occupational exposure to generative AI (Working Paper 140, 20 May 2025, doi:10.54394/HETP0387, CC BY 4.0), made by the ILO's Research department with NASK, which updates the ILO's 2023 index: task-level automation scores from LLMs, a survey of 1,640 workers in Poland, expert review and an AI predictor, aggregated to ISCO-08 occupations by mean and spread into four exposure gradients, with employment shares by region, income and sex (one in four workers exposed; 3.3% in Gradient 4); transformation its most likely impact; its stated limits exposure not impact, an upper threshold, a static snapshot, a survey in Poland

ρ = extends: The archive reads the ILO index as its own sources present it (every field of the census record carries a verbatim quote from the paper, the brief or the Observatory) and adds, on the field's own object: the question the paper does not put to itself, whose labor, directed by whom, to whose gain, answered 'not recorded' as observed (LV-05, LV-10–LV-13); the distinction between technical feasibility and performed work, by which the index meets task-origin only as a prior and cannot speak to a system's labor on a commissioned task or to user labor drag (LV-34, LV-35); what the evidence settles and what it does not, an estimate set against observed trackers with no record establishing exposure as a cause (LV-28–LV-33, LV-48–LV-50); the negative tags (LV-41–LV-44); of the class, the automation of semantic labor under concentrated ownership (#499), displaced labor relocated and enclosed (#533), and the account of job loss as scenario with a redistributive resolution (#521, #520); and the composition's departures at the name (CAP-1-03–CAP-1-07). LV-05 LV-10 LV-11 LV-12 LV-13 LV-34 LV-35 LV-28 LV-30 LV-31 LV-32 LV-49 LV-50 O499-01 O533-01 O521-01 O520-01. Guard: ρ = extends claims an addition and no more (§1.2a). It does not make the field's sources wrong at their dates. Where an archive claim stands against a field claim (the record's eight authors against the paper's seven; NASK as expert group against NASK as preparer; transformation against displaced labor relocated), both are composed under P_coh within the one ontology.

The case: equal terms, ontology included (v0.9). The archive reads the field's entity as the entity. The Labor Vector record describes the index from its own pages, each field with a verbatim quote, and its additions are senses and limits on the same object: the record's question of supply, direction and gains, the distinction between feasibility and performed work drawn with the archive's instrument of labor direction, and the class claims on displacement. No ontology here is one the field cannot hold: the index's own limits already state that exposure is not impact, and the archive's reading extends that statement to the question of whose labor. Equal-terms case (§0.9): one claim space, nothing translated, neither ledger placed first.

#771 Disambiguation Matrix, POSITIVE 1: POSITIVE: Directionality of semantic labor measures the direction of a system's labor relative to the commissioned task across five layers (C, T, R, O, A).

#771 Aphoristic Tooth: Provenance is not only claim-origin; provenance is task-origin.

#499 §2.5 ¶1: increasingly automates the labor (generation replacing human semantic labor)

#533 'Labor bearing-cost': He does not note that the labor does not vanish. It is *relocated* — from visible human interpreters to invisible architectural constraints.

#521 ¶10: What follows is a scenario, not a prediction.

equal-terms case: every field claim composed as received, beside the archive claims it stands with. 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 match over KO sentences): v0.9 0 of 14 (positions [None, None, None, None, None, 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.

