Mechanism

How a Flag Becomes a Grassroots Task: The Departmental Chain Behind IJOP

Traces IJOP from government direction and data intake to platform flags, grassroots checks, and human disposition, while marking the limits of public evidence.

# How a Flag Becomes a Grassroots Task: The Departmental Chain Behind IJOP

The Xinjiang Integrated Joint Operations Platform is usually known as IJOP. Describing it simply as artificial intelligence that automatically decides who is dangerous misses the institutional process that gives the system its practical force. Public materials can establish a chain in which data enters from different departments, the platform produces leads or flags, grassroots personnel receive investigation tasks, and the results of checks return to the system before people and government organs decide what to do. The public record does not show in full how the backend assigns weight to each item of information or how complete lists are generated. ^[5]

Autonomous-Region Training and Local Project Construction

The platform had entered work arrangements at the autonomous-region level by at least 2016. An Xinjiang government gazette records that, on December 25, 2016, the region's vice chair hosted a training meeting for an integrated joint operations platform supported by “big data.” The record confirms only an autonomous-region-level platform training meeting. It does not disclose the backend rules or explain why particular people were flagged. ^[1]

The construction project appears in a public report by a contractor. Leon Technology's 2017 annual board report said that the company was carrying out monitoring projects in Kashgar and Aksu and had undertaken construction of the Kashgar integrated joint operations platform. Together with the gazette, this shows only a limited construction chain: there is a record of platform training at the autonomous-region level and a record of a technology company undertaking the Kashgar project. A corporate report can establish participation in construction. It cannot by itself show that the company independently decided the platform's purpose or the outcome of individual cases. ^[2]

How the Mechanism Works: Data Intake and Grassroots Tasks

One directly verifiable data entry point came from used-car transactions. Materials from Xinjiang's commerce authorities in 2019 said that the used-car trading system shared information with the political-legal commission platform and IJOP. People, identity documents, vehicles, and images involved in transactions were compared in real time and uploaded to the public-security traffic-management authorities. What can be confirmed here is one specific industry interface. It cannot be expanded into a complete list of all IJOP data sources. ^[4]

The other end of the chain was the community and the grid. A plan for Sayibak District placed community IJOP, convenience police stations, and grid management within a normalized counterterrorism and stability-maintenance system, linked to rapid-disposal arrangements. Industry departments supplied data. Communities, police stations, and grid workers were responsible for discovery, checking, or handling. IJOP was therefore not an isolated piece of software but a work interface embedded in a local governance network. ^[3]

Human Rights Watch's reverse engineering of the 2017 IJOP mobile application filled in part of the operating layer visible to grassroots workers. The application could collect personal information, report people or vehicles, search by conditions such as name, identity-card number, or address, and receive “investigation tasks” whose results could then be sent back. Tasks included fields for people, devices, locations, and relationships. This indicates that the platform did more than store records. It organized leads into checks that personnel could carry out. ^[5]

That finding also has a clear boundary. The researchers had no username and password, did not log into the IJOP app, and did not connect to IJOP servers to obtain data to populate the app; their analysis was based on the mobile application and its source code. Its fields can show what workers could collect and what tasks they could receive, but they cannot reveal the complete backend algorithm, the actual lists, every data source, or each level of permission. Public evidence supports a process of platform flagging followed by human investigation. It does not support interpreting every result as a final decision made automatically by a machine. ^[5]

The Key Facts: From Lists and Checks to Disposition

Four internal bulletins from 2017 published by the International Consortium of Investigative Journalists moved the visible process beyond the mobile application. They described IJOP generating lists, workers conducting investigations, and people then being detained or removed from the lists according to the results. Bulletin No. 14 recorded that 24,412 people were marked as “suspicious” within one week. Of them, 15,683 were sent to vocational education and training centers, while another 706 were formally arrested. These figures come from leaked documents examined by experts. They are not public judicial statistics and cannot replace the judicial record of each individual case. ^[6]

The United Nations Office of the High Commissioner for Human Rights used a different level of evidence in its 2022 assessment. It said that documents in the public domain appeared to detail IJOP aggregating diverse data and flagging people based on specific behaviors and indicators. Taken together, it said, these materials suggested that certain behaviors could be automatically monitored and flagged to law enforcement for police follow-up, including potential referral to a vocational education and training center or another detention facility. The assessment examined this operation within a differentiated surveillance environment directed at Uyghurs and other Muslim groups. This is a factual and legal assessment made by a UN body on the available materials. It is not equivalent to a conclusion accepted by the Chinese government. ^[7]

