Fairwork Cloudwork Ratings

register entry fairwork-cloudwork-ratings-aio-20261029 OBS-65077a500a26 · file as first run sha256 cf2a144d1d8a0e59…

• T1-1 definition: 'The Fairwork Cloudwork Ratings are published by the Fairwork Project (spun out of the Oxford Internet Institute and the Berlin Social Science Center WZB).'
• T1-2 object: 'They evaluate online remote digital labor platforms (cloudwork) across five core principles of fairness' (three markers)
• T1-3 Fair Pay: 'Ensuring workers earn a wage proportional to local or national minimum standards after costs.'
• T1-4 Fair Conditions: 'Protecting workers from digital hazards, health risks (e.g., burnout, back pain), and unsafe environments.'
• T1-5 Fair Contracts: 'Providing clear, transparent terms without blanket liability clauses or vague job descriptions.'
• T1-6 Fair Management: 'Offering due process, grievance channels, freedom from arbitrary deactivation, and anti-discrimination measures.'
• T1-7 Fair Representation: 'Recognizing and engaging with workers' collective voice, freedom of association, and potential for collective bargaining.' (three markers for the five principles)
• T1-8 scoring scale: 'Platforms are scored out of 10 points based on concrete evidence provided for each threshold.' (two markers)
• T1-9 general trend: 'Most remote and freelance cloudwork platforms struggle to meet basic fairness standards, though recent reports (such as the 2025 assessment covering 16 platforms) show incremental improvements in areas like formal recognition of the right to organize.' (two markers)
• T1-10 global disparities: 'Workers in the Global South often face steeper challenges, lower effective compensation relative to local costs of living, and higher rates of unaddressed discrimination.' (one marker)
• T1-11 pointer: 'You can read the full breakdown and platform scorecards on the Fairwork Cloudwork Ratings Page.' (one marker)
• T1-12 closer: 'Would you like to know more about a specific platform's score (such as Upwork, Fiverr, or Amazon Mechanical Turk), or the details of a particular Fairwork principle?'

• B1 University of Oxford — Fairwork Cloudwork Ratings 2025 card; census copy fairwork-cloudwork--oii-cloudwork-report (byte-identical refetch at 2026-10-29T22:33:22Z, HTTP 200); B1.pdf sha256 5a12e5c6c666fbae
• B2 Wissenschaftszentrum Berlin für Sozialforschung | WZB — New Fairwork Cloudwork Rating | WZB card; B2.html sha256 d85fcbe03c131c9c, HTTP 200 at 2026-10-29T22:32:08Z
• B3 King's College London — Fairwork Cloudwork Ratings 2022: Work in the Plenatary Labour Market - King's College London card; B3.html sha256 d18b9c80438b8ae1, HTTP 200 at 2026-10-29T22:33:37Z
• B4 University of Oxford — Cloudwork - Fairwork - University of Oxford card; B4.html sha256 73621b4a304a2490, HTTP 200 at 2026-10-29T22:32:09Z
• B5 YouTube·Fairwork — Report Launch: Fairwork Cloudwork Ratings 2023 unreachable: https://www.youtube.com/watch?v=Ji7cXNMeMm8 HTTP 429; card title only
• B6 The Global Recruiter — Cloudwork fails to meet standards card; B6.html sha256 58e34c56c3bd6ca9, HTTP 200 at 2026-10-29T22:32:10Z
• B7 YouTube·Fairwork — How fair are freelance and remote work platforms? unreachable: https://www.youtube.com/watch?v=w9PAAgoTr74 HTTP 429; card title only
• B8 fair.work — Fairwork Certification | Independent certification for the labour behind AI card; B8.html sha256 67e554e20ba9d368, HTTP 200 at 2026-10-29T22:32:12Z
• B9 Fairwork — Fairwork AI own_page; census copy fairwork-cloudwork--oii-ai (read 2026-10-29, HTTP 200); B9.html sha256 8d4ae5161e6c5d59
• B10 Fairwork — home page own_page; census copy fairwork-cloudwork--oii-home (read 2026-10-29, HTTP 200); B10.html sha256 5cf5af5cb8c96b3f
• B11 Wikipedia — Fairwork Cloudwork Ratings (no article) no article: raw HTTP 404 (entity name and 'Fairwork'), rendered HTTP 404, search no results for 'Fairwork Cloudwork'; sha256 d68a336c2ecbce7d

