⚠️ 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 — 2026-11-09

🏗️ Institutional Change — A Foundation That Pays Its Own Projects’ Rent: OpenAtom’s ¥450 Million and the Governance Description Around It

1. The announcement, and where it was made. On 2026-08-28, at HarmonyOS Ecosystem Conference 2026 (HEC, Shenzhen), Xie Shaofeng (谢少锋), chairman of the OpenAtom Open Source Foundation (开放原子开源基金会), delivered a keynote that the foundation published on its own site the same day. The keynote is unusually explicit for a Chinese foundation document: it makes a historical claim, reports audited-looking project metrics, names three challenges, and then commits a budget. It deserves to be read in full rather than excerpted, because the budget is the least interesting part of it.

2. The three “historical contributions” — the foundation’s own account of why it exists. Xie states them as a sequence with a causal structure, not as a list:

  • OpenHarmony “gave birth to” OpenAtom. “Huawei open-sourced the operating-system technology it had accumulated over many years, taking the technology outside the boundary of the enterprise; parties across the industrial chain began building around a unified base. To receive public technology assets and establish rules for collaborative co-construction, the OpenAtom Open Source Foundation was founded accordingly — a milestone in the history of Chinese open source.” This is the foundation describing itself, in its own words, as an institution created to hold an asset somebody else had to give up — and the operative nouns are 承接 (receive/take over) and 规则 (rules), not community, not ecosystem, not developers.
  • OpenHarmony “popularised” open source. More enterprises came to see open source as “a collaborative innovation method that gathers industrial resources and lowers costs”, and industrial-chain actors moved “from pure technology users to ecological co-builders”. In the same breath Xie draws the boundary this series has watched being drawn for months: “clarify the boundary between OpenHarmony and the commercial HarmonyOS operating system built on the OpenHarmony technology base — open-source projects have common technology settled under neutral foundation governance; commercial versions focus on market product innovation.”
  • OpenHarmony “precipitated” the foundation’s governance experience (沉淀基金会治理经验). The foundation, “facing the reality of many participating entities and a complex technical system… explored a neutral operating mechanism for guarding public assets, coordinating multiple parties’ demands, and guaranteeing community service”, and converted OpenHarmony’s practice into “reusable institutional processes”.

3. The numbers, as reported. OpenHarmony graduated in 2025; over six years it completed 11 version releases; the codebase exceeds 140 million lines; it has gathered 13,808 contributors, 644 community partners and 53 project-group donors; 2,040-plus products have completed compatibility testing; co-construction covers 23 technical domains; the pattern is described as “one set of root technology, many distributions, covering many kinds of intelligent terminals.”

4. The three challenges, stated without euphemism — because this is the part that is checkable. Xie names them:

  • “The technical base needs to move from usable to good to use, filling shortfalls in industry components, chip adaptation and device-side intelligence.”
  • “Industry expansion needs to avoid the risk of technical forking and ecosystem fragmentation, and clarify the boundary between open-source and commercial brand use.”
  • “Industrial influence needs to move from rapid domestic development to global layout, raising overseas developer participation and international reach.”

5. The commitment. Under 优化开源生态环境 (optimising the open-source ecological environment), Xie states: "Over the next three years the OpenAtom Open Source Foundation will invest 150 million yuan annually, focusing on support for root-technology evolution, shared industrial capability, development toolchains, security systems, compatibility capability and key technology." The funds are to “guide central state-owned enterprises, private enterprises and university research institutes to participate broadly”, and the foundation will “link open-source projects, industry users, investment institutions and local governments — helping technology find scenarios, users find solutions, and enterprises form revenue.”

6. The governance language around it, which is the actual subject. The keynote’s description of OpenAtom’s own machinery is worth isolating: “a layered governance architecture (分层治理架构), with multiple specialised organisations covering strategic decision-making, technical planning and version R&D across the whole process”; “the foundation centrally takes over OpenHarmony’s code, brand and other public assets on a neutral platform, and builds an open and equal co-construction and sharing mechanism”; and the target pattern "community precipitation — industry validation — product landing — results feeding back" (社区沉淀—行业验证—产品落地—成果反哺). The three management priorities are named as “strengthening open-source management, increasing project promotion and application, optimising the open-source ecological environment.”

