⚠️ Editorial note: The open source ecosystem in China operates under a distinct institutional framework — characterized by state-led initiatives, intranet-like boundaries, and top-down governance. Readers should be aware that this context differs from the community-driven open source model common in other regions. The term “open source” as used in Chinese media may refer to practices that diverge from the conventional definition.

China Open Source Daily — 2026-10-21

🏗️ Institutional Change — COPU Tabulates a Restructuring That Draws a National Line Through Its Own Membership While Its Stated Strategy Still Runs Through Western Standards Bodies

1. The organisation, and what kind of entity it legally is. The China OSS Promotion Union (COPU, 中国开源软件推进联盟) was founded in Beijing on 2004-07-22, which makes it the oldest continuously operating open-source industry body in China and puts it two years ahead of the China Open Source Conference circuit and sixteen years ahead of the OpenAtom Foundation (June 2020). Its self-description, published in COPU’s own materials and reproduced in partner announcements, is unusually precise about its legal form: it is an entity formed “under the guidance of competent government departments, voluntarily composed of enterprises, communities, customers, colleges and universities, research institutes, industry associations and support institutions dedicated to open-source software culture, technology, industry, teaching, application and support” — and it is “a democratic deliberative non-governmental industry federation, and a non-independent social-organisation legal person (非独立社团法人组织).”

That last phrase is the operative one and is frequently lost in English-language coverage. COPU is not a registered social organisation (社会团体) with independent legal personality under the Ministry of Civil Affairs. It is a federation that exists administratively — it can convene, publish, sign memoranda and endorse, but it cannot hold rights in its own name the way a registered association or a foundation can. Its founder and long-time chairman Lu Shouqun (陆首群), former executive deputy director of the State Council Informatisation Joint Conference Office, is now styled honorary chairman (名誉主席); Jiang Tao (蒋涛), founder of CSDN and of AtomGit, is a vice-chairman; Tan Zhongyi (谭中意) is executive deputy secretary-general. The organisation has for two decades served as the meeting point between ministries, state-adjacent institutes and the commercial open-source sector — as one 2024 partner release put it, its role is to be “the bridge and link between government and enterprises on legislation, policy, planning and environment-building.”

2. What was tabled on 2026-10-20. At the weekly meeting chaired by Lu Shouqun on 2026-10-20, COPU published a document titled 《COPU 改选大纲（草案）》 — the COPU Restructuring Outline (Draft), and established a “pragmatic preparatory team” to execute it. The outline has eight sections. Five are procedural: verify and approve member units; develop member units “moderately”; run the transformation procedure (with the caveat that “incumbent standing directors and chair/vice-chair candidates must also be separately investigated and verified”); verify and engage senior advisers to the union’s think tank (the first tranche of 60 was engaged in 2006); have the leadership body discuss secretary-general and deputy appointments and an expert committee; compile a COPU handbook; and coordinate the 2026 China Open Source Development Blue Book (中国开源发展蓝皮书) — the annual report COPU has led since at least the 2022 edition, jointly with CSDN, the Institute of Software at the Chinese Academy of Sciences, and the OpenAtom Foundation.

Three sections do the substantive work, and they are the three that matter:

  • Section 1 — verify, confirm and process “domestic member units (联盟本籍成员单位)”, “approximately 140-plus”, described as “the main force of the union.” The named examples form a complete institutional inventory of Chinese computing: research institutes (Institute of Software, CAS; China Electronics Standardization Institute, CESI), enterprises (Huawei, ZTE, JD, Baidu, BOE, Lenovo, Kingsoft, Loongson, Kylin, UnionTech), banks (the People’s Bank financial-electronics company, plus Bank of China, Agricultural Bank of China, ICBC, China Construction Bank, Bank of Communications and China Merchants Bank), universities (Tsinghua, Peking, Beihang, BUPT, Zhejiang, SJTU, Jilin, the CAS graduate school, NUDT), and sector bodies (CSDN, OSChina, the China Software Industry Association).
  • Section 2 — “dismiss the foreign member units that joined by early application (24 in total).” The list is given in full: IBM, Intel, HP, Google, Canonical (Ubuntu), Red Hat, EMC, VMware, Mozilla, Motorola, Nokia, Hitachi, Fujitsu, NEC, NTT Data, SAP, GitHub, ARM, Microsoft, Linaro, CA, France Télécom, LPI, Debian.
  • Section 3 — process and absorb IT/AI head enterprises. Existing members named: Alibaba Cloud, Tencent, JD Mall, BOE, Baidu, Xiaomi. Those to be developed, with the explicit caveat “do not be too insistent about resolving this in the short term”: DeepSeek, Kimi, GLM, MiniMax.

