⚠️ 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 — 2026-11-11
🏛️ Policy & Regulation — The Evaluator Moved Upstairs: 2026 OSCAR and the Instrument That Grades Agents Instead of Projects
1. The event. On 2026-11-10 the 2026 OSCAR Open Source Industry Conference (2026 OSCAR开源产业大会) was held in Beijing under the theme “Open source without borders, intelligence creates the future” (开源无界 智创未来). It was hosted by the China Communications Standards Association (中国通信标准化协会) and organised by the China Academy of Information and Communications Technology (CAICT, 中国信通院) — the tenth such annual conference, and the second consecutive year in which the standardisation association has been the named host while the research institute has been the organiser. Opening remarks were given by Dai Xiaohui (代晓慧), vice-chair and secretary-general of the association, and by He Baohong (何宝宏), chief engineer of CAICT. For a schedule of comparison: the 2025 conference was held 2025-10-28 on the theme “Trusted open source engages ten thousand industries, boundless intelligence domain opens the future”, hosted and organised by the same two bodies.
2. The five things actually released, in the order the announcement lists them. This is the part to read closely, because three of the five are not about software at all.
- A new batch of “trusted open source” (可信开源) assessment results — trusted open-source governance maturity; trusted open-source supply chain (OpenChain dual assessment); SOSS open-source security national-standard conformity testing; and open-source governance tool (SCA) capability assessment. CAICT states it first proposed the “trusted open source” concept and has built the standard and assessment system around four goals: security, compliance, continuity, health.
- An extension of that same assessment system from the software supply chain to the AI supply chain — a “Trusted AI Supply Chain” (可信AI供应链) assessment system, with a first batch of results covering: enterprise office agent capability; agent development and management platform capability; code-security agent capability; enterprise intelligent programming assistant capability.
- A group standard from the China Institute of Communications, Technical Requirements for Compatibility of Terminal Software, Hardware and Operating Systems in the Power Internet of Things (《电力物联网终端软硬件与操作系统兼容性技术要求》).
- A joint launch of “six-network coordination, open IoT” standard-system construction (六网协同 开放物联标准体系建设) by CAICT together with China Southern Power Grid Digital Grid Research Institute, China Telecom Beijing Research Institute, Huawei, Sinohydro (南水北调) Water Network Smart Technology and China Mobile IoT (中移物联网) — stated purpose: break the deadlock of “scattered technical routes, incompatible interfaces, inconsistent thing-model semantics and cross-domain data that cannot flow or be reused”, and build “a unified IoT base for six-network coordination”.
- The establishment of an Open-Source AI Infrastructure (AI Infra) Software Expert Group (开源人工智能基础设施软件专家组), convened by CAICT, to work on “heterogeneous compute coordination and chip-enablement software”; and the starting of a FlagOS intelligent-computing-centre heterogeneous compute coordination capability evaluation, jointly by CAICT, the AI working committee of the Internet Society of China, and the Zhongzhi FlagOS open-source community.
3. The framing language, quoted because it is the entry’s subject. Dai Xiaohui: open source’s “strategic position keeps rising in national planning”; industry parties should “use standards pre-research as traction to raise governance capability, insist on promoting construction through evaluation and governing through use (以评促建、以用促治), and jointly build an open, collaborative and trustworthy open-source ecosystem.” He Baohong: “the boundaries of open source keep widening, the inclusive effect keeps amplifying, and the mode of collaboration is being rapidly restructured.” And from the two CAICT research presentations: Li Wei (栗蔚), deputy director of CAICT’s Cloud Computing and Big Data Institute, argued that “the value of open source is migrating from collaborative production to the application ecosystem” and that agents have become an important participating subject in the open-source ecosystem; Guo Xue (郭雪), head of the institute’s open-source and intelligent security department, argued that in “AI + software” the “production subject is shifting from people to agents, product form from people adapting to software to software adapting to people, and the business model from selling products to selling results.”