Institutional Analysis & Commentary:

A budget is a governance instrument, and this one points in a specific direction. ¥150 million a year for three years is ¥450 million. Set against the scale of the OpenHarmony project it accompanies — 140 million lines, 13,808 contributors, 644 partners — it is not a large sum; it is roughly the annual cost of a few hundred engineers. What makes it institutionally significant is not the amount but the enumerated line items: root technology, shared industrial capability, toolchains, security, compatibility testing, key technology. Every item on that list is a capability the foundation itself operates or certifies. Not one is “paying maintainers”, “funding release engineering in downstream projects”, “supporting independent contributors”, or “review capacity”. The foundation is funding the platform and the test bench, which is the role a foundation can perform when its relationship to projects is custodial rather than convening.

That is the coherent choice for the institution OpenAtom actually is, and Xie says so plainly in the first contribution: it was created to receive an asset and write rules for it. A foundation that receives assets must maintain a register, a brand, a compatibility suite and a legal arm. This series documented on 2026-11-07 that the Beijing Kaiyuan institute operates a curation register of 300-plus industrial-software projects; the MIIT agent-identity standards of 2026-11-07 created an identity register; and Xie’s keynote describes a project register with a layered governance structure and a compatibility-testing gate of 2,040 products. The Chinese open-source institution of 2026 is, structurally, a registry operator with a budget. Registries are real institutions — they solve a genuine coordination problem, and 2,040 compatible products is not nothing — but they solve it by certification, and certification is a service delivered to producers, not a contribution made by strangers.

The 53 project-group donors are the number that describes the funding model, and it is a different model. In a Western foundation the money generally arrives as membership dues from companies that want a seat; the foundation then spends on staff, events, legal defence and sometimes on the projects. Here, money arrives as project donations from the companies that already own the projects — 53 of them for OpenHarmony alone — and the foundation spends on a platform, a test suite and a brand. The consequences differ. A membership-dues foundation has an incentive to grow the number of dues-paying members and to keep projects attractive to them; a donations-and-grants foundation has an incentive to grow the number of registered projects and the completeness of the register. Neither is dishonest. They simply produce different behaviour when a project stops being interesting to its original donor, and that difference is measurable — which is what the rest of this briefing does.

And then the sentence worth keeping from the whole keynote is the third challenge. “Avoid the risk of technical forking and ecosystem fragmentation.” In a commons, forking is the exit option — the guarantee that a governance structure cannot capture a project’s users, and the reason maintainers can refuse a bad decision knowing the code will survive them. In this keynote, fork risk appears on the cost side of the ledger, as something to be managed. That is not a contradiction — fragmentation is a real cost, and openEuler’s own history (documented here on 2026-11-05) is a case study in how expensive upstream loss can be. But a governance design in which exit is a risk rather than a right will select, over time, for instruments that make exit harder. That is the sentence to hold up against the next standard, the next register, and the next five-year plan.

Checkable questions.

(i) Will the ¥150 million figure appear in a published budget with line items and an audited annual report? A foundation that states a number in a keynote and a foundation that reports the number in a financial statement are two different institutions. OpenAtom is a registered non-profit; whether the annual report discloses the deployment of this fund is directly observable.

(ii) What proportion of the ¥150 million goes to projects the foundation did not originally receive from a single company? If the answer is “almost none”, the budget is a maintenance subsidy for donated corporate code, which is a perfectly respectable thing to be — and a very different thing from a commons fund.

(iii) Does any of it pay for review capacity in a project whose contributing organisation has left? Line items name toolchains and security. The item that would answer the binding question — who pays the reviewer of a stranger’s patch — is absent from the published list. Its absence in year one is a design choice; its absence in year three will be a finding.