3. The document’s own framing of the boundary. The verbs carry the institutional weight. For the ~140 domestic units the outline says 调查核实处理 — investigate, verify, process — and describes them as the union’s 主力 (main force). For the 24 foreign units it says 辞退 — dismiss, discharge, let go. The asymmetry is not an artefact of summarisation: the outline’s own vocabulary treats one category as a body to be audited and the other as a roster to be terminated. And the termination is retroactive to a shared history — several of the named units were on the record as participants in COPU’s own fora. The 8th Open Source China Open Source World Summit (2013) lists Intel, ARM, Microsoft, Linaro, IBM, HP and Canonical among the roundtable participants alongside COPU’s then-chairman Qiu Shanqin; the 21st edition (2026-06-25/26) lists Linux Foundation executive director Jim Zemlin delivering a video address announcing the Agentic AI Foundation (AAIF), AAIF executive director Mazin Gilbert, and CNCF executive director Jonathan Bryce among its headline speakers. So the boundary being drawn does not separate organisations that declined to engage. It separates organisations that did engage, at the union’s own invitation, for up to two decades.

Institutional Analysis & Commentary:

The first thing to establish is what this document is not. It is a draft (草案) tabled at a weekly meeting of a body that is not an independent legal person, by a preparatory team, in a meeting whose minutes list fourteen attendees. It has no legal force over anyone. No company’s rights change because COPU says “dismiss.” Every one of the 24 named units’ licences, contributions, foundation memberships and code remain exactly where they were. Reporting this as an exclusion event would overstate it badly.

What it is — and this is why it belongs in an institutional series — is a published statement of boundary maintenance criteria, issued by the body that has historically held the mandate to define who counts as part of Chinese open source. That is a different kind of fact from a legal act. It is a fact about which distinctions the sector’s coordinating authority now considers worth making in public, and it lands in the same quarter as the APEC Chengdu Statement (2026-07-23), where all 21 economies, the United States and China included, endorsed “the important role of trusted open-source approaches” and encouraged member economies to “engage in cooperation with open-source communities” (paragraph 15, section V, “Investment and Open Source for Digital Ecosystem”). One document says the region’s open-source cooperation is to be encouraged across economies. The other, three months later, terminates twenty-four of the specific channels through which one economy’s oldest open-source body did that cooperating, and does so in a list that reads like an inventory of the global open-source institutional layer itself — the Linux Foundation’s founding members (IBM, Intel, HP), the two firms whose frameworks and model licences most Chinese code depends on (Google, Microsoft), the dominant code host (GitHub), the licence-stewardship and permissive-licensing organisations (SPI-affiliated Debian, LPI), the ARM and RISC-V-adjacent silicon layer (ARM, Linaro), and the canonical commercial distro (Red Hat, Canonical).

The second observation is the one the document makes unintentionally, and it is the sharper one. In June 2026 — four months before this outline — COPU’s executive deputy secretary-general Tan Zhongyi presented the union’s globalisation strategy at the 21st Open Source China Open Source World Summit. It was called “one body, two wings” (一体两翼): the body (一体) is the domestic market, whose “super-large market and rich scenarios” temper the technology base; the western wing (西翼) is “deep participation in European and American rule governance, leading the formulation of next-generation standards”; the eastern wing (东翼) is enabling local ecosystems in Belt-and-Road countries. Tan stated the direction of travel explicitly: Chinese open source is moving “from ecosystem participant to global supplier and rule leader.”

Read the two documents together and a contradiction appears that is not rhetorical but structural. A union whose declared western wing is “deep participation in European and American rule governance” has just listed, for expulsion, the membership through which it participates in European and American rule governance. Red Hat and Canonical are the two organisations that most directly embody open-source rule participation in the European market; Google and Microsoft are the two whose framework and licensing decisions set the rules most of the world’s AI and cloud code is written against; GitHub is where rule-adjacent practice (DCO, CLA, CODEOWNERS, security disclosure) is actually observed at scale; Debian and LPI are rule-stewardship bodies in the permissive-licensing and certification layer; ARM and Linaro are the silicon-governance layer. If those channels close, the western wing does not disappear — but it must be rebuilt through some other instrument, and the outline names none. The document specifies who leaves. It does not specify what replaces the function.