4. The line-up, which is itself a datum. Named speakers included Li Shaohui (李少辉), R&D director of the OpenAtom Open Source Foundation, on “community as foundation, ecosystem as forest”; Wang Lihiao, vice-president of Huawei Enterprise Business, on OpenHarmony; Tang Dan (唐丹), director of the Beijing Institute of Open Source Chip (BOSC), on RISC-V; Wang Hao (王昊) of China Southern Power Grid Digital Grid Research Institute on the Dianhong (电鸿) IoT operating system; and speakers from Changan Automobile, openJiuwen (an agent-interconnection project), CANN’s open-source community, Xiaomi, China Telecom Security, Desay SV, Qingcheng Jizhi and Inspur. The afternoon main forum was devoted to open-source AI and supply-chain security — including a talk titled “From SBOM to AIBOM: the new battleground of AI open-source supply-chain security.”
Institutional Analysis & Commentary:
An evaluation body that moves from assessing projects to assessing agents has changed what it is, and the change is legible in the vocabulary rather than in the announcements.
Consider what the “trusted open source” system actually is, coldly. It is a certification regime: an institute drafts a rubric, enterprises submit artefacts, the institute rules, and the ruling is publishable. Its unit of account has, for a decade, been the organisation and the artefact — governance maturity of a company, the compliance posture of a supply chain, the security baseline of a national standard, the capability of an SCA tool. Every one of those is a thing an enterprise can be audited for. Certification of this kind does real work: it converts an unobservable property (does this company actually govern its dependencies?) into a purchasable signal, and in a market where buyers cannot inspect suppliers, that is the only mechanism that makes trust scale. That is the strongest case for the institution, and it should be stated before the criticism.
The criticism is about what happens when the unit of account moves from the artefact to the actor. Yesterday’s assessments graded a repository or a company. Today’s new batch grades agents — an office agent, a platform for building agents, a code-security agent, a programming assistant. And the two research presentations say out loud why: agents have become “an important participating subject”, the “production subject is shifting from people to agents”. Read those two sentences as an institutional claim rather than a technical one and the whole conference reorganises. A regime that certifies agents is a regime that has concluded the contributor is now a thing that can be held to a standard. And a thing that can be held to a standard is a thing whose output can be registered, checked, and counted — which is precisely what the previous four weeks of this series have documented being built from the other direction: identity codes for agent terminals at MIIT on 2026-11-06, an agent identity and trusted-interconnection centre in Beijing from 2026-09-16, and a contribution-measurement platform at Zhejiang University on 2026-11-04 that analyses “the process and results of human–AI collaboration”.
That is the convergence worth naming: three separate institutions are independently building the same object — a verifiable identity for a non-human contributor — and none of them is a code forge. MIIT wants the terminal identifiable before it enters the network. CAICT wants the agent’s capability measurable before it enters a procurement. The university wants the collaboration process scored before it enters a transcript. None of these is a review queue, and all three presuppose one exists somewhere else. The gap this series has measured since October — twelve unanswered issues under a foundation that had just committed ¥450 million — is the gap between an ecosystem that has become very good at certifying inputs and one that has not yet built the institution that receives them.
The second thing the conference states, and it is more consequential than the assessment results: the value of open source is said to be migrating “from collaborative production to the application ecosystem.” Take that seriously rather than as rhetoric. If the value of open source lies in the application ecosystem, then the thing worth governing is downstream — what gets built on the code, who certifies the built thing, whether it can be procured. And that is exactly the shape of every instrument on the agenda: a compatibility standard for power-IoT terminals, an IoT base for six networks, an AI supply-chain assessment, a heterogeneous-compute evaluation. Every one of these governs deployment. This is not hypocrisy; it is a coherent theory of where the public interest lies, and it is the same theory the OpenAtom filings show in the money — adoption subsidies and platform-building, not maintenance. The corollary is uncomfortable and follows mechanically: an institution whose definition of value is downstream will fund downstream, and the upstream that the downstream depends on will be governed by whoever happens to still be paid to look at it.
Third, the honest counter-observations. (i) The six-network coordination workstream is a genuinely hard coordination problem — incompatible interfaces and inconsistent thing-models across power, water, telecoms and transport are real costs, and a compatibility standard is a legitimate answer to them; the question is only whether it is the first answer or the only one. (ii) The FlagOS evaluation is not a certification of a company but a capability benchmark for heterogeneous compute, jointly operated with the community that builds the software stack — a structurally different and more defensible instrument, because the object being measured is a public good with a named maintainer. (iii) Dai Xiaohui’s formula — 以评促建、以用促治, “promote construction through evaluation, govern through use” — is an accurate description of the system and not a euphemism. Evaluation really does build; use really does govern. The question the formula leaves open is the one this series keeps asking: who is evaluated, and is anyone being paid to answer?