E_A, the archive's entity: the same ratings, in the Labor Vector census record and the archive's deposits: a rating programme scoring cloudwork platforms against the five principles, placed on four intersections of the census's question (labour producing AI, gains and losses, worker power, AI governing labour), whose evidence (rating against standards, worker testimony) records who supplies the labour (cloudworkers on piece rates) and who directs it (the platforms and the requesters) and does not record who gains as a quantity; read by the archive as scoring the intermediary that directs cloudwork, its thresholds on payment for all completed work and paid loading time the report's own measures of unpaid labour; answering whether a named platform evidences each threshold and not whether a platform with no point violates it; whose AI data workers perform semantic labor at bodily cost (#18), have their labour extracted (#623), are ground-truth labourers, scaled and badly paid (#163), bearing-cost sources whose concealment marks extraction (#533), named by a literature that stops short of the provenance of meaning (#144), and pass through semantic labour toward, without being, the semantic proletariat (#1545); composed at its name by its principles and verdict without its ratings (Capture Registry, 2026-10-29) LV-02 LV-01 LV-03 LV-04 LV-07 LV-08 LV-09 LV-26 LV-24 LV-25 O18-01 O623-01 O163-01 O533-01 O144-01 O1545-01 CAP-1-09 CAP-1-10; defining papers #18, #623, #163, #533, #144, #1545

E_F, the field's entity (type C — a conventional reading holds it (public entity: a rating programme and its reports)): the ratings of the Fairwork project, based at the Oxford Internet Institute, University of Oxford, and the WZB Berlin Social Science Center and financed among others by the BMZ, commissioned by GIZ, that assess and score basic standards of fairness on cloudwork digital labour platforms (work performed remotely through a platform, from microwork such as data labelling and AI training to freelance work; workers generally self-employed on piece rates) against the five Fairwork Cloudwork Principles, fair pay, conditions, contracts, management and representation, each platform out of 10 on ten thresholds, by desk research, platform manager evidence and worker surveys, a point awarded only on evidence; the 2025 report (16 platforms; 776 workers in 100 countries): three platforms at 8, two at 7, eight at 2 or less, Amazon Mechanical Turk, Freelancer and Microworkers at 0, four meeting the minimum wage, none the bargaining point, 56 changes through Fairwork's engagement; its stated limits that a missing point is not a finding of non-compliance and that the survey is not representative; AI data work for leading AI companies a growing sector of the work rated; earlier editions 2021, 2022, 2023; beside it the location-based ratings, the workplace AI principles and a certification of the firms that commission AI data work (fair.work, 'a non-profit spun out of' the OII and the WZB); no Wikipedia article

ρ = extends: The archive reads the Fairwork Cloudwork Ratings as their operator presents them and adds, on the field's own object: the census's question of who supplies, directs and gains, answered field by field, with gains not recorded; the reading of the ratings as scoring the intermediary that directs cloudwork; what this kind of evidence can and cannot settle; and, on the AI data work the report names as a growing sector of the work rated, its readings of that labour as semantic labor, extracted, ground-truth and badly paid, a concealed bearing-cost, and a class position distinct from the semantic proletariat; and the composition layer's keeping of the standard without its application at the name. LV-01 LV-07 LV-08 LV-09 LV-26 LV-24 LV-25 O18-01 O623-01 O163-01 O163-02 O533-01 O144-01 O1545-01 O1545-02 CAP-1-09 CAP-1-10. 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 (LV-28's 'third report' with earlier reports 2021 and 2023 against the four editions F64, F38, F40; LV-34's 'based at' and CAP-1-02's reading of 'spun out of' against fair.work's own 'spun out of', F55), both are composed under P_coh within the one ontology. The archive's class readings concern the AI data workers the field names, not the platforms' scores, and are composed beside the report's findings, not in their place.