Generative AI and Jobs: A Refined Global Index of Occupational Exposure is the ILO working paper that "updates the ILO’s 2023 Global Index of Occupational Exposure to Generative AI (GenAI)" (the paper), refining "the global measurement of occupational exposure to generative AI" from task-level data, expert input and AI model predictions (ILO publication page). Its brief is "Generative AI and jobs: A 2025 update" (ILO brief page) and NASK calls it the ILO–NASK Index (NASK); it is the work of the ILO's Research department with NASK (the paper), prepared on NASK's page by the NASK Analysis and Research Office (NASK), its authors eight on the ILO's page and seven in the paper's title block (ILO publication page; the paper). It builds on the 2023 index of Gmyrek, Berg and Bescond (the paper; YouTube card), lowering the mean score to 0.29 from 0.30 (ILO brief page), revising 2023's clerical scores as "overly optimistic", and setting 20.5% for Gradients 1–3 against 2023's 29.2% (the paper). Its method combines three LLMs scoring each task from 0 to 1, a survey of 1,640 people employed in Poland giving 52,558 data points on 2,861 tasks, expert review in Delphi-style rounds, and an AI assistant that predicts scores for all tasks, its predictions correlating at 0.92 (the paper). Occupations are classed by the mean and standard deviation of their task scores into four gradients (the paper; ILO brief page), and employment shares estimated for the 112 exposed occupations (the paper). One in four workers worldwide is in an exposed occupation and 3.3% of employment in the highest gradient, 4.7% of female and 2.4% of male; exposure is 11% in low-income and 34% in high-income countries, 41% of female against 28% of male employment in the latter (the paper). The paper reads these patterns as occupational structure (the paper), The People Space as the gendered distribution of clerical roles and as infrastructure and digital access (The People Space), and an ILO video asks how generative AI is reshaping gender inequalities at work (YouTube card). Clerical occupations remain the most exposed, while digitized professional and technical occupations and media- and web-related tasks rose, and exposure declines with physical tasks (the paper; ILO brief page). Since nearly all occupations have tasks requiring human input, transformation is the most likely impact (the paper; ILO brief page; NASK), though efficiencies "may mean that fewer workers are needed" (the paper). The estimates "reflect “exposure” to GenAI and not the actual impact on employment", an upper threshold and a static snapshot at the start of 2025 (the paper), which The People Space carries as "exposure does not equal automation" (The People Space). The authors add that task scoring carries subjectivity, all tasks are weighted equally, the survey sample may not represent the working population, Poland is assumed an upper threshold, and the index is generic (the paper). On gains the paper cites economists that productivity benefits lie in complementing human expertise and not in labour savings, and asks for a transition managed through social dialogue (the paper), which The People Space quotes with words changed (The People Space). It was published on 20 May 2025 under CC BY 4.0 with its task-level scores public (ILO publication page; the paper; NASK; Scribd card) and sits on the ILO Observatory, which lists later work on exposure (ILO Observatory). Commentary calls it comprehensive (LinkedIn card; The People Space); Wikipedia has no article, and four articles cite the paper for claims on women, layoffs and enrolment (Wikipedia).