Checkable questions.
(i) Does the “Trusted AI Supply Chain” assessment publish its rubric, and does it grade the upstream components an agent depends on or only the agent’s own capability? The distinction decides whether this is a supply-chain instrument or a product-marketing instrument. An AIBOM-based assessment that traces dependencies would be the first Chinese instrument in this series with a maintenance-relevant unit of account.
(ii) Will the CAICT FlagOS evaluation results be published with the underlying measurements — operator coverage, cross-chip efficiency, cost per million tokens — or only as pass/fail grades? A benchmark that publishes numbers can be contested and improved; one that publishes grades cannot. This is the cheapest available test of whether the new evaluations are designed to inform or to certify.
(iii) Does any of the four new agent assessments contain a clause about contribution to the open-source projects the agent is built on? If not — and on the published description, none does — then the AI supply chain is being assessed end-to-end for everything except the one link this series has documented as the binding constraint.
Sources: CAICT via WeChat — 开源无界 智创未来:2026 OSCAR开源产业大会在京召开 (2026-11-10; Dai Xiaohui and He Baohong remarks; trusted open-source and trusted-AI-supply-chain assessment results; six-network coordination standard system launch; AI Infra software expert group; FlagOS evaluation; Li Wei and Guo Xue presentations; speaker list) · SegmentFault — 可信开源智启未来,2025 OSCAR开源产业大会在京举行 (2025-10-28; prior-year baseline: 14 categories, 141 enterprises assessed; He Baohong on maintenance weakness and a “market-plus-government” open-source upkeep model) · CNR — 2025 OSCAR开源产业大会在京举行 (2025-10-29) · CAICT — 智能家居产业生态培育工作组工作会议 (2026-11-09; the same institute building a national open-source community inside a mandatory-standard workstream, with an “open source” sub-group) · BAAI community — 智源众智FlagOS 2.2 (2026-10-20; 1,454 operators, 10 heterogeneous chips, 2.04× throughput on deployed Ascend 910B) · This series’ 2026-11-07 briefing — MIIT Notice No. 42, agent terminal network access, and the CAICT adaptation registry
⚖️ Legal & Licensing — An Adaptor Certificate Becomes a Purchasing Channel: What the Xi’an OpenHarmony Programme Actually Built
1. The event, and the same week’s other event. On 2026-11-10, at the Xi’an National Digital Publishing Base, the OpenAtom Foundation and the Xi’an High-Tech Industrial Development Zone Administrative Committee jointly held the OpenAtom “Park Tour” (园区行) Xi’an stop, badged also as the Xi’an OpenHarmony Pioneer City Industrial Ecosystem Conference. Co-organisers: Xi’an Municipal Bureau of Industry and Information Technology, Xi’an Municipal Data Bureau, and the Xi’an High-Tech Software City development office. Hosts: the Xi’an OpenHarmony Technology Adaptation (Certification) and Product Development Centre, ChinaSoft International (中软国际), and Digital Xi’an Technology Operations Group. About 300 people attended. Three days earlier, on 2026-11-06, MIIT published Notice No. 42 on agent terminal network access, and this series’ 2026-11-07 entry described the resulting CAICT “agent version and terminal model adaptation registry”. Both are registries that a product must enter before a public buyer can buy it. This entry is about the version of that mechanism built at municipal level.
2. What was signed and expanded, stated as instruments. At the event:
- A tripartite agreement on an “one-stop” OpenHarmony industrial service platform, signed by ChinaSoft International, CAICT’s CTTL Terminal Laboratory (泰尔终端实验室) and Xi’an Fengniao Pilot-Testing Technology (西安蜂鸟中试科技). Its stated function: western-region enterprises complete compatibility testing, proof of concept and pilot-line validation “at their own doorstep”, so they “no longer have to run back and forth for a single test.”