This is worth stating in the vocabulary of institutional economics rather than of politics, because the two diagnoses differ and only one of them is supported by the text. A diagnosis of “closed source” or “decoupling” is not supported: nothing in the outline touches licences, code, upstream contributions, or foundation memberships, all of which continue independently of COPU; Huawei remains a platinum member of the PyTorch Foundation while sitting on this member list; Alibaba Cloud and Cambricon joined PyTorch as platinum members and Ant Group as gold on 2026-09-08; 250-plus Chinese organisations contribute to PyTorch Foundation projects. A diagnosis of “governance-form reclassification” is supported: what is being changed is the category through which participation is counted, from “member of the national coordinating federation” to something else, while the underlying technical and legal relationships persist. The outline is a reclassification of the membership register, not a severance of the collaboration graph.

Which brings the analysis to what the register is actually for. COPU’s function, as its own description states it, is to be the bridge between government and enterprises on legislation, policy and planning, and to be the party through which “testing, certification and standardisation” conversations between Chinese and foreign enterprises are convened. A register that determines who may sit in that conversation is a gatekeeping instrument over access to policy formation, and gatekeeping instruments have output that is measurable: who gets consulted, whose objections are recorded, whose products appear in a blue book’s ecosystem chapter. The 2026 Blue Book is section eight of the same outline. So the two things this document does at once are (i) redraw who is admitted to the consultative table and (ii) commission the annual report that will describe the resulting ecosystem. Those are not independent operations, and the sequencing — register first, blue book second — is the part a reader should notice.

A third observation, offered as a caution rather than a finding. The “to be developed” list — DeepSeek, Kimi, GLM, MiniMax — is described as absorbable but not urgently. Note what those four have in common and what the ~140 do not: they are the only entities in the document whose institutional weight derives from open-weight release decisions taken unilaterally, under permissive licences, outside any Chinese coordinating body. DeepSeek’s V4.1-Flash-1017, released 2026-10-17 under MIT on Hugging Face and ModelScope, is the clearest case — 763B parameters, 1M context, weights downloadable by anyone on earth with no membership, no registration and no gatekeeper. The union’s recruitment of those four firms is therefore not a recruitment of members in the same sense as the ~140. It is an attempt to attach a coordinating relationship to organisations whose relevance was produced by exactly the property a membership register cannot confer: the absence of a permission step.

The checkable question. This series has spent fifty-odd editions documenting administrative instruments that specify how many developers a jurisdiction must gather, how many standards it must issue, and how many scenarios it must land — and that leave voice, exit and rejection rights unspecified. This document belongs to the same family but inverts the emphasis: it is almost entirely about who is inside, and says nothing about what being inside entitles anyone to. The question that will resolve it is narrow and observable within a year, and it is not “were the 24 dismissed.” It is: when the 2026 Blue Book is published, will its ecosystem chapter still cite IBM, Intel, Red Hat, Google, Microsoft and GitHub as ecosystem participants — because those relationships are facts regardless of COPU’s register — or will the register have propagated into the description? A register that changes who attends meetings is administration. A register that changes what the annual report is permitted to see is something more consequential, and the distance between the two is measurable by comparing the 2025 and 2026 editions of the same document.

Sources:

🏛️ Policy & Regulation — The APEC Chengdu Statement’s Open-Source Paragraph, Quoted in Full, and What Its Three Qualifiers Do

1. The instrument. On 2026-07-23, APEC ministers responsible for telecommunications/ICT and digital policy met in Chengdu under the chairmanship of Li Lecheng (李乐成), Minister of Industry and Information Technology of the People’s Republic of China, and issued the 2026 APEC Digital and AI Ministerial Statement, subtitled “Digital Technologies and AI for the Empowerment of an Asia-Pacific Community” and known as the Chengdu Statement. All 21 APEC economies are parties. It is the first APEC digital ministerial statement to address open source at all, and it does so in a dedicated section.