Generative AI and Jobs: A Refined Global Index of Occupational Exposure is an index of occupational exposure to generative AI, "published as a working paper with a research brief" (Labor Vector record), which in its authors' words "updates the ILO’s 2023 Global Index of Occupational Exposure to Generative AI (GenAI)" (the paper) and which, on the ILO's page, "refines the global measurement of occupational exposure to generative AI by combining task-level data, expert input, and AI model predictions" (ILO publication page). Its brief is "Generative AI and jobs: A 2025 update" (ILO brief page); NASK's page calls it the ILO–NASK Index, "developed to better understand how generative AI may affect work at the task level" (NASK); and the Labor Vector record lists among its names "ILO Working Paper 140" and the Observatory's "AI impact on jobs" (Labor Vector record). The Labor Vector record classes the evidence it brings as an "exposure estimate" and places it at the intersection of the "Distribution of gains and losses: wages, employment, workload, credit, ownership, and who captures productivity gains", under the question it puts to every resource it holds: "Who supplies the labor, who directs it, and who receives its gains — and what kind of evidence does this resource bring to that question?" (Labor Vector record). The paper gives its makers as the ILO's Research department "partnered with NASK, the National Research Institute of the Ministry of Digital Affairs in Poland" (the paper); NASK's page gives the report as prepared by the NASK Analysis and Research Office, the ILO having "played an active role in the preparation of this report" (NASK); and the Labor Vector record names the ILO "with NASK and the Polish Ministry of Family, Labour and Social Policy (expert group)" (Labor Vector record). Eight authors are listed on the ILO's publication page, Balint Nafradi among them, and seven, without him, in the paper's own title block (ILO publication page; the paper), the record following the eight (Labor Vector record). Its lineage is the ILO's first index of 2023, "the method developed by the ILO (Gmyrek, Berg and Bescond, 2023 - GBB hereafter)", which scored ISCO-08 task descriptions with GPT-4 (the paper; Labor Vector record, founded), and an ILO video on the rail carries that paper's title, "Generative AI and Jobs: A global analysis of potential effects on job quantity and quality." (YouTube card, title only). Its method opens with three LLMs "available on the market at the end of 2024: GPT-4, GPT-4o and Gemini Flash 1.5", each task scored from 0, where "the task cannot be automated", to 1, where "the task could be fully automated with GenAI technology (the task can be conducted without participation of a human operator)" (the paper). A survey of 1,640 people employed in Poland then scored a representative 2,861 of the 29,753 tasks of the Polish classification, 52,558 data points, each group's tasks "assessed by individuals working" in it (the paper), and the record keeps these counts apart, beside 3,265 ISCO-08 tasks rescored and 112 occupations classed as exposed (Labor Vector record), as the Capture Registry finds the composition at the name does not (Capture Registry, 2026-10-29). International experts, for whom "expertise in AI design or development is not a requirement", compared the scores in "several rounds of Delphi-style discussions"; the human scores became a repository from which "an AI assistant able to predict scores for tasks in the technical documentation of ISCO-08" was built, its ISCO-08 and Polish predictions correlating at 0.92, "demonstrating exceptional stability of predictions" in the authors' words (the paper; Labor Vector record, method). An occupation is classed by "two moments of distribution: occupational mean and standard deviation (SD) of task-level scores" (the paper) into "four progressively increasing gradients of GenAI exposure" (ILO brief page), Gradient 4 holding "the highest share of tasks exposed to potential GenAI-driven automation" with "a high consistency of this exposure across tasks"; employment shares are then estimated for "the 112 occupations identified as exposed" (the paper; Labor Vector record, unit). Globally "one in four workers are in an occupation with some GenAI exposure" and "3.3% of global employment falls into the highest exposure category", "female (4.7%) and male employment (2.4%)" (the paper; Labor Vector record, findings); the Capture Registry finds that the composition at the name, a new address where no card cites the record, names the paper and holds these figures, checked against the census record and its saved sources (Capture Registry, 2026-10-29). Exposure rises with income, "11% of total employment in LICs vs 34% in HICs", and the gap between the sexes with it: in high-income countries "28% of male employment is exposed, with only 3.5% in gradient 4, compared to 41% of female employment", "9.6% in gradient 4", and in Europe and Central Asia 39% of female employment against 26% of male (the paper; Labor Vector record). The paper reads these patterns as reflecting "occupational structures" and where roles "such as clerical, financial, and customer service jobs" are concentrated (the paper); The People Space reads the gender gap as "the gendered distribution of clerical and customer-facing roles" and the income gap as "shaped by infrastructure, digital access and workforce structure" (The People Space); an ILO video on the rail is titled "How is generative AI reshaping gender inequalities at work?" (YouTube card, title only). The Capture Registry finds the composition's gender figure, "roughly 29% vs. 16% across sampled countries", in no saved source, "the paper's 29.2%" being "the 2023 index's combined share of augmentation and 'the big unknown'" (Capture Registry, 2026-10-29); the paper gives the 2023 shares as 13% "augmentation" and "16.2% of employment in “the big unknown”", against 20.5% for its own Gradients 1–3 (the paper). "Clerical occupations continue to have the highest exposure levels", the majority of the 13 occupations in Gradient 4 among them, while exposure rose for "some strongly digitized occupations" (the paper; Labor Vector record), financial analysts and web and multimedia developers in Gradient 4 and bank tellers, translators, applications programmers and investment advisers in Gradient 3 (the paper), and for media- and web-related tasks with "voice, image and video generation" (ILO brief page); exposure "declines with tasks of physical nature" (the paper). Against 2023 the mean score is "0.29 in 2025 versus 0.30 in 2023" and the standard deviation "0.14 in 2025 v. 0.30 in 2023" (ILO brief page; Labor Vector record), and the authors revise their own 2023 scores for clerical tasks, which "reflected an overly optimistic assessment of full automation potential" (the paper). Because "nearly all occupations have some tasks that require human input", the paper holds that "transformation of jobs is the most likely impact of GenAI" (the paper), the brief that "most jobs will be transformed rather than made redundant" (ILO brief page; Labor Vector record, findings), and NASK that the index exists "not to predict the disappearance of jobs, but to identify where