0
Subsystems
Following the 1994 tax-sharing reform, the State Taxation Administration (STA) of China began developing the "China Tax Administration Information System" (CTAIS) in 1996. Based on specifications such as the Tax Collection and Administration Operational Procedures and the Municipal-level Collection and Administration Business Requirements, its goal was to support a modern collection and administration model "built on tax filing and optimized services, underpinned by computer networks."
CTAIS comprises 7 subsystems (1 system maintenance + 6 business subsystems), covering grassroots collection and administration plus municipal-level management and monitoring across five major series — management services, collection monitoring, tax inspection, tax legal affairs, and tax enforcement. It is a large-scale application serving tax authorities nationwide at both grassroots and municipal levels.
Rolled out nationwide from 2000. Taking Wuhan as an example, in 2001 it adopted a "fully centralized" data distribution model to process citywide collection and administration data centrally, and passed acceptance review by STA leadership and an International Monetary Fund (IMF) expert panel, earning high praise.
Serving tax authorities nationwide and hundreds of millions of taxpayers, the system demands extremely high requirements for high-concurrency transaction processing, data consistency, and availability — a quintessential "must-not-fail" core systems scenario.
The process coupling, permission systems, and inspection linkages across the five major business series require the architecture to distill complex business rules into maintainable system structures — not scattered if-else logic in code.
The evolution from CTAIS to the unified architecture of Golden Tax Phase III — "one platform, two-tier processing, three coverages, four system categories" — is a textbook case of how large-scale systems achieve versioning and smooth migration.
Participating in the R&D of a system at this scale left me with three things I still draw on today: a reverence for high concurrency (every design must account for peak load and failure scenarios), an intuition for data consistency, and the ability to collaborate on engineering within a massive system.