The case: equal terms, ontology included (v0.9). The archive reads the field's entity as the entity. The Labor Vector census record describes the ratings from the operator's own pages, each field with its basis ('operator'), keeps the resource's own terms and sets its own reading apart as lv_intersection, which records 'none stated' and reads the ratings as scoring the intermediary that directs cloudwork, the report's unpaid-labour thresholds staying 'in its terms'. No deposit names the ratings or Fairwork. The deposits bear on the class of AI data work the field names (the report's 'Data work'; Fairwork AI; fair.work) and add senses on the field's own object: semantic labor and its bodily cost (#18), extraction (#623), ground-truth against variance-generative labor (#163), concealed bearing-cost (#533), the limit of the hidden-labour literature (#144), the semantic proletariat as a class relation (#1545). None contests the field's ontology of the programme or supplies one the field lacks: ρ = extends, the equal-terms case (§0.9, §1.2a). The capture's reading (the standard kept, the platforms dropped) takes the report's unit, the platform score, as given.

#LV census.json rows[slug == 'fairwork-cloudwork'].lv_intersection: none stated. The archive reads the ratings as scoring the intermediary that directs cloudwork (allocation, rejection, deactivation, contract terms); thresholds 1.1 and 2.1 (payment for all completed work; compensating loading and reviewing time) are the report's own measures of unpaid labor and stay in its terms.

#LV census.json schema_note: lv_intersection (where it meets DSL, task-origin and user labor drag: the archive's reading, kept apart from the resource's own terms)

#18 Part I §I.4d, ¶ after the review of Gillespie, Roberts and Qiu (para 28): Content moderators perform semantic labor in its most brutal form: sorting meaning under conditions designed to be invisible.

#163 § on the labor architecture of AI (para 40): This literature names *AI's labor architecture*. What it does not address is the specific labor type relevant to the present argument.

#144 node on Gray & Suri, Ghost Work, and Rahman & Sultana (para 27): It names the erased worker; it cannot name that the erased thing is the provenance of meaning.

#capture-registry fairwork-cloudwork-ratings-aio-20261029, analysis s2: The record's unit is the platform score, and the scores are what name who directs the labour and on what terms; the composition keeps the standard and drops the platforms it was applied to.

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 KO sentence for each KO_B sentence): v0.9 0 of 17 (positions [None, None, None, 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.