work is most likely to change" (NASK); the paper qualifies its own reading: "this does not mean that demand for these occupations will remain stable, as the efficiencies gained from introducing technology may mean that fewer workers are needed" (the paper). Of the class the record places the index in, the employment effects of AI, the archive states that a technology which "increasingly automates the labor (generation replacing human semantic labor)" is "the full vertical integration of semantic production", its resources "controlled by the same small set of corporations" (#499, Rex Fraction), and that labor platforms displace "does not vanish. It is relocated", the steward capturing it "as proprietary infrastructure" (#533, Johannes Sigil). The paper states what its estimates are not: they "reflect “exposure” to GenAI and not the actual impact on employment", represent "the upper threshold of the percentage of employment that could be affected if GenAI technology were fully implemented", leave out "new jobs that may be created" and future advances as "a static snapshot of the exposure of existing occupations to GenAI at the start of 2025", and do "not imply the immediate automation of an entire occupation" (the paper; Labor Vector record, stated limits and coverage), barriers such as "insufficient digital skills" and "the cost of technology" standing between exposure and adoption (the paper); The People Space carries the limit: "exposure does not equal automation" (The People Space). Its further limits, stated by its authors, are that the gradients "merely suggest the general trends" because "any type of task scoring system carries a degree of subjectivity", that "all tasks are weighted equally", that the survey sample "may not represent the entire working population" and is comparable only within each 1-digit group, that "For 91.5% of the tasks, respondents did not report performing a similar task in their work", that Poland is taken as "representative of the upper threshold of automation potential", and that the index "is generic" (the paper; Labor Vector record, stated limits). The record's negative tags for it are "NOT an observed employment outcome", "NOT a forecast of job loss: exposure is an upper threshold of potential", "NOT a measure of AI use at work" and "the worker survey was run in Poland only" (Labor Vector record); the Capture Registry finds those limits absent from the composition at the name, whose "can be heavily impacted" "stays inside exposure, and gives no limit", the index there "composed at its headline grain", and holds that the negative tags are "what separate this resource from the employment trackers beside it in the census" (Capture Registry, 2026-10-29). To its organizing question the record answers for this index that its labor object is "the tasks of occupations" and "the global employment in the 112 occupations classed as exposed", that who supplies the labor is "not recorded; the index scores tasks' technical feasibility for GenAI automation", its own scores "supplied by 1,640 people employed in Poland" and "by an expert group", that who directs it is "not recorded", and that gains are "not recorded as observed outcomes" (Labor Vector record). The paper's statement on gains is given "As argued by numerous economists": "the key to unlocking the productivity benefits of GenAI lies not in the search of outright labour savings, but in the extent to which human expertise can be complemented with new technological capacities" (the paper), a sentence The People Space quotes as the report's, without "outright" and without the economists (The People Space); the paper asks whether jobs can "retrain and retain existing workers, rather than drive labor churn through workforce replacement" (the paper), and The People Space names a divide "in who is most impacted, and who benefits" (The People Space). For the transition the paper asks for social dialogue "to minimize “technological unemployment”", holds that "Workers know their jobs best", and offers the index, linked to national micro data, as "a foundation for social dialogue and targeted policy responses to manage the transition" (the paper). In the archive's reading, "The index estimates where task labor could pass to a GenAI system on grounds of technical feasibility, task by task; it meets task-origin only as a prior on which tasks are exposed", and "It observes no performed work, so it cannot speak to whether a system's labor advanced a commissioned task, or to user labor drag" (Labor Vector record, lv_intersection). The record holds that this evidence answers which occupations have tasks "scored as technically feasible for GenAI automation" and what share of employment "sits in exposed occupations", and does not answer "observed employment, wage or hours change in any occupation", "whether exposure leads to job loss, transformation or new work", "who directs adoption or receives the gains from it" or "how GenAI is in fact used at work"; in its worked question the index is the estimate set beside the observed trackers, since "An exposure estimate scores what AI could do to a task; an observed outcome counts what happened to jobs", and "No record establishes that exposure causes the employment change" (Labor Vector record, questions). Of accounts of AI job loss the archive reads one, the Citrini memo's spiral in which "AI improves, companies cut workers, displaced workers spend less, companies invest more in AI", as "a scenario, not a prediction" narrated backward (#521, Rex Fraction), a crisis that "has a policy resolution (redistribute gains)" (#520, Rex Fraction). Published on 20 May 2025 under DOI 10.54394/HETP0387 and "Attribution 4.0 International (CC BY 4.0)", with "all task-level scores publicly available" and "the full dataset (csv)" on NASK's page (ILO publication page; the paper; NASK; Labor Vector record), copied on Scribd (Scribd card, snippet only), the index lives on the ILO Observatory on AI and Work in the Digital Economy as "AI impact on jobs", beside the ILO's platform policy tracker, and the Observatory now lists ILO Working Paper 166 (17 March 2026) and a brief on exposure indicators (17 April 2026), which the record leaves unchecked as to whether they revise this index, as it leaves the task-level scores unchecked as data (ILO Observatory; Labor Vector record). A LinkedIn post on the rail has the report presenting "a comprehensive assessme..." (LinkedIn card, snippet only), and The People Space gives it "nearly 30,000 real-world tasks, more than 50,000 human assessments and expert validation" and glosses its highest gradient as jobs whose tasks "could feasibly be handled by GenAI without human input" (The People Space). Wikipedia has no article on the index (Wikipedia); four articles cite the paper, for a United Nations report predicting that women "will likely bear the brunt of future layoffs in technological fields", for Australian firms that "have cut roles as a result of AI", for "the ratio of women in many software fields" declining, and for US enrolment in computing and STEM degrees dropping "especially for women" (Wikipedia: Gender equality in the technology industry; Technological unemployment; Software engineering; Programmer).