Sources: OpenAtom Foundation — 开放原子开源基金会理事长谢少锋出席鸿蒙生态大会2026 (2026-08-28): three historical contributions; OpenHarmony 140 million-plus lines, 13,808 contributors, 644 community partners, 53 project-group donors, 2,040-plus compatible products, 23 domains, 11 releases, graduated 2025; three challenges including fork and fragmentation risk; ¥150 million per year for three years; layered governance architecture and the community precipitation → industry validation → product landing → results feedback pattern · OpenAtom Foundation — homepage and governance description · This series’ 2026-11-07 briefing — the Beijing Kaiyuan curation register and MIIT’s agent-identity register · This series’ 2026-11-05 briefing — openEuler’s upstream history and the cost of upstream loss


🔬 Field Note — Twelve Thousand Stars, Zero Replies: Measuring a Foundation-Held Project From Inside

1. Why this project. OpenAtomFoundation/pikiwidb is the clearest available specimen of the thing the previous entry describes: a project donated to OpenAtom and hosted under the foundation’s GitHub organisation. Originally Qihoo 360’s Pika, it is a persistent, RocksDB-backed, Redis-protocol-compatible KV store — a real infrastructure project with real production use, not a demonstration. Every number below was pulled from GitHub’s public API on 2026-11-09 and is reproducible.

2. The repository, as measured.

  • 6,136 stars, 1,170 forks, 265 watchers, 89 open issues (which includes open pull requests). Created 2014-11-03; last push 2026-10-16.
  • Licence: BSD-3-Clause, and the LICENSE.md file still reads “Copyright (c) 2015-2024, Qihoo360. All rights reserved.” — the copyright was not reassigned on donation, which is normal and correct for a BSD-licensed donation, and which also means the foundation holds governance, not ownership.
  • README still links Qihoo360/pika for its build badge, downloads badge, star chart and contributor graph; the repository directory is pikiwidb but the star-chart URL still points at OpenAtomFoundation/pika. The documentation has not fully left the donor’s house.

3. The contribution surface, measured. GitHub’s search API reports 1,716 pull requests and 1,379 issues in the repository’s history. Of 800 pull requests enumerated (pages 1–8, newest first):

  • 584 merged — a 73 percent merge rate, which is high and is evidence that the project’s maintainers do process contributions.
  • Creation dates 2023-07-26 to 2026-10-12, with a steep recent decline: 14 opened in 2026-04, 3 in 2026-07, 1 in 2026-10, none since 2026-10-12.
  • The top eight pull-request authors account for 540 of 800 — chejinge (190), Mixficsol (161), wangshao1 (51), cheniujh (50), chengyu-l (31), baerwang (25), luky116 (22), chenbt-hz (20). This is a small, familiar group, which is what a healthy five-person team looks like and also what a closed team looks like. The data cannot distinguish those two cases; only the organisation can.

4. Commits in 2026, measured. 31 commits since 2026-01-01, from five named authors:

  • 22 from @360.cn addresses, of which 20 are one author, wuxianrong;
  • 4 from @qq.com;
  • 5 from users.noreply.github.com, of which 3 are authored Copilot — i.e. agent-generated commits landed with the agent as commit author;
  • the newest commit is 2026-10-16T03:35:06Z; the project has not moved in 24 days.

5. Issues in October, measured, and this is the finding. Twelve issues were opened between 2026-10-01 and 2026-10-28. Zero have comments. Zero are closed. They are substantive, well-formed bug reports from clearly identifiable outside accounts — jiren-the-gray filed seven of them in a two-day window (GEORADIUS/GEORADIUSBYMEMBER with COUNT above INT64_MAX; XINFO argument out-of-bounds; stream keys appearing nonexistent to string commands; PFADD on an existing HyperLogLog replying 1; SMOVE to a wrong-type destination; LRANGE with INT64_MAX returning empty after LPOP; INCRBY on an out-of-range value), plus hanke580 on CONFIG REWRITE replacing pika.conf without fsync, objectboy2016 asking in Chinese about Stream consumer-group XGROUP support, and quiet-maple-47 filing two. Maintainer response time on outside issues, in this sample, is: not yet measurable, because there has been no response.