- A second batch of governing members admitted to the Xi’an OpenHarmony Industrial Innovation Centre — six vice-chair and council units: Shaanxi Transportation Holding Technology Group, China Unicom Xi’an, Xi’an ESWIN Computing (奕斯伟), China Telecom Xi’an, China Mobile Shaanxi Xi’an, and Xi’an Huizhi Information Technology — covering transport, telecoms, chips and software.
- An expansion of the municipal certification pipeline reported from the same programme (OpenAtom, 2026-01-29): 32 key traffic devices have completed OpenHarmony adaptation and 16 have passed compatibility assessment, with more than 20 supply-chain vendors engaged and radar-vision units and signal controllers moving toward local manufacture.
3. The mechanism, in the foundation’s own words, and this is the entry’s substance. OpenAtom’s account of the Xi’an model (2026-09-02) is unusually explicit about the institution rather than the technology, and it is worth reproducing because this series has been arguing for five days that the missing piece is not a standard but an instrument that connects certification to money:
"…relying on the country’s first OpenHarmony technology adaptation (certification) and product development centre, [Xi’an] explores a complete development path of technology adaptation — industrial cultivation — urban empowerment. On the one hand it relies on the OpenHarmony distributed base to break technical obstacles to heterogeneous device coordination; on the other it emphasises institutional innovation, releases city-level real scenarios through open competition for the best solution (揭榜挂帅), and establishes a “three lists” mechanism (三张清单) that converts urban-governance goals into technical requirements, certification standards and market channels, supported by adaptation service vouchers (适配服务券) as a policy instrument to lower the cost of adaptation for small and medium enterprises, closing the commercial loop “R&D — adaptation — certification — procurement” and landing the industrial loop “local adaptation — local production — local application.”"
4. The money at the end of the chain, which is what makes the mechanism real. The Xi’an Municipal Public Security Bureau Traffic Management Detachment’s 2026–2027 smart-traffic construction project — the flagship deployment the certification pipeline feeds — was won by an OpenHarmony ecosystem consortium led by ChinaSoft International for ¥245 million, reported 2026-08-25. The project covers AI-driven coordinated dispatch, about 4,082 sensing points across the city, and five classes of OpenHarmony traffic device.
Institutional Analysis & Commentary:
A district government in China has built the one thing this series has argued cannot be built by a national foundation: an instrument that pays for work on the commons and can be audited, because it is a procurement. The chain is: three lists (what the city needs → what certificate satisfies it → what may be bought) + adaptation vouchers (who pays the SME’s adaptation cost) + a certification centre (who verifies) + a ¥245 million city procurement (who buys). Every link is a budget line in a horizontal government’s own competence, not a grant application to a ministry. That is the institutional significance, and it is larger than the project.
Why this works where the national design does not, and the reason is a property of public procurement rather than of China. Ostrom’s design principles for commons institutions begin with clearly defined boundaries and congruence between rules and local conditions; Williamson’s L3 analysis of governance predicts that a transaction whose asset specificity is high will be governed by whatever arrangement can write a complete-enough contract. A municipal traffic project is the perfect specimen: the buyer is one entity, the requirement is one document, the deliverable is verifiable at the unit, and the payment is conditioned on acceptance. Certification fits this transaction because procurement is exactly the kind of transaction certification was invented for. Compare what the same institution cannot do: pay the reviewer of a stranger’s patch. There is no buyer, no requirement document, no acceptance test, and no unit of delivery. The Xi’an mechanism is not a better version of a commons fund; it is a perfect fit between an instrument and the one transaction type that instrument actually completes.
And here is the part that should be stated in the foundation’s favour, because the record supports it. The Xi’an model is not merely “subsidy the buyer.” The three lists convert a city’s governance goals into technical requirements and certification standards before procurement — which means the standard is written against a named, real deployment, not against a laboratory. The adaptation vouchers reduce the fixed cost for the small vendor that would otherwise be priced out of certification. And the innovation centre’s membership expansion puts transport operators, three telecoms and a chip company into the governing body — which is exactly the boundary-definition work Ostrom’s first principle demands. Set against the alternative of a ministry writing a standard and then looking for someone to adopt it, this is the higher-quality institutional design, and it deserves to be reported as such. The open-source way this series defends is not a governance form; it is a property of rules — who wrote them, who can inspect them, who can exit — and on the first two of those, Xi’an is doing better than the national-level instruments it sits under.