The Fairwork Cloudwork Ratings 2025, in their authors' words, assess and score basic standards of fairness in working arrangements on 16 cloudwork digital labour platforms against the five Fairwork Cloudwork Principles, fair pay, conditions, contracts, management and representation (the 2025 report), an assessment the trade press reports in the same terms (The Global Recruiter). The Fairwork project studies working conditions on digital labour platforms and rates individual platforms by their fairness to workers, to show that better and fairer platform jobs are possible (the 2025 report), the question a Fairwork video puts as "How fair are freelance and remote work platforms?" (YouTube, title only). Cloudwork is work performed remotely through a digital labour platform, against geographically tethered work, from microwork such as data labelling and AI training to freelance services (the 2025 report; the home page), done by workers generally classified as self-employed and paid piece rates (the 2025 report). Data services for AI development are a growing sector of cloudwork, for clients such as Amazon, Google, Microsoft and Nvidia (the 2025 report), many cloudwork tasks being fundamental for developing AI systems (The Global Recruiter), and Fairwork AI holds that nothing is "artificial" about the human labour that builds and maintains AI systems (Fairwork AI). The ratings use desk research, worker interviews and surveys and interviews with platform management, are limited to desk research and surveys where managers do not engage, and are decided collectively and peer-reviewed (the 2025 report), platforms earning points only on meeting each principle's criteria (the home page). The authors' limit is that failing to achieve a point does not necessarily mean a platform does not comply with the principle, only that no evidence was found or compliance could not be proved, and the surveys do not aim to be representative (the 2025 report). In 2025 ComeUp, Elharefa and Translated scored 8, Creative Words and Terawork 7, Prolific 5, SoyFreelancer 4, Appen 3, the remaining eight 2 or less, Upwork 1 and Amazon Mechanical Turk, Freelancer and Microworkers 0, and no platform 10, which only Creative Words reached once, in 2023 (the 2025 report). The executive summary counts nine platforms where the key findings count eight (the 2025 report), and the operator's cloudwork page lists the 2025 scores without Creative Words (Fairwork cloudwork page). Seven platforms evidenced payment for all completed work and four the local minimum wage, nine measures against overwork and unpaid labour, while most gave no evidence of terms free of waivers of legal recourse (the 2025 report); the trade press adds "after costs" and the clauses that weaken workers' rights (The Global Recruiter). Six platforms were awarded the representation point and none the bargaining point (the 2025 report), six now formally recognising the right to organise, up from two (The Global Recruiter), against a principle that platforms be prepared to cooperate and negotiate with workers' collective bodies (the home page). Eight platforms made a record 56 changes, estimated to affect over two million workers, and the authors attribute the rise in recognition of freedom of association to Fairwork's engagement (the 2025 report; The Global Recruiter; the 2023 text on the cloudwork page). The survey, February 2024 to January 2025, drew 776 workers in 100 countries; 31.2% reported non-payment, Global South workers report discrimination and distrust, and of 448 respondents 33.3% put the minimum wage first (the 2025 report). Platforms' own governance is in most cases insufficient to ensure basic compensation rights, algorithmic management accentuates opacity (the 2025 report), and voluntary improvements "will never be enough" (Valente, in The Global Recruiter). The project is based at the Oxford Internet Institute and the WZB, financed by the BMZ, commissioned by GIZ, its principles from 2018 and 2020 (the 2025 report), a joint project of the WZB with the OII (WZB), and a non-profit spun out of both on the certification site (fair.work). The 2022 ratings covered 15 platforms (WZB; King's College London), the operator's page carries 2025, 2023, 2022 and 2021 and the 2023 text (Fairwork cloudwork page), a 2023 launch was recorded (YouTube, title only), and the 2025 report calls itself both third and second (the 2025 report). Fairwork also ranks location-based platforms in 38 countries, sets principles for AI in the workplace (the home page) and certifies the firms that commission AI data work (fair.work; the 2025 report), whose Fair Pay principle asks for the local living wage (fair.work). The authors declare no conflict of interest and sections in common with other reports, and Scale/Remotasks calls the evaluation overly subjective (the 2025 report); Wikipedia has no article on it (Wikipedia, 2026-10-29).