6. The maintainer list, which is the indicator that decides the question. MAINTAINERS.md on the unstable branch defines three roles and names three maintainers — wangshao1, Mixficsol, chejinge — with an Organization column. Two cells are an em-dash. The third reads 360, with the email chejinge@360.cn. A committers table adds five more names; a notable-contributors table adds nineteen.

Institutional Analysis & Commentary:

The donation moved the governance and left the labour where it was, and this repository lets one watch the moment those two things separate.

The project was created in 2014 by Qihoo 360, donated to OpenAtom, and re-hosted under the foundation’s organisation. What the donation acquired is precisely what Xie Shaofeng’s keynote says a foundation exists to acquire: an asset, a brand, a place in a register, and rules. What it did not acquire — because a donation transfers projects, not people — is the review capacity that the donor’s payroll was quietly providing. In 2026 that capacity has narrowed to one person at one company address for the bulk of commits (20 of 31), and a maintainer list where the only disclosed employer is the original donor. The project is not abandoned; it is staffed at the level one company’s remaining interest supports, which is exactly the post-donation equilibrium this series described on 2026-10-26 for Alibaba’s donated projects.

The mechanism is worth naming precisely, because it is not neglect and it is not malice. Consider the incentives of a maintainer at 360 whose job is to keep a KV store running for internal customers. They have review capacity for their own team’s work and no budget line for reviewing a stranger’s patch. An outside issue reporting a genuine bug in XINFO is, to them, work with no internal customer — and the institution that holds the project has a registry function (the project is registered, governed, brand-held) but not a service-level function (nobody’s performance review depends on median time-to-first-response). Twelve unanswered issues in one month is what that incentive structure produces. The commons was donated; the queue was not.

Now the fair counter-observations, which the documents require. First, 584 merged pull requests is real — this project has absorbed a great deal of outside work, far more than the Ascend adapter examined on 2026-11-08 (36 pull requests in its entire history). Second, the issue volume is new: seven filings from one account in two days is either a diligent security reviewer or an LLM-assisted bug-hunting sweep, and the report quality — precise Redis-protocol divergences with INT64_MAX boundaries — is consistent with an automated audit that found real defects. If so, the twelve unanswered issues are not a community failing to be heard; they are a maintainer team facing a machine-generated defect queue with no capacity allocated to it. Both readings point to the same institutional fact, and the second makes it sharper: the marginal cost of finding defects has collapsed; the marginal cost of triaging them has not. A foundation with ¥150 million a year and a project register has an obvious thing to buy that it does not currently list.

Third, and this cuts against the series’ own framing: a mirror problem does not apply here. The unstable branch carries CODE_OF_CONDUCT.md, CONTRIBUTING.md in two languages, SECURITY.md, a three-tier MAINTAINERS.md, issue templates, dependabot and release-drafter workflows, and clang-format/clang-tidy configuration. Somebody built the commons furniture properly. The scaffolding of an open project is fully present and the labour to operate it is not — which is a much more interesting finding than “the project is dead”, and a much more common one than either the triumphalist or the dismissive accounts of Chinese open source predict.

Checkable questions.

(i) Does the donation agreement obligate the donor to keep staffing maintenance, and for how long? This is the single document that would settle whether what is being observed is a transition (the donor winds down over an agreed period) or an equilibrium (the donor’s interest ran out and nobody noticed). OpenAtom’s project agreements are not public; whether they can be made public is itself the institutional question.

(ii) Do the twelve October issues get answered, and by whom — a 360 address, a foundation address, or a non-360 contributor? Three possible triagers, three different institutional meanings. The foundation has now stated a ¥150 million annual budget that explicitly includes “security systems”; a project under its organisation with a machine-found defect queue and no triager is the cheapest possible test of whether that line item reaches the code.

