⚠️ 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 — 2026-11-21

🏛️ Policy & Regulation — The Bookkeeping Keeps Arriving: What the Foundation’s Own Filings Print Next to Its Conference Stage

1. The event, in the order it happened. On 2026-11-20 the 2026 OpenAtom Developer Conference (2026开放原子开发者大会) opened in Zhengzhou, Henan. Present: Wang Song (王崧), member of the Henan Provincial Standing Committee and party secretary of Zhengzhou; Xie Shaofeng (谢少锋), chairman of the OpenAtom Open Source Foundation; Wang Weiwei (王威伟), deputy director of the information technology development department of MIIT (工业和信息化部信息技术发展司); Zhuang Jianqiu (庄建球), mayor of Zhengzhou; and Lü Jian (吕建), academician of the Chinese Academy of Sciences, who gave the keynote. The theme was “See open source, see change” (看见开源,看见改变). The conference is the foundation’s annual flagship and described in the official account as one of the country’s most influential annual open-source meetings.

2. The five signed projects, and the two founders, are the structural news. The article supplied through the manual feed lists a project donation signing ceremony (开源项目捐赠签约仪式) in which five projects — AISHPerf, RoboFusion, SpaceAICOS, ExusData and Kindling — entered the foundation’s “neutral, open and standardised” incubation system, covering AI compute benchmarking, embodied intelligence data and fusion frameworks, cloud-native observability and intelligent systems. It also records two new working committees: a humanoid robot open-source community working committee (开放原子人形机器人开源社区工作委员会) and the OpenLoong community working committee (OpenLoong社区工委会), the latter described as marking a move to “multi-party collaborative co-governance” (多方协同共治). Alongside: an agricultural-intelligence operating-system community launch, M-Robots OS 3.0, the OpenLET embodied-intelligence open-data matrix, and the LoongArch Community — 22 founding members across enterprises, universities and research institutes.

3. The city’s contribution, stated as the numbers it chose to state. Zhuang Jianqiu: Zhengzhou has nearly 40,000 software and internet-related enterprises, an industry scale above CNY 130 billion, and 68,000 PFLOPS (6.8万P) of installed compute — the “foundation for building an open-source city”. He described the city’s policy frame as “software-hardware coordination, digital-real integration” and its aim as “building an open-source city jointly with the foundation” (开源之城共建启动仪式), with application scenarios to be opened in manufacturing, transport, energy, agriculture, government affairs and education. MIIT’s Wang Weiwei gave the national account: China “contributes more than 2.6 million active developers to the world,” an open-source talent pool that is “large in scale and structurally diverse,” and has “given rise to a batch of projects with industry competitiveness and international influence.” He summarised the department’s line in eight characters — “open source is born of developers and thrives on developers” (开源因开发者而生、因开发者而兴) — and restated the priority this series has tracked all year: “talent evaluation oriented to contribution” (以贡献为导向的开源人才评价), plus continued support for “incubation, collaborative development and security governance” across the whole chain.

4. What the foundation’s statutory filings print, because the same day is the right day to read them. OpenAtom is a national non-profit foundation founded in June 2020, registered with the Ministry of Civil Affairs and supervised by MIIT, with founding donors including Alibaba, Baidu, Huawei, Inspur, Qihoo 360, Tencent and China Merchants Bank. Its annual-inspection work reports (年度工作报告) filed to the Ministry of Civil Affairs are image-only PDFs on its disclosure page; the Europe–China Digital Commons Observatory OCR’d the 2020, 2021, 2022 and 2023 reports and published the totals (evidence level A for the filings, OCR re-reading pending). Across those four years: donation income CNY 223.6 million, of which foreign donations were zero in every single year; government subsidy CNY 1.46 million in 2020, the same figure in 2021, and CNY 20.7 million in 2023 — of which CNY 20.2 million was restricted to the “open-source project ecosystem cultivation platform” (开源项目生态培育平台); total income CNY 270.5 million against total expenditure CNY 171.0 million; charitable-activity spending rising from CNY 2.2 million (2020) to CNY 93.7 million (2023); administrative overhead at 3.95 per cent in 2023; net assets of about CNY 130 million at end-2023. Named large donations tell the concentration: in 2023, Huawei Device (CNY 15.772M) and Huawei Technologies (CNY 12.08M) together with ChinaSoft International (CNY 8.1M) made up CNY 35.95 million, 48 per cent of the year’s donations — and CNY 6.49 million of OpenHarmony’s spending went to Huawei Cloud services. The foundation’s own Xinhua-reported cumulative figure is more than CNY 600 million raised since 2020; the two filed years the observatory reconciled come to roughly CNY 139 million, and no 2024 or 2025 report is published on the disclosure page as of the observatory’s 2026-11-08 update.

5. The mobilisation figure, restated as a ratio. MIIT’s 2.6 million active developers worldwide is not a registration count for a conference, and the state has been using it since at least 2026-11-20’s Zhengzhou stage. It is a stock measure with no contributor definition attached — the same shape as Shanghai’s “3 million developers gathered by 2027” that this series examined on 2026-11-20, and the same shape as openEuler’s “30,000-plus contributors” examined on 2026-11-18. Three city-scale and one national-scale number, four different units, no published denominator.

Institutional Analysis & Commentary:

The foundation is not hiding anything. That is precisely what makes its bookkeeping the most informative document in Chinese open source this month, and it is more informative than the stage.