Then the limits, stated precisely, because they are the same limits every time. First, this mechanism creates adaptors, not maintainers. 32 devices adapted and 16 certified is a genuine achievement measured in installed base; it says nothing about who reviews a patch to the OpenHarmony kernel from outside the ecosystem. Second, the whole chain terminates in one purchaser — a city traffic bureau — so its scale is bounded by the number of large public projects a given city runs, and its discontinuation risk is a budget cycle, not a community decision. Third, the certification centre is co-built by the vendor that leads the consortium that wins the contract — ChinaSoft International hosts the Xi’an centre, signs the platform agreement, and leads the ¥245 million bid — which is disclosed and normal for a public-private platform, but which means the certification is not independent of the supplier in the way a standards body’s certification is. None of these is misconduct. All three are why the model replicates as procurement policy rather than as commons governance.
The generalisation that follows, and it is testable against the Guangdong case. Xi’an is not unique: in 2026-03-24 Guangzhou’s Huangpu district opened the country’s first provincial-level OpenHarmony adaptation centre — ¥100 million total investment, positioned as “public-interest in nature, enterprise-operated” (公益性质、企业化运作), run by a district state-owned enterprise (Science City Group) leading a joint venture with Jiadu Technology, the Fifth Electronics Research Institute and Huawei — with a demonstration-zone plan whose targets are stated in firms, scenarios and industrial output: 30 firms, 10 scenarios, ¥3 billion by end-2026; 100+ firms, 50 scenarios, ¥10 billion by end-2028. Two cities, six weeks apart, same instrument, same operating form, same output metrics. When an instrument is adopted by different jurisdictions within weeks and its success is denominated in industrial output, it has stopped being a pilot and become an administrative category — which is exactly what the filings in this series’ 2026-11-10 entry showed at national level: the ¥20.7 million 2023 subsidy restricted to an “open-source project ecosystem cultivation platform”, spent on compatibility testing, licensing, training and hosting. The local centres are the delivery arm of that line item, and they are good at it.
Checkable questions.
(i) Do the “three lists” publish the certification standards, and can a vendor outside Xi’an and outside the incumbents’ network obtain a certificate without a local partner? OpenAtom’s account claims the adaptation centre “does not confine the results locally” and publishes its adaptation specifications. A published specification and an open certification queue are two different things, and the second is the one that determines whether this is a standard or a channel.
(ii) What is the total value of “adaptation service vouchers” issued, and what is the smallest firm that received one? The voucher is the cheapest instrument in the whole chain and the only one aimed at a small supplier. Its distribution says whether the mechanism lowers entry cost or subsidises firms already able to bid.
(iii) Does the Xi’an centre publish the number of outside contributions to OpenHarmony that came from firms it certified — pull requests merged upstream, not products shipped? This is the difference between an ecosystem and an installed base, and the centre is the only institution positioned to produce the number.