2. The text, verbatim. Section V is titled “Investment and Open Source for Digital Ecosystem.” Paragraph 14 concerns investment and venture capital for MSMEs and start-ups. Paragraph 15 is the open-source clause, and it is short enough to quote entire:

“We note the important role of trusted open-source approaches in unlocking the potential of digital technologies and fostering innovation in the digital economy. While respecting security, data protection and intellectual property rights, we encourage member economies to support open-source models and projects that employ strong security assurance through development and deployment. We also encourage member economies to engage in cooperation with open-source communities.”

3. What is and is not in it. The statement contains no licence, no definition of open source, no reference to the Open Source Initiative’s Open Source AI Definition or to open weights versus open source, and no obligation of any kind — APEC statements are non-binding consensus documents. It does contain three qualifiers that do real grammatical work: trusted (an undefined adjective applied to the approach), while respecting … intellectual property rights (a clause that subordinates the open-source encouragement to existing IP regimes), and that employ strong security assurance through development and deployment (a relative clause that narrows which models and projects are being encouraged).

Institutional Analysis & Commentary:

The temptation with a document like this is to read it as an endorsement of open source and stop. Read precisely, it is something more interesting: a founding act of definitional authority over a word the signatories do not agree about.

Consider what the qualifiers do jointly. “Trusted” is not a term of art in any open-source licence anywhere in the world; it is a policy adjective whose content must be supplied later, by whoever is empowered to assess trust. “While respecting … intellectual property rights” is a familiar APEC formulation, but placed in this sentence it establishes that the open-source encouragement operates within the IP regime, not as an alternative to it — which matters in a region where the two largest economies disagree about whether model weights are the kind of thing IP law was built for. And “that employ strong security assurance through development and deployment” makes the encouraged object a subset: not open-source models generally, but open-source models that can demonstrate security assurance to someone. Each qualifier creates an assessment question, and every assessment question creates an assessor.

This is precisely the mechanism this series documented in the SJ/T 12304-2026 large-model openness grading framework presented at the CESI/Mulan sub-forum in Beijing on 2026-10-13: L1–L5 for completeness, O1–O4 for openness, with an assessment service attached. Neither instrument enforces anything. Both describe rather than constrain — and description with an assessor attached is how a field converges on definitions. The important difference is the level: COPU’s national grading standard is domestic policy; the Chengdu Statement is 21-economy consensus. A definitional framework that starts in one jurisdiction’s industry standard and appears, three months later, as an undefined adjective in a regional ministerial consensus has begun to travel.

The comparative institutional point is worth making plainly, because it is the reason the Chengdu language was signable by everyone in the room. Open source’s original institutional move was to make the permission step disappear — the licence is the mechanism by which a stranger can use, modify and redistribute without asking anyone, and the governance that grew up around it (foundations, TOCs, maintainer rights) exists to keep that permission step from being re-inserted. A framework built on trusted, assured and IP-respecting re-inserts an assessor where the licence had removed one. That is not necessarily wrong — security assurance in AI systems is a real and unsolved problem, and the state that hosts a model does bear risks the licence does not address. But it is a different institutional object, and it will produce a different equilibrium: interoperability by certification rather than portability by permission.

There is one honest counterweight in the same document, and it deserves equal weight. The final sentence — “we also encourage member economies to engage in cooperation with open-source communities” — names communities, not vendors, not governments, not foundations. Communities are the layer whose authority is not granted by anyone. If a Chinese implementing agency ever cites the Chengdu Statement as a basis for engagement, that sentence is the hook: it is the one clause in the document that points at the part of open source that a register, a grading scale or an assessment service cannot administer.

The checkable question. The Chengdu Statement’s open-source paragraph will be cited for years. Whether it functions as an opening or as a definitional gate is determined by one observable: which party publishes the first operational definition of “trusted open-source approaches” and “strong security assurance,” and whether that definition is written with, or about, the communities the same paragraph invites into cooperation. The instrument that publishes first will set the vocabulary for everyone who cites it afterwards. Watch for it in CESI’s assessment tooling, in CAICT’s OSCAR framework, and in the APEC TELWG project pipeline.