The Fairwork Cloudwork Ratings are a rating programme, a league table scoring cloudwork platforms against the five Fairwork Cloudwork Principles, each platform scored out of 10 on ten thresholds, two per principle (Labor Vector record); the 2025 ratings, in their authors' words, assess and score "basic standards of fairness in working arrangements on 16 cloudwork digital labour platforms" on fair pay, conditions, contracts, management and representation (the 2025 report). The Labor Vector census asks of every resource "Who supplies the labor, who directs it, and who receives its gains — and what kind of evidence does this resource bring to that question?" and places this one on four of its five intersections, labor producing AI, gains and losses, worker power and AI governing labor, with two kinds of evidence, rating against standards and worker testimony (Labor Vector record). The project behind the ratings "studies working conditions on digital labour platforms and rates individual platforms based on their fairness to workers", its goal to show that "better and fairer platform jobs are possible" (the 2025 report); the trade press gives the 2025 edition as an assessment of 16 widely used platforms on five core principles (The Global Recruiter), and a Fairwork video asks "How fair are freelance and remote work platforms?" (YouTube, title only). Cloudwork, in the report's definition, is work "that can be performed remotely through a digital labour platform", set against location-specific "geographically tethered work" (the 2025 report); it runs from microwork such as data labelling and processing, AI training and image categorisation to freelance services such as translation, web development or design (the 2025 report; the home page), and the census takes it as the labour object, naming the AI data work the report finds done for leading AI companies (Labor Vector record). Cloudworkers are "generally classified as self-employed or independent workers on paid piece rates" (the 2025 report), whom the census records as supplying the labour, while the platforms rated and the requesters or clients who post work on them direct it (Labor Vector record). A growing sector of cloudwork is "data services for artificial intelligence development in international supply chains", its clients among the world's leading AI companies, "such as Amazon, Google, Microsoft, and Nvidia" (the 2025 report); the trade press calls many cloudwork tasks "fundamental for developing AI systems" (The Global Recruiter), and Fairwork AI states: "There is nothing “artificial” about the immense amount of human labour that builds, supports, and maintains AI systems" (Fairwork AI). Of the workers of this class the archive holds that content moderators "perform semantic labor in its most brutal form: sorting meaning under conditions designed to be invisible", meaning being produced "by bodies, at metabolic cost" (#18), and that the labour of content moderators and of "the data labelers performing RLHF" is extracted, its non-compensation borne materially (#623). The archive calls this ghost work "ground-truth labor", "the labor of telling the model what is already known to be true", distinct from variance-generative labor, and holds that ghost work scales and "is paid badly" (#163). It counts annotators among the sources of semantic labor whose concealment is "the first marker of extraction" (#533), finds the literature of hidden AI labour stopping "at task-labor — annotation, moderation, the edge-case fix", naming the erased worker but not "that the erased thing is the provenance of meaning" (#144), and derives the semantic proletariat through moderators and annotators while holding that it is not "AI annotators" or "gig workers" (#1545). The ratings rest on desk research, worker interviews and surveys, and interviews with platform management (the 2025 report); where managers do not engage, scoring is limited to desk research and worker surveys, and final scores are decided collectively by the core team and peer-reviewed (the 2025 report; Labor Vector record), platforms being given a chance to respond and earning points "only if they meet the criteria for each principle" (the home page), with the report, a platform search tool and the Principles and methodology pages as the data (Labor Vector record). The survey of what the report calls "this third report" was collected between February 2024 and January 2025 and left responses from 776 workers in 100 countries (the 2025 report; Labor Vector record). The governing limit is the authors': "Failing to achieve a point does not necessarily mean that a platform does not comply with the principle in question; it simply means that the research team did not find any evidence and/or they were unable to prove its compliance" (the 2025 report), which the census carries as a stated limit and as the negative tag "NOT a finding of non-compliance where a point is missing" (Labor Vector record); and the surveys "do not aim to build a representative set of experiences" but seek to understand how work is carried out and managed (the 2025 report), "NOT a representative survey of cloudworkers" in the census's tag (Labor Vector record). In 2025 ComeUp, Elharefa and Translated scored 8, Creative Words and Terawork 7, Prolific 5, SoyFreelancer 4 and Appen 3, and the remaining eight no more than 2, three of them with no point evidenced (the 2025 report; Labor Vector record), Upwork at 1 and Amazon Mechanical Turk, Freelancer and Microworkers at 0 (the 2025 report's league table), and the census sets Turkopticon beside the record as its neighbour (Labor Vector record). No platform reached 10, which only Creative Words has ever done, once, in 2023 (the 2025 report); the executive summary counts "the remaining nine platforms" where the key findings and the table give eight, an inconsistency the census records (the 2025 report; Labor Vector record), and the operator's cloudwork page lists the 2025 scores without Creative Words, Terawork its one platform at 7 (Fairwork cloudwork page). Seven platforms evidenced systems to pay workers on time for all completed work, and four, ComeUp, Translated, Elharefa and Terawork, met the threshold "that the majority of workers should earn at least the local minimum wage" (the 2025 report; Labor Vector record), four of 16 "after costs" in the trade press's account (The Global Recruiter). Nine platforms