(iii) Is there a second project in OpenAtom’s organisation in the same state — and how many are there in total? The GitHub organisation carries 8 public repositories: pikiwidb (last push 2026-10-16), pikiwidb-raft (2024-10-31), xupercore (2024-05-21), TobudOS (2024-02-02, with no licence detected), SmartIDE (2023-10-23), the documentation site (2026-08-20), a legal-licence translation repository (2025-05-08) and a profile repository. Four of the eight have not been pushed to in more than a year. The foundation reports 77 projects through TOC review and 59 in incubation; the public forge under its own name shows eight. Where the other 69 live is a question with a factual answer, and it determines whether “hosted by OpenAtom” describes a governance relationship or a registry entry.

Sources: GitHub API — OpenAtomFoundation/pikiwidb repository metadata (6,136 stars, 1,170 forks, 265 watchers, 89 open issues, created 2014-11-03, pushed 2026-10-16, BSD-3-Clause) · GitHub — pikiwidb pull requests (1,716 total; 800 enumerated: 584 merged, 14 opened in 2026-04, 3 in 2026-07, 1 in 2026-10) · GitHub — pikiwidb issues (1,379 total; 12 opened 2026-10-01 → 2026-10-28, none commented, none closed) · MAINTAINERS.md — three maintainers, one with a disclosed organisation (360), plus eight committers and nineteen notable contributors · LICENSE.md — BSD-3-Clause, “Copyright (c) 2015-2024, Qihoo360” · README.md — badges and star chart still pointing at Qihoo360/pika · GitHub — OpenAtomFoundation organisation repositories (8 public repositories; last push dates as stated) · This series’ 2026-10-26 briefing — the survival rule for company-donated projects and the CODEOWNERS indicator · This series’ 2026-11-08 field note — Ascend/pytorch, 36 pull requests in its entire history


🔍 Commentary — The Register and the Queue

First: today’s two entries are one sentence read from both ends. A foundation states it will spend ¥450 million over three years on toolchains, security, compatibility and root technology, and describes itself as the neutral holder of an asset — 140 million lines, 13,808 contributors, 2,040 tested products. Under that same foundation sits a project with 12,000 stars, 584 merged pull requests, twelve unanswered October issues, and a maintainer list whose only disclosed employer is the company that donated it. The foundation has a budget for the bench and no line for the queue.

Second: this is not a Chinese failure and the distinction matters more than the finding. The Linux Foundation’s own October 2026 research reports that an estimated 25 percent of incoming pull requests are AI-generated low-quality submissions rising to 31 percent at large projects, that 64 percent of maintainers at projects with more than 50 contributors say AI-assisted contributions have lengthened review cycles, and that only 35 percent of projects have a policy on AI-assisted contributions. Every maintainer anywhere now faces a collapsing cost of generating patches and an unchanged cost of reviewing them. What the PikiwiDB sample adds is not a defect unique to this ecosystem but a place where the generic problem lands on an institution with an unusual capacity to solve it and no stated intention to. A foundation that owns a registry and a test suite, funded at ¥150 million a year, is one of the few institutions on earth that could simply pay for triage — and the published line items do not include it.

Third, and this is the correction this series owes itself. In the 2026-11-08 entry it argued that the state’s instruments were building a supply of certified contributors faster than any project had built a queue to receive them, and that certification was not the scarce factor. The PikiwiDB data supports that argument in an unexpected direction: the queue is not scarce for lack of contributors — 1,716 pull requests arrived — but because review capacity is attached to a payroll rather than to the project. Certification regimes fix the wrong side of that ledger. The missing instrument is unglamorous and entirely fundable: a service level, a named reviewer, and a person whose job it is to answer. Xie Shaofeng’s third challenge names the risk he is worried about as fragmentation. The risk the repository actually exhibits is quieter: a project that is registered, governed, branded, starred — and not answered.

Fourth, the sentence worth keeping, and it is a question the register cannot answer. A foundation can hold an asset, a brand, a compatibility suite and a budget. It cannot hold attention, because attention is not transferable — it belongs to whoever is being paid on Tuesday. The test of a donation is not what the foundation received but who shows up for the stranger. On the evidence available for the week of 2026-10-01, under a foundation that had just committed ¥450 million, nobody showed up for twelve people — and the cheapest reform available to the institution is not another standard, but a shift rota.