The Consequences It Produced: People Still Decided After Software Errors

An Associated Press investigation in 2025, based on leaked emails, internal documents, procurement records, and interviews, described IJOP using relationship analysis and risk classification. It said that engineers had repaired a software problem that caused hundreds of people to be incorrectly listed as high risk. The cases presented in the investigation show how software classification could directly expand the scope of grassroots checks. Once an incorrect list entered the process, however, it still required judgment and execution by workers and government organs. ^[8]

This is why responsibility cannot be assigned only to a piece of code. Contractors built and maintained the system. The political-legal and public-security systems organized data and tasks. Industry departments supplied information. Community workers, police-station personnel, and grid workers conducted checks. The authorized organs carried out the final disposition. Public materials can piece together these stages, but they are not sufficient to attribute every flag to one superior official, one company, or one algorithm. ^[2][3]

How the Chinese Government Explains the Relevant Policies

The Chinese government's public explanation uses a framework of counterterrorism, deradicalization, and vocational education. A 2019 white paper described vocational skills education and training as counterterrorism and deradicalization measures carried out according to law. It also emphasized vocational skills and education in the national common language. This explanation answers how the government defines the goals and legal basis of the training work. It does not disclose IJOP's flagging rules, error rate, or procedures for reviewing lists. ^[9]

After the UN assessment was released, China's Ministry of Foreign Affairs fully rejected it at a press conference on September 21, 2022, calling it illegal and invalid and saying it had been assembled from false information. The response clearly stated the Chinese government's position on the assessment's legality, objectivity, and credibility. It did not provide IJOP rules, individual lists, correction results, or appeal records that outside parties could check. The government's rationale and the UN assessment therefore remain part of a public dispute. One side's judgment cannot be written as an admitted fact of the other side. ^[10][7]

Responsibility and the Evidence Boundary: The Route for Correction Remains Unclear

Existing evidence can reconstruct a limited but coherent sequence. There is a record of platform training at the autonomous-region level and a record of a technology company undertaking the Kashgar project. The political-legal and public-security systems organized data and tasks. Industry departments and grassroots units supplied information or conducted checks. After investigation, flags entered classification and disposition. The used-car interface, mobile-app fields, internal bulletins, and district-level planning each illuminate a different part of this chain. They cannot substitute for one another. ^[1][2][4][6]

The hardest part to find in public materials is how people learned why they had been flagged, how they could request correction of inaccurate information, and who reviewed an appeal. The existence of human decisions after automated flagging does not mean that a reviewable, correctable, and accountable remedy procedure existed. The software error described by the Associated Press makes this gap more concrete. Public evidence still cannot establish an error rate or confirm that every affected person experienced the same disposition. ^[8]

Sources

Government and corporate records include the Xinjiang government gazette, Leon Technology's annual report, the Sayibak District plan, materials from Xinjiang's commerce authorities, the vocational education and training white paper, and the Ministry of Foreign Affairs press conference. Technical and investigative materials include Human Rights Watch's reverse engineering of the application, the internal bulletins published by ICIJ, the Associated Press investigation, and the assessment by the UN Office of the High Commissioner for Human Rights. Each type of material is used in the main text according to its evidentiary level. ^{[1][2][3][4][5][6][7][8][9][10]}

Sources

1. Xinjiang Government Gazette, 2017 Issue 1primary-recordUnchecked
2. Leon Technology 2017 Board Work Reportprimary-recordUnchecked
3. Sayibak District 14th Five-Year Plan and 2035 Objectivesprimary-recordUnchecked
4. Measures for Orderly Development of Xinjiang's Automobile Sectorprimary-recordUnchecked
5. China's Algorithms of Repression: Reverse Engineering a Xinjiang Police Apptechnical-researchLive

   View captured page
6. China Cables: Operating Manuals for Mass Internment and Arrest by Algorithminvestigative-reportingUnchecked
7. OHCHR Assessment of Human Rights Concerns in Xinjiangofficial-findingUnchecked
8. How Silicon Valley Enabled China's Digital Police Stateinvestigative-reportingUnchecked
9. White Paper on Vocational Education and Training in Xinjiangprimary-recordUnavailable
10. Foreign Ministry Spokesperson Wang Wenbin's Regular Press Conference on September 21, 2022primary-recordUnchecked

×