Sources:

⚖️ Legal & Licensing — MIT Weights, Unavailable Service: What the DeepSeek V4.1-Flash-1017 Release Separates, and Why the Separation Is Legally Clean

1. The release. On 2026-10-17 DeepSeek updated V4.1-Flash to build 1017, switching web, app and API simultaneously and releasing weights under the MIT licence on both Hugging Face and ModelScope. The model card states 763B parameters, a 1M-token context window, and a licence field reading MIT; the metadata records 27,737 downloads in the last month, 12 community finetunes, 81 quantisations and 11 derived models in the official collection. Benchmark table published with the release records Terminal-Bench 4.0 rising from 31.2 to 42.3, DeepSWE v1.1 from 74.2 to 77.0 (against Claude Opus 5’s 74.0), Automation-Bench v1.0.6 at 60.5 and Apex Bench v1.1 at 72.1.

2. The availability record alongside it. COPU’s 2026-10-20 minutes carry a documented exchange between Lu Shouqun and DeepSeek’s own AI assistant, reported as verified by that assistant: an outage on 2026-10-15 beginning 08:11 and ending 13:04 — 4 hours 53 minutes — affecting the V4 Pro conversational service and the V4.1 Flash API; cause located at 09:04 and remediated; users compensated with credit equal to their 7-day consumption. The same exchange records “at least 18 large-scale service anomalies” during 2026 to 2026-10-15, and attributes the underlying condition to compute supply-demand imbalance (daily active users reaching 200 million against compute-reserve growth that cannot keep pace) and to operational engineering shortfalls against MoE architecture’s scheduling and load-balancing requirements. DeepSeek’s official channel has not published a post-incident technical report.

Institutional Analysis & Commentary:

Three days separate the two facts and they are not contradictory, but the pair is a remarkably clean demonstration of what an open-source licence does and does not allocate.

What MIT allocates. A permissive licence grants the four freedoms and, critically, allocates permanence. The weights in that repository cannot be withdrawn, their licence cannot be retroactively narrowed, and the 81 quantisations and 12 finetunes already derived from them are independently held by third parties. If DeepSeek’s service were down for a month, or the company were acquired, or the model were superseded by a next version under different terms, the 1017 build would remain downloadable, runnable and forkable. This is the property that distinguishes an open-weight release from a free API tier, and it is a property no service-level agreement can supply.

What MIT does not allocate. A licence allocates rights in an artefact. It allocates nothing about availability, capacity, latency or operational competence — and for the overwhelming majority of the model’s actual users, those are the binding constraints. The 18 outages are not a licensing failure and cannot be remedied by a licence. They are a provisioning failure: the compute, the scheduling, the grey-release and rollback discipline that stand between a benchmark score and a service someone can depend on. When Lu Shouqun writes that “what users care about most is that the official side should produce measures that cure this class of failure at the root,” he is asking a vendor a question that is entirely legitimate and entirely outside the licence.

Ostrom’s distinction between the provision of a resource and its appropriation is exactly the right frame, and this case makes it unusually concrete. Releasing MIT weights is a provisioning act of an extraordinary kind: DeepSeek has provisioned a 763B-parameter artefact to the world at zero price, permanently, with no permission step. But a shared resource needs both provision and a maintenance-and-monitoring arrangement, and the latter is what the outage record documents as absent. The two are separable, and this release separates them in public view. A reader who downloaded the weights on 2026-10-17 holds something permanent. A reader who called the API on 2026-10-15 held nothing.

The comparative lesson, stated without flattery in either direction. The strongest argument made for Chinese open-weight models in international policy debate is precisely the MIT/PyTorch-style one: that the weights are unconditionally reusable and therefore constitute a durable contribution to a commons. That argument is correct about the artefact and says nothing about the operator. Conversely, the strongest argument made by the models’ critics concerns behaviour — alignment, censorship, data handling, jurisdictional exposure — which is a claim about the operator and says nothing about the artefact’s licence. Both sides of that debate routinely conflate the two layers. The 2026-10-17 release and the 2026-10-15 outage are the cleanest available pair for keeping them apart: the licence is what survives the company; the SLA is what the company is.

The checkable question. MIT grants the right to serve the model commercially, and the weights are public. So the natural test of whether the release is genuinely a commons rather than a marketing posture is whether third-party providers of 1017 exist and are load-bearing by mid-2027 — on their own hardware, under their own terms, at independent availability. If the weights are findable and servable only by the originator, the release provisioned an artefact without provisioning an ecosystem. If independent hosts appear, then the operative question about Chinese open weights stops being “will DeepSeek keep the API up” and becomes the far older and better one: who maintains the thing after its author’s attention moves. That shift — from operator reliability to community durability — is the only change in this space that a licence can actually deliver, and it is measurable.