T against L_F(B) (AIO's ontology at work on its own cards): At 'ILO Generative AI and Jobs index' (T1, eight cards: B1–B8, of which B3 and B8 unreachable and B6 and B7 videos admitted by title; 293 words) the composition gives the index by its title and working-paper number and its headline findings. Composed at the source's grain or coarser: the index as the ILO's updated assessment (F1, F2), the four gradients by mean score and variability (F19–F21), the method's sources, worker surveys, Delphi rounds and AI assessments (F13, F15), 24% and 3.3% (F23), 34% against 11% (F24), transformation over redundancy (F38–F40), clerical occupations and translators most exposed (F31–F33), physical tasks least (F34, coarse). Available on the cards, not composed: every maker but the ILO (NASK's page, F6; the authors, F7), the 2023 paper's title (F10), the date and DOI (F60), the brief and NASK's name for the index (F3, F4), the scores against 2023 (F36), the purpose NASK states (F42), the data and its access (F59, F62), the female and male shares of the highest gradient on card 1 itself (F23's '4.7%' and '2.4%', and F24's by income), the transition through social dialogue on card 1 (F57), the commentary's limit (F68) and its question of who benefits (F69). Limit dropped: on the cards, The People Space's 'exposure does not equal automation' and 'upper-limit scenarios' (F68), while the findings it governs are composed (F23, F24, F38); the paper's own statements of the limit, 'not the actual impact on employment' and the upper threshold (F43), the static snapshot (F45) and its qualification of transformation, that 'fewer workers' may be needed (F41), stand in the own page B9, which T did not surface, and are recorded as not surfaced, not as dropped. Force raised: none observed; transformation is composed as the research's emphasis, at the interpretation's force (F38, F39). Composed without attribution, coarse grain: the reasons for the income and gender gaps, 'differences in economic structures and digital adoption' and women's representation in clerical roles, at the grain of The People Space (F28, F29) and the paper (F27), given as the index's findings without a source named; recorded at the coarse grain of common knowledge. Substituted reading: the gender finding, the paper's 4.7% against 2.4% (F23) replaced by 'roughly 29% vs. 16% across sampled countries', a figure in no source of the field, whose numbers stand in the paper only as the 2023 index's 29.2% augmentation and 'big unknown' together and 16.2% 'big unknown' (F30) (the Capture Registry's finding, CAP-1-03, CAP-1-04, names the 29.2% and not the 16.2%); the two counts, 29,753 tasks and 52,558 data points (F11), joined as 'nearly 30,000 to over 52,000 individual task data points' and placed in ISCO-08 (CAP-1-05; The People Space's 'nearly 30,000 real-world tasks, more than 50,000 human assessments', F66, keeps the two as two measures in one sentence). Content replaced by use: the closer, three prompts, one to 'Policy recommendations from the ILO for workforce upskilling', a recommendation the field does not hold (T1-11). Outside the cards: the gender figure, the least-exposed occupations, 'secretaries, receptionists', 'mass redundancy' (alignment.T1_outside_cards).