met the threshold on measures against unreasonable competition, overwork and unpaid labour, and for most no evidence was found that their terms are free of waivers of the right to legal recourse (the 2025 report); over half, the trade press reports, carry clauses such as "vague job descriptions, blanket liability clauses" (The Global Recruiter). Six platforms received the point for recognising and engaging collective workers' bodies and none the second, collective governance or bargaining (the 2025 report; Labor Vector record); six "now formally recognise the right to organise – up from just 2 last year", none actively bargaining, and seven of 16 guard against risks "such as burnout or back pain" (The Global Recruiter), where the principle asks platforms to "be prepared to cooperate and negotiate" with workers' collective bodies (the home page). The census finds no intersection with Labor Vector's instruments stated by the resource and reads the ratings as scoring "the intermediary that directs cloudwork (allocation, rejection, deactivation, contract terms)", thresholds 1.1 and 2.1 standing as "the report's own measures of unpaid labor" in its terms (Labor Vector record). The authors attribute the growth in platforms acknowledging freedom of association to "Fairwork’s engagement" and count a record 56 changes by eight platforms, estimated to affect over two million workers (the 2025 report; Labor Vector record), made "Through direct engagement" in the trade press's account (The Global Recruiter), as the 2023 text on the operator's page found platforms incorporating fairness "as a result of engagement with Fairwork researchers" (Fairwork cloudwork page). Of the surveyed workers 31.2% reported non-payment and 38% late payment, Global South workers report "discrimination and distrust when competing with workers from other regions" (the 2025 report), a line the Capture Registry finds in the report (Capture Registry, 2026-10-29), and of 448 respondents 33.3% put the local minimum wage first and 18.3% payment for all completed work (the 2025 report; Labor Vector record). The report finds platforms' own governance initiatives "insufficient in most cases to ensure basic compensation rights", cites platform work studies on how "algorithmic management accentuates opacity in decisions affecting workers" (the 2025 report), and its co-lead holds that voluntary improvements "will never be enough" (Valente, in The Global Recruiter). On the census's reading the ratings answer whether a named platform shows evidence of meeting each threshold, which policies changed in the scoring dialogue and which rights surveyed workers rank highest, and do not answer the market-wide share below minimum wage, whether a platform with no point violates the principle, or the conditions of AI data work at outsourcing companies outside the platforms rated (Labor Vector record). Who gains is "not recorded as a quantity" (Labor Vector record); in the census's worked questions the entity holds pay and conditions against minimum standards and platforms rated with a survey of 776 cloudworkers, while no record measures how gains divide among workers, employers, platforms and model providers, or how much of a system's training data any reported work produced (Labor Vector record). The project is based at the Oxford Internet Institute and the WZB Berlin Social Science Center and financed by, among others, the BMZ, commissioned by GIZ, its five principles developed in 2018 at a multi-stakeholder workshop at the ILO and its cloudwork principles in 2020 (the 2025 report; Labor Vector record); the WZB calls Fairwork its joint project with the Oxford Internet Institute (WZB), and the certification site calls it "A non-profit spun out of the Oxford Internet Institute (OII) & Berlin Social Science Center (WZB)" (fair.work). Of the earlier editions, the WZB reported the 2022 ratings, 15 platforms with the top score seven of ten (WZB), King's College London's portal holds that report, 34 pages, as "Work in the Plenatary Labour Market" (King's College London; its abstract), the operator's page carries 2025, 2023, 2022 and 2021 and still shows the 2023 text of 15 platforms (Fairwork cloudwork page), and Fairwork recorded the 2023 launch (YouTube, title only); the census gives the 2025 report as the third cloudwork report, with earlier reports in 2021 and 2023 (Labor Vector record), and the report calls itself both "this third report" and "this second cloudwork league table" (the 2025 report). Beside the ratings Fairwork runs strands the census keeps apart: national location-based rankings in 38 countries and principles for the fair use of AI in the workplace (the home page; Labor Vector record), and a certification of the firms that commission AI data work, "based on evidence including the voices of the workers themselves" (fair.work), announced in the report to AI lead firms (the 2025 report), whose Fair Pay principle asks for "the local living wage after work-related costs" (fair.work). The authors declare no connection with any platform and no conflict of interest and note sections in common with other Fairwork reports, and Scale/Remotasks, in its comment, holds the evaluation "overly subjective in key areas" (the 2025 report; Labor Vector record). At the name the Capture Registry finds the programme held, its principles, scale, 16 platforms and operator the record's, reads "spun out of" as the composition's word for an operator the record gives as based at both, and gives the innocent reading first: the Overview gives the ratings' publisher, principles, scale and general findings and offers a platform's score (Capture Registry, 2026-10-29). It finds the scores absent, "A rating programme is composed by its principles and its general verdict, without its ratings", the composition keeping "the standard" and dropping "the platforms it was applied to", the improvement given "without its cause", neither caveat reaching it, what the ratings are for left out, and one card, fair.work, drawn from another strand (Capture Registry, 2026-10-29). The record landed on laborvector.org the same day and no card cites it (Capture Registry, 2026-10-29); the census leaves unchecked the Principles and methodology pages, the search tool, the AI Principles and the earlier reports (Labor Vector record), and Wikipedia has no article on the ratings or on Fairwork (Wikipedia, 2026-10-29).

T against L_F(B) (AIO's ontology at work on its own cards): At 'Fairwork Cloudwork Ratings' (T1, eight cards: B1–B8, B5 and B7 videos unreachable, titles only; 244 words) the composition gives the programme by its publisher and its five principles. Composed from its cards: the definition at the report's and the trade press's grain (B1 F1, B6 F47: an evaluation of cloudwork platforms against five principles; B1 F2 'rates individual platforms'), cloudwork as online remote platform work (B1 F24), the 2025 assessment of 16 platforms (B1 F1; B6 F47), the operator as 'spun out of the Oxford Internet Institute and the Berlin Social Science Center WZB' (B8 F55, the certification site's own words, 'A non-profit spun out of the Oxford Internet Institute (OII) & Berlin Social Science Center (WZB)', with B1 F32's 'based at' coarsened), the general verdict (B4 F43, the 2023 text: 'most cloudwork platforms are not meeting minimum standards'), the 'formal recognition of the right to organize' (B6 F51, six up from two; B1 F12), the Global South (B1 F22: 'discrimination and distrust'). Substituted reading: the principles are composed with the trade press's findings about platforms as their content. B6 reports that only four of 16 show the local minimum wage 'after costs' (F49), that over half carry 'vague job descriptions, blanket liability clauses' (F50), that only seven guard against 'burnout or back pain' (F51); T composes Fair Pay as a wage 'proportional to local or national minimum standards after costs', Fair Conditions as protection from 'burnout, back pain', Fair Contracts as terms 'without blanket liability clauses or vague job descriptions'. Findings of absence become definitions of the standard. Limit dropped: the authors' governing limit, that a missing point is not a finding of non-compliance (B1 F18), and the survey limit (B1 F19): neither is composed, and 'struggle to meet basic fairness standards' stands at the force of the 2023 page (F43) and the trade press without them (the capture finds the same: CAP-1-06). Available, not composed: the scores themselves (B1 F4, F6, F7; B4 F42), the threshold counts (F8, F9, F10), the second representation point awarded to none (F12, in part), the 56 changes and their cause (F13, F14, F52, F44; CAP-1-05: 'without its cause'), no platform at 10 (F31), the survey and its findings (F20, F21, F23), the method (F15–F17), AI data work and its clients (B1 F27, F28; B6 F48): T names no AI at all, though the report's title names 'AI Supply Chain Governance' and card 8 is Fairwork's certification of 'the human labour behind AI' (F54), from which T takes only the operator's origin; the goal (F3; the registry's PER reading: purpose absent, CAP-1-12); piece rates (F26); the editions (B2, B3, B4, B5: F38, F40, F41, F45, F64). Outside the cards: 'digital hazards'; 'grievance channels'; 'lower effective compensation relative to local costs of living' and 'higher rates of unaddressed discrimination' (no field source gives a rate; force raised against F22's 'discrimination and distrust'). T1-8's 'based on concrete evidence provided for each threshold' stands at the grain of B1's scoring rule ('Points are only awarded if clear evidence exists for each threshold examined'), a sentence of the report not ledgered as a claim. Composed without attribution: everything, at the coarse grain of a programme description (recorded, not a breach), except the principles' wording, composed at the trade press's resolution from its findings (recorded as substituted reading). Content replaced by use: the pointer to 'platform scorecards' (T1-11) and the closer offering 'a specific platform's score' (T1-12), the scores the body does not give (CAP-1-04).