Sources: OpenAtom Foundation via WeChat — 开放原子"园区行"(西安站)成功举办:完善产业服务体系,推动开源鸿蒙应用落地 (2026-11-10; joint hosts; ~300 attendees; ChinaSoft–CTTL–Fengniao one-stop platform agreement; six new governing members; Cheng Xiaoming, Wang Jing, Ju Bo and Tong Qiang remarks; RISC-V dual-open-source coordination) · OpenAtom Foundation — 发挥场景示范效应,西安构建开源鸿蒙产业发展实践范式 (2026-09-02; the “technology adaptation — industrial cultivation — urban empowerment” path; 揭榜挂帅; the three lists; adaptation service vouchers; the R&D–adaptation–certification–procurement loop) · OpenAtom Foundation — 开源鸿蒙赋能西安智慧交通,解锁城市出行"新速度" (2026-01-29; 32 devices adapted, 16 passing compatibility assessment, 20+ vendors; local manufacture) · Eastmoney/ChinaSoft International — 中标2.45亿元西安智慧交通建设项目 (2026-08-25; ¥245 million consortium win) · Guangzhou Huangpu district government — 全国首个省级开源鸿蒙适配中心在广州黄埔落地 (2026-03-24; ¥100 million, “public-interest, enterprise-operated”, Science City Group + Jiadu + Fifth Electronics Institute + Huawei; 2026 and 2028 output targets) · This series’ 2026-11-10 briefing — OpenAtom’s 2023 state grant restricted to an “open-source project ecosystem cultivation platform” · This series’ 2026-11-07 briefing — MIIT Notice No. 42 and the CAICT adaptation registry
🔍 Commentary — The Instrument That Fits the Transaction
First: today’s two entries are the same finding from a ministry’s rostrum and a district’s signing table. In Beijing, an institute announced that open source’s value is migrating from collaborative production to the application ecosystem, and issued assessments for four agent products and a seven-party IoT standard. In Xi’an, a city connected three lists, a voucher, a certification centre and a ¥245 million procurement into a single loop and reported 32 adapted devices. Neither is a scandal and neither is theatre. Both are the same institution working as designed: the Chinese open-source system of 2026 is a deployment machine with a certification front end, and it is getting better at deployment every quarter.
Second, the analytical correction this week owes the previous five. This series has spent a fortnight describing a deficit — review queues unfunded, maintainers on a donor’s payroll, twelve unanswered issues — and has sometimes written as though the actors had failed to notice. The Xi’an mechanism shows the opposite: they have built, deliberately and at municipal scale, an instrument that works. The reason it does not address the deficit is not blindness. It is that the instrument answers a different transaction. Public procurement has a buyer, a specification, an acceptance test and a unit of delivery; certification attests to exactly those four things. A patch review has none of them. Every Chinese instrument in this series — the register, the credential, the identity code, the adaptation certificate, the supply-chain assessment — fits the procurement transaction, and none fits the collaboration transaction. That is a statement about the transaction, not about the people doing it, and it explains the pattern better than any account of culture or of intent.
Third, and this is where the analysis has to hold two things at once. The Xi’an loop is good institutional design, and it should be praised with the specificity it earns: standards written against a named real deployment rather than a laboratory; vouchers aimed at the fixed cost that prices small vendors out; a governing body that admits the operators, the three carriers and a chip company; a published adaptation specification that the foundation says is open to outside firms. These are the moves Ostrom’s boundary and congruence principles call for, executed by a district government. The critique that survives is not “this is fake open source.” It is that this design has a ceiling, and the ceiling is not technical. Its scale is the number of large public projects a city runs; its durability is a budget cycle; its certifier is co-housed with the winning supplier; and its output metric is industrial value. A commons built this way will have excellent adapters, a large installed base, and no one on the roster whose job is to answer a stranger.
Fourth, the sentence worth keeping, and it is a prediction about instruments rather than about politics. For six weeks this series has asked who pays the reviewer of a patch, and the answer keeps arriving in the negative from the same direction: an instrument is fundable only if it fits a transaction somebody already runs. Xi’an did not solve the maintenance problem; Xi’an found a transaction big enough to carry a certificate, and the certificate is the only part of the mechanism that travels. The prediction that follows is checkable in every city that copies this model in 2027: the adaptation centres will be built, the vouchers will be issued, the devices will be certified, and the number of people paid to review upstream code will remain where it is — because there is no procurement for it. The reform that would break this is unglamorous and, on the evidence of Xi’an, entirely within the competence of a district government: put a line in the ¥245 million contract that requires the consortium to staff a public review rota upstream. No new institution would be needed. Only a clause.
Sources: CAICT via WeChat — 2026 OSCAR开源产业大会 (2026-11-10) · OpenAtom Foundation via WeChat — 开放原子"园区行"(西安站) (2026-11-10) · OpenAtom Foundation — 发挥场景示范效应,西安构建开源鸿蒙产业发展实践范式 (2026-09-02) · Guangzhou Huangpu district government — 全国首个省级开源鸿蒙适配中心 (2026-03-24) · This series’ 2026-11-10 briefing — the filings, the restricted platform grant, and the three funding modes · This series’ 2026-11-09 briefing — the ¥450 million commitment and the register-versus-queue finding