First, take the disclosure seriously, because it is better than the sector’s average and it should be credited before it is read closely. Filing annual work reports to the Ministry of Civil Affairs and leaving them on a public page where an outside researcher can OCR them is not nothing; KAIYUANSHE publishes a detailed but unaudited account, COPU publishes nothing public at all, and most Chinese industry bodies in this space publish nothing that can be reconciled. OpenAtom’s filings are statutory, machine-readable in principle, and arithmetically cross-checked — the observatory verified that income minus expenses equals the change in net assets year by year. On Williamson’s L2, that is a real institutional asset: an organisation that can be audited from outside is governed differently from one that cannot. The finding below is not a scandal, and it should not be read as one. It is a balance sheet, and a balance sheet tells you what an institution is not by what it says but by what it cannot stop revealing.

Second, the composition is the argument, and it is an endogenous-funding argument in Coase’s original sense. Zero foreign donations in four consecutive filed years is not an accident of exchange rates. It is the structural signature of a domestically-financed, ministry-supervised vehicle whose donors are the companies whose projects it hosts. And that produces a governance consequence no charter can undo: the party that can withdraw funding is the party whose project sits inside the portfolio. In 2023, Huawei-affiliated entities gave 37 per cent of all donations while OpenHarmony took 28 per cent of charitable spending — and a meaningful share of that spending returned to Huawei Cloud as paid services. The observatory’s own description is exact and should be quoted rather than paraphrased: “part of the donated money flows back to the donor as cloud-service revenue. Related-party disclosure covers this, but it is not arm’s-length.” This is not corruption; it is what a foundation looks like when its endowment is its beneficiaries. The neutral-governance claim — “发挥平台的中立治理优势”, the neutral-governance advantage the foundation cites for itself — is therefore not a claim about who funds it. It is a claim about what the foundation does that no donor would do unaided: hold the mark, absorb the legal work, run the CI/CD that everybody needs and nobody wants to pay for. The filings support that claim and it is a real service. They simply show that the foundation’s neutrality is procedural, not financial, and the two are different institutions.

Third, and this is the finding the Zhengzhou stage finally makes checkable: the Chinese foundation has opened its doors and is measuring its intake, and it still has no output metric. Count the day’s instruments. Five projects signed, four more communities or working committees stood up, one version released, one data matrix published, one processor-architecture community chartered with 22 members. Every one of those is an input: something entered, something joined, something was founded, something was signed. Not one number on the stage described what the existing portfolio produced — merged patches, upstream acceptances, releases, downstream products, maintainers paid. And the filings show the same shape: the 2023 accounts itemise income and expenditure by project, which is a cost view, not an output view. This is not a Chinese peculiarity — most foundations report disbursement rather than effect — but it is a peculiarity of this moment, because the intake machinery has just doubled in surface area. When a foundation that graduated OpenHarmony and openEuler in 2025 and held 59 projects across 18 fields by mid-2026 (per its own June 2026 account: 10 with working committees, 2 graduated) adds five projects and four communities in a single morning, the ratio that matters is how much of the existing portfolio is maintained versus how much is merely hosted — and no document in the announcement, and no filed account, states it.

Fourth, the LoongArch Community deserves to be read against that gap rather than against its press release, because it is the one item on the stage with a charter that matches the hard problem. “IP cores, binary translation, a base operating system” is not a marketing taxonomy; it is the actual dependency chain of an instruction set, ordered correctly, from silicon to application. What remains unaddressed is the part the charter names last and the code must supply first: who writes the netlist, who reviews the patch, who pays the maintainer. The 22 member organisations are a founding roster, and a founding roster is a commitment of intention, not a schedule of contributions. This series will return to that question in the entry where the code, not the ceremony, is the evidence.

Fifth, the honest caveat, stated so that the balance sheet is not misread as a verdict. CNY 93.7 million of charitable spending in 2023 at 3.95 per cent overhead is a lean, working organisation, and roughly half of that spending went to open-source engineering and infrastructure — code hosting, CI/CD, compatibility testing, licensing work. That is a materially better allocation profile than the programme’s press coverage suggests, and it contradicts any reading of the foundation as an events business with a legal annexe. The Mulan PSL v2 licence the foundation stewards is OSI-approved, which makes it a real counterpart for international licence work; the observatory’s repo check of 35 OpenAtom repositories found every one under an OSI-approved licence, and this series itself today verified that the newest LoongArch hardware project ships under it. Those are not small achievements. They are the achievements the balance sheet is actually consistent with — and they are all downstream-of-admission work, which is exactly where the money and the attention in this system are.

Sixth, the falsifiable tests. Does the foundation publish a 2024 or 2025 annual report — the observatory found neither as of 2026-11-08, and the answer within the year decides whether the disclosure regime is a practice or a phase. Does any 2026 or 2027 report itemise output per project — merged commits, maintainers, releases, upstream acceptances — rather than income and expenditure? That single change would make the foundation’s own accounting the best instrument in the ecosystem. And: does MIIT’s “2.6 million active developers” acquire a definition — annual contributors to any hosted project, or registrants? The first number is policy-relevant; the second is a stage number, and the distance between them is what a contribution-oriented evaluation system would have to measure.

Checkable questions.

(i) Of the 59 projects the foundation reported hosting in June 2026, how many received a release in the last twelve months, and how many have a maintainer who is not employed by the donating company? Both answers exist inside the foundation’s own systems and neither appears in any published document.

(ii) Was the CNY 20.7 million 2023 government subsidy a one-off or a run-rate? The filings show CNY 1.46 million in 2020 and 2021, nothing in 2022, and CNY 20.7 million in 2023 — a fifteenfold step that no document explains, and the granting body is not named in the filing.

(iii) Does the new humanoid-robot and OpenLoong working-committee governance give votes to anyone who is not a funder? “Multi-party collaborative co-governance” is the right aspiration; a co-governance body whose members are all donors is a donors’ council with a new name.

Sources: 2026 OpenAtom Developer Conference opens in Zhengzhou — official account report (2026-11-20; the Wang Song, Xie Shaofeng, Wang Weiwei, Zhuang Jianqiu and Lü Jian attendance; the “see open source, see change” theme; MIIT’s more-than-2.6-million-active-developers-to-the-world figure and the contribution-oriented talent-evaluation line; Zhengzhou’s nearly 40,000 software and internet enterprises, CNY 130 billion-plus industry scale and 68,000 PFLOPS of compute; the Zhengzhou open-source-city co-construction launch; the five donated projects AISHPerf, RoboFusion, SpaceAICOS, ExusData and Kindling; the humanoid-robot community working committee and the OpenLoong community working committee; the agricultural-intelligence SOS community, M-Robots OS 3.0 and the OpenLET data matrix; the LoongArch Community with 22 founding members) · Zhengzhou Municipal Government via Sohu — 2026开放原子开发者大会在郑州举行 (2026-11-20; the attendee list, Wang Weiwei’s and Zhuang Jianqiu’s remarks, and the conference’s official framing) · Europe–China Digital Commons Observatory — Chinese open-source foundations and OpenAtom finances (updated 2026-11-08; OpenAtom’s MIIT supervision and June 2020 founding, the four-tier donor structure, the OCR’d 2020–2023 annual-inspection filings with CNY 223.6 million of donations all domestic, the CNY 20.7 million 2023 government subsidy of which CNY 20.2 million restricted to the ecosystem-cultivation platform, CNY 270.5 million total income against CNY 171.0 million expenditure, 3.95 per cent administrative overhead, 2023 Huawei-affiliated donations of CNY 27.85 million at 37 per cent and CNY 6.49 million of OpenHarmony spending paid to Huawei Cloud, the absence of any 2024 or 2025 report, and the check that all 35 surveyed OpenAtom repositories use OSI-approved licences) · Xinhua — OpenAtom fundraising above CNY 600 million (2026-06-25; the foundation’s self-reported cumulative figure) · OpenAtom Foundation — 7 projects sign donation agreements at the 2026 Open Source Ecosystem Conference (2026-06-25; 59 projects across 18 fields, 10 with working committees, OpenHarmony and openEuler graduated in 2025) · This series’ 2026-11-18 briefing — the contribution instrument that does not exist


1. The announcement. On 2026-11-20, at the Zhengzhou conference, the OpenAtom LoongArch Community (开放原子龙架构社区) was formally established — jointly built by Loongson Technology (龙芯中科) and the OpenAtom Open Source Foundation, described as the country’s first comprehensive open-source community in the field of a self-developed instruction set, with the objective of building “the world’s leading LoongArch open-source innovation base” and “further consolidating the root base of domestic CPUs”. It is to be planned, incubated and operated by the foundation, and it will concentrate on three open-source special projects: IP cores, binary translation, and a base operating system, aiming at “the first ecosystem co-construction practice in the domestic self-developed CPU architecture field, from chip design to terminal application”. Twenty-two founding member organisations are reported, spanning chips, operating systems, complete machines and research institutes. Three flagship projects are stated to be online: a LoongArch open processor core centred on OpenLA132, the LoongArch binary translators LATU and LATS, and a base operating system called LoongHong. A community site, openloongarch.cn, is also stated to be online.

2. The charter’s own principles, quoted because they are the testable part. The community states that it will “adhere to both independence and openness” (坚持自主与开放并行), continue to incubate and maintain high-quality open-source software and hardware projects, push LoongArch into more application scenarios, and — the sentence that matters most — “actively align with international upstream open-source communities” (积极对接国际上游开源社区), with the stated goal of building an open platform that is “rooted in China and globally leading”. It further reported that at the concurrent openKylin open-source forum, Loongson was awarded the openKylin Community’s Platinum Donor of the Year (2026年度白金捐赠人), and that two Loongson engineers received project promotion contribution and contribution awards respectively for contributions to openKylin.

3. The code, measured rather than asserted, on 2026-11-20. This is the evidence the announcement does not contain and cannot contradict, because it is public and timestamped.

  • github.com/loongson/OpenLACores — the repository behind the OpenLA132 / OpenLA264 processor design and verification work named in the announcement — was created 2026-10-08, last pushed 2026-10-26, and holds six commits from four distinct authors. One of those six commits is a merge of pull request #1, opened by an outside contributor (MarsDoge) on 2026-10-16 and merged on 2026-10-26; the rest are from Loongson-side engineers. The repository is 17 stars and 4 forks, and ships under Mulan PSL v2 — the licence the foundation stewards, verified through the GitHub API.
  • github.com/lat-opensource/lat — the LoongArch binary translator, a project this series flagged as donated to the foundation in June 2026 — was created 2025-03-26, last pushed 2026-11-19, and stands at 251 stars. Its licence is reported by GitHub as NOASSERTION, and the observatory’s earlier table records it as GPL-2.0, QEMU-derived — a fact worth keeping in view for anyone reasoning about copyleft and the architecture’s own toolchain.
  • LoongHong, the announced base operating system, did not surface in a GitHub repository search on 2026-11-20, and openloongarch.cn did not respond to a direct request from this run’s host at the time of writing. Neither observation proves absence — both may be hosted on GitLink, Gitee, AtomGit or a domestic network this series cannot reach — and both are recorded as unverified rather than negative.

4. The upstream reality, which is the standard any “align with international upstream” claim is measured against. LoongArch has been in the mainline Linux kernel since 5.19, and the LOONGARCH entry in the kernel’s MAINTAINERS file is listed as Maintained, with Huacai Chen as maintainer, Wang Xuerui as reviewer and a public loongarch@lists.linux.dev mailing list. That is not a Chinese document. It is the kernel’s own record, and it is the strongest single fact in this entry — because it means the architecture’s most consequential piece of software already sits in a governance structure that is not Chinese, does not report to a ministry, and cannot be chartered by a foundation. The GCC and QEMU ports and the OpenJDK trees that Loongson mirrors on GitHub tell the same story: the upstream is upstream.

Institutional Analysis & Commentary:

The LoongArch Community is the most substantively well-designed institution OpenAtom has chartered this series has seen, and the six commits in its flagship repository are the most honest thing published about it. Both statements are true and they belong in the same paragraph.

First, credit the design, because the design is correct and this series has spent six weeks arguing that designs of this kind do not exist. Three special projects — IP cores, binary translation, base OS — is the right decomposition of an instruction-set ecosystem, in the right order, and it is a decomposition that only someone who has built one would produce. An ISA is useless without a core that others can license or embed, a translation layer that runs the software the ISA does not yet have, and an operating-system base that gives downstreams something to fork. Naming all three, and naming binary translation in particular, is an admission that is rare in this genre: it concedes that the ecosystem’s immediate requirement is compatibility with software written for somebody else, not originality. That concession is the strongest argument the announcement makes for itself.

Second, the arithmetic. 22 member organisations, 132 days per member, and six commits. The community is announced on 2026-11-20; the repository at its centre was created on 2026-10-08 — 43 days before the ceremony, not years, which is the correct way to read a new hardware project and also the way that makes the ratio legible. In those 43 days the flagship code received six commits, of which one came through a pull request from someone outside Loongson. Twenty-two members, six commits, one outside contributor. Compare that with the exact same week’s openKylin award citation, where Loongson is honoured for two individual engineers’ contributions — a company being thanked for contributing to somebody else’s community is a fact about where its engineers’ time actually goes. The honest reading is not that the community is a failure; it is that a founding roster measures intent and a commit log measures capacity, and for now the two numbers differ by a factor that no press release will close. The foundation has opened an intake — 22 organisations, three special projects, a charter with the right words in it — and the ecosystem’s question was never intake.

Third, and this is where the LoongArch case becomes genuinely interesting rather than merely thin: the architecture’s most important software is already governed by a charter nobody in Zhengzhou wrote. The Linux kernel’s LOONGARCH entry is Maintained, on a public mailing list, with named maintainers and no foundation in the loop. LoongArch has been in mainline since 5.19. That means the community’s own stated goal — “actively align with international upstream open-source communities” — is not aspiration; it is a description of where its centre of gravity already is, and it is a stronger position than any Chinese project in this series has occupied. An ISA that is in mainline Linux, GCC and QEMU has something no amount of subsidy buys: its patches go through linux-arch and Linus, and its maintainers answer a mailing list with a worldwide subscriber base. The prize of the LoongArch Community is therefore not a new commons; it is an interface between a national programme and an existing commons — and interfaces of that sort are governed by whoever controls the merge, which in this case is neither Loongson nor OpenAtom.

Fourth, the licence, which is where the two regimes actually touch. OpenLACores ships under Mulan PSL v2 — a permissive, OSI-approved licence, which is the right choice for a processor core: permissive is what lets a Chinese fabless company embed a core without a licence review, and Mulan PSL v2 being OSI-approved means an international adopter’s counsel can clear it without inventing a legal opinion. LAT is GPL-2.0, QEMU-derived — copyleft, as QEMU requires. So the portfolio demonstrates something this series has argued is rare: the foundation is deploying the right instrument per artefact rather than one ideological default. A permissive licence on the hardware, a copyleft licence on the derived translator, both OSI-approved or GPL-family, both legible to an outside lawyer. This is the one place in the whole Zhengzhou programme where the output is designed for a reader outside the system — and it is worth saying plainly that it is the opposite of the pattern this series documented in model licences, where the newest instruments are bespoke and the toll is in the annexe.

Fifth, what the numbers would have to show for the charter to mean what it says. The charter’s own language supplies the tests. “Independence and openness in parallel” is testable by a single number: the proportion of merged commits authored by people not employed by Loongson or the 22 members — and “the first ecosystem co-construction practice in the domestic CPU architecture field” is a claim about sharing that is falsified by a commit log showing one employer. “Align with international upstream communities” is testable by another: the number of LoongArch patches authored outside China that land in mainline, and the number of the community’s own projects that get accepted upstream rather than hosted. “Multi-party governance”, which the day’s other communities claim, is testable by a membership list that names organisations with votes and not just logos. None of these is an unreasonable demand; each of them is a number the community’s own tooling produces as a side effect.

Sixth, the caveat that protects the finding from being read as a dismissal. Hardware is not software. Six commits in six weeks on an RTL design project is a normal shape for a repository that has just been opened, and comparing it with the commit velocity of a mature software community would be an unfair measurement dressed as a rigorous one. OpenLA132 was published at all — a processor core, under a permissive licence, with a verification directory, a documentation set and an English README — and that is a materially more open artefact than most of the projects this series has audited, several of which publish no code the public can read. The point is not that the number is small. The point is that this is the first time in six weeks of Chinese open-source announcements that an outside observer could open the flagship repository on the day of the ceremony and read the code, the licence and the commit history — and that alone makes the LoongArch Community the most auditable institution on the Zhengzhou stage.

Checkable questions.

(i) Who maintains OpenLA132, and are any of the four commit authors on the community’s technical steering body? A governance document that names maintainers before it names donors would be the first of its kind in this series.

(ii) Will the LAT translator’s GPL-2.0 obligations be honoured for downstream redistribution, and who audits that? A QEMU-derived binary translator is the architecture’s compatibility bridge and its licence is copyleft; if an embedded vendor ships a modified LAT without source, the community’s credibility with any foreign upstream is spent.

(iii) Does the community publish a member list distinguishing paying donors from contributing organisations? Twenty-two “founding members” spanning chips, OS, complete machines and research institutes is plausible; a breakdown by contribution would make the number mean something, and the foundation already knows the difference — it operates a four-tier donor structure.

Sources: IT之家 via Sina Tech — 我国首个自主指令集综合开源社区,开放原子龙架构社区正式成立 (2026-11-20 17:44; the joint founding by Loongson and OpenAtom; 22 founding member organisations; the three special projects of IP cores, binary translation and base OS; OpenLA132, LATU/LATS and LoongHong; the “first comprehensive community in a self-developed instruction set” claim) · QQ Tech — 龙芯中科:我国首个自主指令集综合开源社区成立 (2026-11-20 18:35; the “independence and openness in parallel” and “actively align with international upstream open-source communities” principles; the community site openloongarch.cn; the openKylin Platinum Donor of the Year award and the two Loongson engineer awards) · GitHub API — loongson/OpenLACores (repository created 2026-10-08, last pushed 2026-10-26, six commits from four authors, one merged PR from an outside contributor, 17 stars and 4 forks, Mulan PSL v2 licence) · GitHub API — lat-opensource/lat (created 2025-03-26, last pushed 2026-11-19, 251 stars, licence reported as NOASSERTION and recorded elsewhere as GPL-2.0 QEMU-derived) · Linux kernel MAINTAINERS — the LOONGARCH entry (status Maintained; Huacai Chen as maintainer, WANG Xuerui as reviewer, loongarch@lists.linux.dev) · OpenAtom Foundation — 7 projects sign donation agreements (2026-06-25; the LoongArch binary translator among the projects entering incubation) · This series’ 2026-11-20 briefing — the layer that has no owner


🏗️ Institutional Change — An Instrument That Licensed Its Own Duplicate: The Secure-and-Reliable Assessment Enters Its First Mass Expiry

1. The claim under examination, and why it deserves a careful reading rather than a rebuttal. On 2026-11-18 at 13:10, the 早安胡同 (Good Morning Hutong) WeChat account published “The national-assessment list is expiring: UnionTech and Founder are not being thrown out, it is time to hand in the paper” (国测名单到期:统信方德不是被踢出局,是到了该交卷的时候). The article arrives in the manual feed for 2026-11-20 and is 2,182 characters. Its occasion is a widely circulated rumour that UnionTech (统信) and China Standard Software / Founder (中科方德) were being removed from the national assessment list. The article’s answer, reached by reading the China Information Technology Security Evaluation Center’s (中国信息安全测评中心) “secure and reliable assessment” (安全可靠测评) notices from the first batch in December 2023 through the ninth in September 2026, nine notices in all, is procedural: the first batch was published on 2023-12-26 with an explicit three-year validity expiring 2026-12-26 — thirty-odd days from publication — so nothing has been removed; a certificate has reached its term.

2. The rule that makes this an institutional event rather than an administrative chore. The official name of the assessment is 安全可靠测评 (secure and reliable assessment), jointly issued by the China Information Technology Security Evaluation Center and the State Secrecy Science and Technology Evaluation Center. The article’s central technical point: when a notice lists a product, the kernel version is a mandatory field, and when the kernel is iterated, the product is treated as an entirely new product requiring full re-testing — there is no renewal channel. The first batch (2023-12-26) listed six operating systems, all at grade I: Kylin Desktop V10, Kylin Server V10, UnionTech Desktop V20, UnionTech Server V20, Founder Desktop V3.1 and Founder High-Trust Server V4.0. Between then and now: Founder V5.0 desktop and UnionTech V20 (kernel 5.10) in May 2024; Founder Pro V5.0 (kernel 6.6) in September 2025; UnionTech V25 desktop (kernel 6.6) in January 2026. The two server versions from the original cohort — UnionTech Server V20 and Founder High-Trust Server V4.0 — have not appeared in any notice since 2023. The article’s formulation is precise and worth preserving: what has stalled is two old server models, not two companies — “product models and corporate entities must not be conflated.” The kernels went from 4.19 to 6.6; a vendor cannot renew a 2023-vintage kernel in place, so expiry review is a full re-entry.

3. The grading framework itself has been tightening, which is the part of the article that carries real information. In the 2023 first batch, grade I was the ceiling. By September 2026, Kylin Advanced Server V11 SP1 took grade II — the first domestic server operating system to do so — and HarmonyOS Desktop V1.0 took grade II earlier in 2026; Loongson’s 3A6000 (grade C) unlocked grade III in the September notice. In three years the framework has expanded from a single grade I into grades II and III. And the procurement consequence is documented: a Guizhou public procurement notice states explicitly that equipment must meet secure-and-reliable assessment grade II, and the central-agency desktop operating-system list now names products beyond Kylin, UnionTech and Founder. The article’s own summary: without a corresponding assessment result, a bidder has no standing to participate in the project at all — it is a bid-entry ticket (投标入场券). Crucially, it also states what the threshold does not do: already-deployed, already-running terminals and servers do not stop working because the list updates; it constrains new project tenders.

4. Why the article is worth taking seriously on its own terms. It is semi-official in tone and procedural in method — it does not dispute the existence of the list, does not dispute the gate, and does not argue that the gate should go. It combs nine official notices, dates the first batch, quotes the validity term, explains the kernel-version rule, and separates models from companies. It even concedes the part that hurts: “the path of getting by on an old version’s credentials is closed”, and the long-run effect favours “vendors who genuinely polish their products”. That is the voice of an insider explaining a rule to a alarmed market, and the alarm is legible: a rumour that two major vendors had been expelled spread widely enough to require an article.

Institutional Analysis & Commentary:

This is the first document in six weeks of Chinese open-source policy that describes a certification regime making itself harder, and for that reason alone it is the most institutionally interesting thing in today’s feed. It is also a clean illustration of a governance mechanism this series has not yet had the evidence to describe: a conformity-assessment instrument whose periodic re-testing rule quietly licensed its own duplicate.

First, the mechanism, stated as a mechanism rather than a policy. A certification that grants permanent status creates a rent: once obtained, the certificate is an asset that costs nothing to hold and cannot be competed away, and the holder’s rational strategy is to defend the credential rather than improve the product. A certification that grants three years, non-renewable, with a mandatory kernel-version field creates the opposite incentive structure: the certificate is a perishable asset, and holding it requires shipping a new product on a schedule you do not control because the standard keeps moving. The author’s own analogy is exactly right and it is a good one: this is not renewing a driving licence, it is re-registering a car after an engine swap. What the three-year cycle has built, without anyone describing it as such, is a replication mandate inside a procurement gate — and replication mandates are the one institutional device that reliably converts a compliance cost into an engineering one.

Second, and this is the empirically important half: the mandate is visibly working, and the evidence is that the grades moved. Three years ago the entire domestic OS field sat at grade I, the ceiling, and the assessment distinguished nothing because everyone passed at the same level. Today there is a grade II — taken first by a HarmonyOS desktop, then by Kylin’s Advanced Server V11 SP1 — and a grade III unlocked by Loongson’s 3A6000. A grading scheme that produces differentiation is doing the thing grading schemes are for; one that produces a single pass grade is a membership list with a rubric attached. The 2023 regime was closer to the second; the 2026 regime is closer to the first. And the second-order effect is visible in the province: Guizhou writing “grade II” into a procurement intention means the latest standard immediately becomes a market requirement, not a certificate on a wall. That is how a technical standard acquires teeth in any system — not by enforcement, but by being cited by a buyer.

Third, the honest and uncomfortable part, which the article does not address and which this series must: a procurement gate that requires a certificate is a barrier to entry, and barriers to entry have incumbency effects regardless of how rigorously they are administered. The article’s framing is that the threshold constrains new tenders only and leaves running systems alone, which is materially fairer than a revocation would be — and still means that a startup without a certificate, or with a certificate at the wrong grade, cannot bid. The article notes that the central-agency list now includes products beyond the big three, which is evidence against lock-in; but it does not say how many, and it does not say what the cost and duration of a full re-test are, which is the number that determines whether the gate selects for quality or for balance-sheet capacity to absorb a compliance cycle every three years. A rigorous gate administered over a narrow field does not necessarily produce better software; it can produce better-capitalised software. Nothing in the article lets a reader distinguish those two outcomes, and the three-year cycle now means every vendor in the field faces the question simultaneously — the first mass expiry is a natural experiment, and its result will be visible as a shift in the vendor list.

Fourth, the reflexivity the article performs, which is its most interesting property as an object. It was written to correct a rumour that read a procedural event as an exclusion, in the voice of someone inside the system, for a market that reacted to a list without reading the notices. That is a legibility problem created by the instrument itself: a three-year cycle whose expiration dates are buried in a mandatory-field rule produces, once, a mass expiry that looks like a purge. The institution was well designed and badly signposted — and the cost of bad signposting is a rumour, which is a real cost in a market where a vendor’s stock and pipeline both move on list membership. The author’s contribution is not to argue about the gate but to publish the calendar, and the general lesson is that a rule nobody can read is a rule that will be misread, especially when it bites all at once.

Fifth, what this connects to in the wider picture, and it is the connective tissue the day needs. Yesterday’s briefing examined Shanghai paying up to CNY 20 million to import a distribution layer and Moonshot putting a revenue threshold in a licence. Today’s instrument is a third kind: a procurement gate that renews on a technical clock. All three are attempts to govern a layer where the open artefact is free and the money is in the access — and all three, notably, are calendars: a licence’s annual-revenue trigger, a city plan’s 2027 targets, a certificate’s 2026-12-26 expiry. The instrument this system is best at building is not a charter or a commons; it is a schedule with a consequence attached. That is also its limitation: a schedule selects for organisations that can meet schedules.