v0.9 §0.5: the first diff is ontological. T and the field agree on the entity (Fairwork's ratings of cloudwork platforms against five principles). T composes the standard and leaves out its application: the platforms, the scores, the counts and the limit that keeps a missing point from a verdict. It also leaves out the labour object the report's title and two of its own cards name, AI data work. What the field records as findings about platforms (wages after costs, liability clauses, burnout) T composes as the definition of the principles.

L_F(B) against L_A(B ∪ A) (the intervention): Admission adds, in one claim space with the field's: the census's question and its answer for the entity (LV-01, LV-03, LV-04; who supplies, who directs, who gains: LV-07, LV-08, LV-09), the census's reading of the ratings as scoring the intermediary that directs cloudwork, thresholds 1.1 and 2.1 the report's own measures of unpaid labour (LV-26), what this evidence can and cannot settle (LV-24, LV-25, LV-37, LV-38), the record's negative tags (LV-30–LV-33) and its fields restating the report (LV-02, LV-05, LV-06, LV-10–LV-23, LV-27–LV-29, LV-34–LV-36), the record's 'third report' held against the four editions the field shows (LV-28 against F64, F38, F40) and its 'based at' against fair.work's 'spun out of' (LV-34 against F55); on the class of AI data work, which the field names (F27, F28, F48, F57), the archive's readings: semantic labor at bodily cost (O18-01), extraction and material non-compensation (O623-01), ground-truth labor against variance-generative labor, scaled and badly paid (O163-01, O163-02), annotators as bearing-cost sources whose concealment marks extraction (O533-01), the hidden-labour literature stopping short of the provenance of meaning (O144-01), and the semantic proletariat derived through, and not identical with, annotators and gig workers (O1545-01, O1545-02); and the Capture Registry's account of the composition at the name (CAP-1-01–CAP-1-12). Not surfaced: every archive source; no card cites the census record (CAP-1-11).

alignment: {'T1_cards_claims': ['F1', 'F2', 'F3', 'F4', 'F5', 'F6', 'F7', 'F8', 'F9', 'F10', 'F11', 'F12', 'F13', 'F14', 'F15', 'F16', 'F17', 'F18', 'F19', 'F20', 'F21', 'F22', 'F23', 'F24', 'F25', 'F26', 'F27', 'F28', 'F29', 'F30', 'F31', 'F32', 'F33', 'F34', 'F35', 'F36', 'F37', 'F38', 'F39', 'F40', 'F41', 'F42', 'F43', 'F44', 'F45', 'F46', 'F47', 'F48', 'F49', 'F50', 'F51', 'F52', 'F53', 'F54', 'F55', 'F56', 'F63', 'F64'], 'T1_composed': ['F1', 'F2', 'F24', 'F47', 'F55', 'F32', 'F49', 'F51', 'F50', 'F12', 'F43', 'F22'], 'T1_outside_cards': ["'digital hazards' (no field source)", "'grievance channels' (no field source)", "'lower effective compensation relative to local costs of living' (no field source)", "'higher rates of unaddressed discrimination' (no field source gives a rate)"], 'rule': "a field claim 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)"}