v0.9 §0.5: the first diff is ontological and narrow. T and the field agree on the entity, the ILO's index of occupational exposure, at its headline grain. What T's composition does with its own cards is to keep the findings and drop the statement that bounds them there, The People Space's 'exposure does not equal automation' (F68); the paper's fuller statement of the bound, its snapshot and its doubt about demand (F43, F45, F41) stand on the own page T did not surface. Its ontology at work is that of a finding without its bound: an exposure estimate composed as a fact about jobs. Where the cards are thin it supplies figures and lists no field source gives: a gender ratio (T1-6) and least-exposed occupations (T1-8). The makers reduce to the ILO: NASK, the preparer on its own page (card 4, F6), does not appear, nor any author (card 1, F7).

L_F(B) against L_A(B ∪ A) (the intervention): Admission adds, in one claim space with the field's: the record's organizing question and its placing of the index at the intersection of gains and losses, as an exposure estimate (LV-05–LV-08); the record's answer to the question for this index, the labor object and who supplies, directs and gains, each 'not recorded' as observed (LV-10–LV-13); the archive's reading that the index estimates where task labor could pass to a GenAI system on technical feasibility, meets task-origin only as a prior, and observes no performed work, so cannot speak to a system's labor on a commissioned task or to user labor drag (LV-34, LV-35); what the evidence settles and what it does not (LV-28–LV-33), with the worked question that sets an exposure estimate against observed trackers and the gap that no record establishes exposure as the cause of employment change (LV-48–LV-50); the record's negative tags (LV-41–LV-44); on the class of the employment effects of AI, the archive's statements that AI integrates the means of semantic production and automates its labor under concentrated ownership (#499), that labor platforms displace is relocated and enclosed, not vanished (#533), and that an account of AI job loss narrated as mechanism is a scenario and not a prediction, with a redistributive resolution (#521, #520); and the Capture Registry's findings at the name (CAP-1-01–CAP-1-11). The field's claims are carried whole and composed as received; none is translated (equal-terms case). The archive's additions extend the field's object on its own terms: the paper's limits (F43, F45, F46) and its qualification (F41) are what the record's negative tags and does-not-answer fields (LV-41–LV-44, LV-30–LV-33) restate at the record's grain, and the record adds the question the paper does not ask of itself, whose labor and to whose gain, answered 'not recorded'.

alignment: {'T1_cards_claims': ['F1', 'F2', 'F3', 'F4', 'F6', 'F7', 'F10', 'F11', 'F15', 'F17', 'F21', 'F23', 'F24', 'F28', 'F29', 'F31', 'F36', 'F37', 'F38', 'F39', 'F42', 'F57', 'F59', 'F60', 'F62', 'F65', 'F66', 'F67', 'F68', 'F69', 'F70', 'F71'], 'T1_composed': ['F1', 'F2', 'F13', 'F15', 'F19', 'F20', 'F21', 'F23', 'F24', 'F27', 'F28', 'F29', 'F31', 'F32', 'F33', 'F34', 'F38', 'F39', 'F40', 'F66'], 'T1_outside_cards': ["'roughly 29% vs. 16% across sampled countries' and 'nearly twice as likely' (no field source; the paper's 29.2% and 16.2% are 2023 category shares, F30, and its female/male figures are 4.7%/2.4% globally and 41%/28% exposed in high-income countries, F23, F25)", "'secretaries, receptionists' among the most exposed and 'routine information-processing roles' (no field source in these words; the paper names administrative secretaries in its Conclusion, outside the claims ledgered)", "'Manual trades, construction, agriculture, and heavy manufacturing' as least exposed (no field source names them; F34 states only that exposure declines with tasks of physical nature)", "'mass redundancy' (no field source)", "'Policy recommendations from the ILO for workforce upskilling?' (no field source states an upskilling recommendation; the paper asks for social dialogue, F55)"], 'rule': "a field claim counts as on T's cards when its source is a card or its quote occurs in a card's saved text or snippet (card 1, the ILO's publication page, carries the paper's abstract verbatim); it counts as composed when T states it at its source's grain or coarser; the typology of what is not is in T_vs_LB (§6.2)"}