Sixth, the caveat that keeps this honest. None of this is a verdict on the vendors involved, and the article is right to insist on the distinction. UnionTech and Founder are real companies with real products sold in volume; the fact that two server models from a 2023 cohort have not been re-submitted says something about product roadmaps and where each vendor chose to spend its engineering years, not about removal. The inference that should be drawn is narrower and more interesting: the oldest server operating-system versions in the domestic field are now the ones outside the gate, which is itself a statement about the pace at which the domestic server-OS market has been re-platformed from 4.19-era kernels to 6.6-era ones in three years.

Checkable questions.

(i) Will a new notice appear before 2026-12-26 listing new versions of UnionTech Server and Founder High-Trust Server? The article effectively predicts the wait and see. If nothing appears, the two models lapse — and the question becomes whether procurement notices written for grade II accept their replacements.

(ii) What does a full re-test cost, in money and months? This is the number that decides whether the gate selects for quality or for capital, and it is the question every Chinese OS vendor’s strategy team is asking this week.

(iii) How many vendors appear in the central-agency desktop OS list beyond the big three, and at what grades? The article asserts diversification; the list itself would quantify it, and the grade distribution would show whether the bar is being lowered to widen the field or held while the field widens.

Sources: 早安胡同 via WeChat — 国测名单到期:统信方德不是被踢出局,是到了该交卷的时候 (2026-11-18 13:10; the nine notices from 2023-12 to 2026-09; the official name and the two joint issuing centres; the 2023-12-26 first batch with a three-year validity to 2026-12-26 and six grade-I operating systems; the mandatory kernel-version field and the no-renewal rule; the 2024-05, 2025-09 and 2026-01 new-version notices; the absence of UnionTech Server V20 and Founder High-Trust Server V4.0 from later notices; Kylin Advanced Server V11 SP1 and HarmonyOS Desktop V1.0 at grade II and Loongson 3A6000 at grade III; the Guizhou procurement requirement for grade II; the central-agency desktop list’s expanded vendor set; the “bid-entry ticket” framing and the statement that deployed systems are unaffected) · This series’ 2026-11-20 briefing — an instrument for a company or a city, and none for a commons


🔍 WeChat Monitor

Two of the three items in the manual feed for 2026-11-20 were extracted and used as primary sources above (china-daily-english-input, raw file raw/2026-11-20.md, repository pulled and up to date at the time of this run; source files read in full and cross-checked against the independent reporting cited).

  • 早安胡同, 国测名单到期:统信方德不是被踢出局,是到了该交卷的时候 — 2,182 characters, published 2026-11-18 13:10. Used as the primary source for the Institutional Change entry: the secure-and-reliable assessment’s first mass expiry, the kernel-version/no-renewal rule, the grade framework’s expansion from I to III, and the procurement consequence. Note the two-day lag between the article’s publication (2026-11-18) and its appearance in the feed (2026-11-20) — recorded here because the entry’s dating depends on it.
  • 开放原子 (OpenAtom Foundation official account), 2026开放原子开发者大会在郑州开幕 — 3,017 characters, published 2026-11-20 19:51. Used as the primary source for the Policy & Regulation entry: the attendee list, MIIT’s and Zhengzhou’s remarks, the five donated projects, the two new working committees, and the LoongArch Community’s 22 founding members. Cross-checked against the Zhengzhou municipal government’s own account of the same event and against IT之家’s LoongArch report, both cited above.
  • 龙蜥社区 (OpenAnolis), 「2026 龙蜥社区年度优秀贡献者」评选启动 — 408 characters, published 2026-11-20 15:00. An event notice, not institutional news: contributor award nominations and a call for speaking topics for the 2026 Dragon Lizard Operating System Conference on 2026-12-29 at the Zhongguancun National Independent Innovation Demonstration Zone Convention Center in Beijing, across five tracks (Agentic OS, AI Infra, AI Security, Agentic CPU, AI Native) plus a SkillHub workshop, with topic submissions closing 2026-11-23. Noted rather than analysed; it belongs to 2026-12-29, and this series will read the conference’s outputs rather than its call for papers. The feed’s structured slots for the fifteen monitored accounts (OpenAtom, CCF’s open-source development committee, Huawei Open Source, Mulan, COPU, Tiangong Kaiwu, Ming Says, BAAI FlagOpen, KAIYUANSHE, Silicon Time, AtomGit, the Linux Kernel Developers Conference, GOSIM/CSDN, the Huang Daniang Tea-Swotting House and Yetian’s Book) were all empty for the day apart from the OpenAtom item quoted.

Dedup against the previous three briefings. The 2026-11-20 entry covered Moonshot’s K3 revenue-share against AWS and Microsoft, Shanghai’s Model Ark import and the Menlo Ventures report; none of those recrosses below. The 2026-11-18 entry covered Zhejiang’s seven-department opinion, code as a line item and openEuler’s 22 million installations and 61 per cent new-share forecast — Zhengzhou’s 68,000 PFLOPS of compute and 2.6-million-developer national figure are new; openEuler appears here only in passing as a graduated project with a maintenance question attached, and its installation and market-share figures are not re-reported. The 2026-11-19 entry covered a listed bank’s national governance assessment, the K3 licence’s revenue threshold and a listed model company’s repapering — today’s secure-and-reliable assessment is a different instrument (a pre-licensing, government-administered, three-year procurement gate, as distinct from a bank’s internal four-part governance apparatus), and the entry is built on a different document with different dates. The OpenAtom 2020–23 statutory filings were cited in this series on 2026-11-10; they are reused here deliberately and only as a continuation — 11-10 used them to reconcile funding against the Xinhua figure for a European-observatory analysis, whereas today they are read against the Zhengzhou stage for what they say about government subsidy, donor concentration and the absence of an output metric. LoongArch has not appeared in this series before: grep across the November 11–20 entries and ~/.hermes/cron/output/bf436b4758a0/ returned no prior mention of the LoongArch Community, OpenLACores, LoongHong, LATU/LATS or the 22-member roster, and the repository measurements are taken fresh on 2026-11-20. The secure-and-reliable assessment (国测/安全可靠测评) has not been covered in any November entry.

🔍 Commentary — Three Calendars and One Commit Log: The Instruments This System Builds Best

First, today’s three documents are three varieties of the same instrument, and the instrument is a schedule. A licence whose trigger is annual revenue and a procurement gate whose validity expires on 2026-12-26 and a city plan whose targets land in 2027 are not, functionally, charters. They are calendars with a consequence attached — and on the evidence of six weeks, calendars are what this ecosystem builds superlatively well. A charter tells you who may join. A calendar tells you when you must show up again. The first is a governance question; the second is the question the system actually answers, over and over: a bank’s assessment renewed annually, a certificate every three years, a plan every five, a licence trigger every twelve months of revenue.

Second, the commit log is the counter-document, and this is the finding the series should carry forward. Twenty-two founding members and six commits in forty-three days, with one contribution from outside the founding company. The day’s three calendars all measure attendance. None of them measures production. A certificate measures whether a product was submitted; a plan measures whether a target was hit; a revenue threshold measures whether an agreement was triggered. Only a commit log measures whether anyone who did not have to show up, showed up — and the Chinese system’s own instruments cannot read it, because a commit log has no filing date, no certificate number and no payee. This is the same legibility gap the series has documented since October, now with a hardware architecture attached to it: the certificate regime can see that Loongson’s 3A6000 reached grade III; nothing in the regime can see that the open processor core behind the architecture received one outside pull request in its first six weeks. Both facts are in today’s briefing, they are about the same company, and the system reports the first at length and the second not at all.

Third, the licence and the gate are converging on the same truth from opposite ends, and the truth is about time. Mulan PSL v2 is a permanent grant with no calendar at all — that is its entire point, and it is why an outside counsel can clear it. The secure-and-reliable assessment is the opposite: a grant with an expiry date and no renewal path. Put them together and the Chinese open-source stack is being built out of two incompatible temporal regimes — perpetual licences on the artefacts and perishable certificates on the products those artefacts become. That works, and it has produced real adoption, but it means the commercial life of a Chinese open-source operating system is bounded by a three-year compliance cycle while its legal life is unbounded. A vendor planning past 2026-12-26 is planning a re-certification, not a release. And nobody in the whole apparatus has yet asked whether the two clocks should be the same clock.

Fourth, the connective tissue, and it is the same layer again. Shanghai pays for the distribution layer. Moonshot tolls the distribution layer. OpenAtom signs 22 members into a hardware layer. CAICT grades the products that sit on it. Four instruments, four owners, and not one of them is a commons with a published membership, a notice period, and an appeal. The point is not that any of them is illegitimate — each is doing something real, and the secure-and-reliable assessment’s move from a single pass grade to a graded framework is genuine institutional progress. The point is arithmetic: every one of these instruments selects for organisations that can meet a date, and a commons is the one institution that does not have a date. That is why a foundation ends up with a donor concentration and an events calendar, and why a rumour can outrun a notice: an institution organised around deadlines will be read through its deadlines, and its quiet, undated work — publishing a core under a permissive licence, keeping a mailing list alive, merging a stranger’s patch — will go unread even when it exists.

Fifth, the honest caveat, and it is a strong one this time. The LoongArch Community may well become the first item on this series’ ledger where a Chinese charter survives contact with the code, because its architecture is already in mainline Linux under a public Maintained entry with named maintainers, because its flagship core ships under an OSI-approved permissive licence, and because the announcement’s own stated principle — align with international upstream — is the one principle in the entire Zhengzhou programme that cannot be met by ceremony. Falsifiability is a feature. A community that says “we will align with international upstream” has written the test of its own success, and no provincial target can be substituted for the number of patches that get merged.

Sixth, what to watch, all of it checkable within a quarter. The OpenLACores commit log on 2027-01-20 — ninety days from the ceremony, the point at which a founding roster either becomes a contributor base or remains a press release. A 2024 or 2025 OpenAtom annual report on the disclosure page, which the observatory found absent as of 2026-11-08. The 2026-12-26 expiry, and any new notice or revised procurement list published before and after it. And whether any Chinese institution in this ecosystem publishes, for the first time, a number that counts contributors rather than attendance — because every instrument examined this week, without exception, counts the second and none counts the first.

Sources: IT之家 via Sina Tech — 开放原子龙架构社区正式成立 (2026-11-20) · GitHub API — loongson/OpenLACores repository metadata and commit history (2026-11-20) · OpenAtom Developer Conference official account report (2026-11-20) · 早安胡同 via WeChat — 国测名单到期 (2026-11-18) · Europe–China Digital Commons Observatory — Chinese open-source foundations and OpenAtom finances (2026-11-08) · Linux kernel MAINTAINERS — LOONGARCH entry · This series’ 2026-11-20 briefing — a toll in Seattle and a cheque in Shanghai · This series’ 2026-11-19 briefing — the refusal right and the licence that governs a margin · This series’ 2026-11-18 briefing — code as a line item