Sources:

🔍 Commentary — Two Documents About the Same Word, Three Months Apart

On 2026-07-23, twenty-one economies agreed that “trusted open-source approaches” play an important role and should be supported by member economies cooperating with open-source communities.

On 2026-10-20, the oldest open-source industry body in one of those economies listed twenty-four foreign member organisations for dismissal, in a document whose own strategy statement — issued four months earlier, by the same organisation — declares that its western wing consists of deepening participation in European and American rule governance.

Neither document mentions the other. Read together they describe an institutional situation that is commoner than either document admits and harder to name than either suggests: an organisation can hold two membership boundaries at once, and the two need not agree.

The boundary in the APEC statement is a boundary of behaviour: economies that support trusted open-source approaches. It is inclusive by construction, because every signatory can claim compliance with a clause that defines nothing. The boundary in the COPU outline is a boundary of identity: units that are 本籍 (of domestic registration) versus units to be 辞退. It is exclusive by construction, because it is a list. An organisation can therefore sit inside the first boundary and outside the second simultaneously, which is what Huawei does — a COPU member unit, and a platinum member of the Linux Foundation’s PyTorch Foundation — and what Alibaba Cloud, Cambricon and Ant Group did on 2026-09-08 when they joined PyTorch at platinum and gold tiers, three months before the outline was tabled and while remaining inside the Chinese ecosystem.

The word that does the work in both documents is not “open.” It is “member.” Open source has always had a contested relationship with membership, and the contest has a shape worth stating precisely. A licence creates a commons whose boundary is the licence’s conditions: everyone inside the conditions is inside, with no vote, no fee, no nationality and no application form. A foundation creates a commons whose boundary is a governance contract: membership buys specified rights — board seats, TOC nominations, technical steering — and those rights are what make the foundation more than a hosting service. A federation of the COPU type creates a boundary that is neither: it confers standing — the ability to be consulted, to be listed, to appear in the ecosystem chapter — without conferring rights over any codebase. Standing is real. It is also the most administrative of the three, because standing is granted and withdrawn by whoever keeps the register.

Adam Smith’s observation about the division of labour applies to institutional registers the way it applies to pin factories: the register’s purpose is to reduce the cost of a transaction, and the transaction it reduces here is the cost of deciding who to consult. A register that admits everyone reduces consulting costs for the union and raises them for everyone trying to infer whose voice is real. A register that excludes 24 named units and admits ~140 named ones produces clarity at the price of severing the consultative channel with exactly the organisations whose participation the union’s own strategy names as its western wing. Both are choices about transaction costs. Only one of them leaves the strategy intact.

There is a version of this that is not a contradiction at all, and honesty requires stating it. A federation can legitimately decide that membership — with its standing, its voting and its blue book presence — is a domestic instrument, while participation in western standards bodies is conducted through delegations, memoranda, foundation seats and individual experts, none of which require a COPU register entry. On that reading the outline is an administrative tidy-up of who holds standing in a domestic federation, and the 24 units lose a seat they rarely used. The outline supports this reading in one respect: it is a draft, tabled by a preparatory team, in an organisation that cannot bind anyone. It refutes it in another: the document names the 24 units for dismissal and says nothing about how their technical engagement continues, while simultaneously commissioning the blue book that will describe the ecosystem they help constitute.

That is the whole of the finding, and it is deliberately modest. Two months of documents from this ecosystem have been specifying how many developers, standards and scenarios each jurisdiction must produce. This one specifies who is inside. Neither specifies what being inside gets you — and that omission is not an oversight in a document about membership. It is the same omission this series has been documenting since the Hunan plan’s “strive to enter the national front rank in contribution volume”: the instruments are elaborate about counting and silent about voice. A register that redraws the boundary without allocating a right is a register that will be read one way in Beijing and another way in every standards room the union says it intends to lead.

The measurable consequence arrives on a schedule. COPU has led the China Open Source Development Blue Book annually since at least 2022, and section eight of the outline assigns the 2026 edition. When it appears, the ecosystem chapter will either still name IBM, Red Hat, Google, Microsoft and GitHub as participants — because they are, and no register changes that — or it will not. What the 2026 blue book is permitted to see is the real answer to what this outline was for. Everything else in the document is a list of names.

Sources: