---
title: "When You Cannot Compare the Code: The SPC's RMB 166 Million Centrifuge Case and How to Prove Data Misappropriation Without the Defendant's Data"
author: "DCC Editorial"
published: 2026-08-07T06:00:00.000Z
url: https://datacompliancechina.com/posts/spc-centrifuge-case-proving-data-misappropriation/
description: "AnJie Broad partner Li Yanying reads (2022)最高法知民终1592号, the 'centrifuge case' — selected as one of the SPC's 2025 typical anti-unfair-competition cases on 8 September 2025 — for what it teaches about the burden of proof when the misappropriated asset is data. The plaintiffs never obtained the defendants' software or dataset, never decompiled anything, and never ran a substantial-similarity comparison. Instead they exploited the input-output character of software: enter the claimed impeller basic-stage data into their own program along with performance figures taken from the defendants' published product literature, and the output matched. Under AUCL Article 39(2)(i) that was enough to shift the burden, and the defendants' refusal to permit on-site inspection or produce design files converted into an adverse inference. DCC adds the correction the commentary omits: the SPC did not protect the software as a trade secret at all — the plaintiffs failed to delimit its secret points, so it was rejected on the not-generally-known element, and only the dataset was protected. The software was the measuring instrument, not the asset. For overseas counsel holding datasets in China, this is the enforcement route that actually carries damages, and the case is a lesson in how to delimit a claim before you file."
tags: ["trade-secrets", "technical-secret", "burden-of-proof", "judicial", "case", "supreme-peoples-court", "anti-unfair-competition", "data-assets", "source-code", "punitive-damages", "employee-departure", "商业秘密"]
laws_cited: ["anti-unfair-competition-law", "trade-secret-protection-provisions"]
domains: ["data-economy", "data-security"]
account: "anjie-broad"
original_title: "视点丨涉软件数据商业秘密纠纷的举证责任分配规则解析——最高院案例解读系列⑨"
original_author: "李彦莹 (Li Yanying), Partner, AnJie Broad"
original_publication: "安杰世泽律师事务所 (AnJie Broad) WeChat Official Account"
original_url: "https://mp.weixin.qq.com/s/qGBYslBSaqs2cssSUTaPag"
source_language: "zh"
---

> **Source: Data Compliance China** — https://datacompliancechina.com/posts/spc-centrifuge-case-proving-data-misappropriation/ · China data law, translated and annotated for overseas counsel. Cite as: Data Compliance China, "When You Cannot Compare the Code: The SPC's RMB 166 Million Centrifuge Case and How to Prove Data Misappropriation Without the Defendant's Data", https://datacompliancechina.com/posts/spc-centrifuge-case-proving-data-misappropriation/
> *Editor's Note — DCC.*
>
> This is the ninth installment of AnJie Broad's running series on Supreme
> People's Court technical-secret judgments, written by partner **李彦莹 (Li
> Yanying)**. It reads **(2022)最高法知民终1592号** — the "centrifuge case"
> (离心机案) — for one question: how does a plaintiff discharge its burden of
> proof when the thing taken is a dataset and the defendant will not open its
> systems?
>
> DCC tracks this series because trade secrecy, not data-property-rights
> registration, is still the only route in Chinese law that gives a company an
> enforceable, damages-bearing claim over data it holds. A registration
> certificate is evidence; a trade-secret judgment is money. This one was
> **RMB 166.1 million**.
>
> Two additions of our own. First, a correction: the commentary treats the
> software and the dataset together as "the technical secrets in the case,"
> but the SPC **did not** protect the software as a trade secret — the
> plaintiffs failed to delimit what was secret about it, and that part of the
> claim was rejected. Getting this backwards would teach exactly the wrong
> lesson about how to plead. Second, the commentary does not touch how the
> RMB 166 million was actually built; we set the calculation out below,
> because the multiplier structure is the part that makes the case worth
> reading twice.

## The case

Two Shenyang industrial groups — anonymized in the judgment as **某甲集团** and
**某乙公司** — design and manufacture centrifugal compressors. Over years of
testing they accumulated **impeller model basic-stage data** (叶轮模型基本级数据),
the parameter set that drives compressor selection, together with the in-house
software that consumes it.

Three individuals — 孙某良, 印某洋, 吴某坡 — left the plaintiffs' employment,
then progressively established and controlled two competing Shenyang companies.
The SPC found they had spent more than a decade covertly running competitors
built on their former employer's technical secrets. The court ordered the
infringement stopped, required the defendants to issue a written **undertaking
of non-infringement** (不侵权承诺书), and awarded:

| Component | Amount (RMB) |
|---|---|
| Economic loss | 164,647,802 |
| Reasonable expenses of enforcement | 1,500,000 |
| **Total** | **166,147,802** |

On **8 September 2025** the SPC published the judgment as one of the year's
**typical anti-unfair-competition cases** (2025年人民法院反不正当竞争典型案例),
listing it second and describing it as covert establishment of competing
companies plus a decade-plus misappropriation of a former employer's technical
secrets.

## What the court actually protected — and what it refused

This is where DCC parts company with the commentary, and the distinction is
practical rather than pedantic.

The plaintiffs asserted two things as trade secrets: the **basic-stage data**
and the **software**. The SPC treated them differently.

**The dataset was protected.** The court found the basic-stage data had been
formed through long-term research, experiment and accumulation, carried
commercial value, was not generally known or readily obtainable by relevant
persons in the field at the time of the alleged infringement, and had been
covered by the plaintiffs' confidentiality management system, employee handbook
and employee confidentiality agreements. All three elements — 三性 — held.

**The software was not.** The plaintiffs asserted only the executable, never
produced the source code, and described the program in general or functional
terms. Because they gave no clear and specific delimitation of what, precisely,
was secret about the optimization and iteration methods, the court held it
could not find that the asserted content was not generally known. That part of
the claim failed on the merits.

The commentary's account of the secrecy analysis — that the defendants raised
no substantive objection to commercial value or non-public status, so the SPC
examined only whether confidentiality measures had been taken — describes the
court's treatment of the **dataset**. It does not describe the software, which
foundered on a different element entirely.

The consequence is the single most useful sentence a China litigator can carry
out of this case: **the software was not the asset. It was the measuring
instrument.** The plaintiffs won because they had delimited one thing precisely
— a dataset — and could demonstrate its use. They lost the other because they
gestured at it.

On the confidentiality element the SPC was, by contrast, generous. It held that
measures need not be "airtight and foolproof" (严丝合缝、万无一失); measures
ordinarily sufficient to prevent disclosure will do, assessed against the nature
of the secret and its carrier, its commercial value, how identifiable the
measures are, how well they correspond to the secret, and the holder's evident
intent to keep it. A plaintiff's confidentiality evidence need not map onto the
asserted secret with perfect precision so long as it makes the scope and the
intent recognizable.

## The burden-shifting architecture

**Article 39 of the Anti-Unfair Competition Law (2025)** — effective 15 October
2025, and carrying forward Article 32 of the 2019 version, which is what the
court actually applied — restructures who must prove what:

> In civil trial procedures for infringement of trade secrets, where the trade
> secret holder provides prima facie evidence that it has taken confidentiality
> measures with respect to the claimed trade secret, and reasonably indicates
> that the trade secret has been infringed, the suspected infringer shall prove
> that the trade secret claimed by the holder does not fall within the trade
> secrets provided for in this Law.
>
> Where the trade secret holder provides prima facie evidence reasonably
> indicating that the trade secret has been infringed, and provides one of the
> following types of evidence, the suspected infringer shall prove that it has
> not committed an infringement of the trade secret:
>
> (i) evidence indicating that the suspected infringer had a channel or
> opportunity to obtain the trade secret, and that the information it uses is
> substantially identical to that trade secret;
>
> (ii) evidence indicating that the trade secret has been disclosed or used by
> the suspected infringer, or is at risk of being disclosed or used;
>
> (iii) other evidence indicating that the trade secret has been infringed by
> the suspected infringer.

Two separate shifts operate here, and it is worth keeping them apart.

**On the right itself**, the holder need not prove all three elements. It
proves confidentiality measures, and the burden of *negating* the other two
moves to the defendant. A defendant that does not affirmatively attack
commercial value and non-public status will find them established by default —
which is exactly what happened to the dataset in this case.

**On the infringing act**, the holder need not produce direct evidence that the
defendant did the thing. It reasonably indicates infringement, satisfies one of
the three limbs, and the defendant must then show a **lawful source** for the
information it uses and give a reasonable explanation of the technical
correspondence. Failure to carry that rebuttal lets the court find infringement
established.

## Reverse derivation: proving use without the defendant's data

The evidentiary position was, on paper, hopeless. The plaintiffs held no source
code comparison. Neither instance conducted a substantial-similarity review. A
first-instance **lawyer's investigation order** sent plaintiffs' counsel to the
defendants' customers, who refused to produce anything, citing confidentiality
clauses. On appeal the SPC retrieved a criminal-referral document from the
former Shenyang industry and commerce bureau and a case-dismissal decision from
the public security bureau, both recording that parties in a related case had
obtained the plaintiffs' compressor drawings and software by improper means —
but neither pointed at the technical information actually used during the
period complained of. The defendants had also licensed a functionally
comparable American program in November 2016, giving them a ready alternative
explanation.

So the plaintiffs stopped trying to compare artifacts and worked the software's
**input-output behavior** instead.

The rights software only produces output once specific basic-stage data has
been loaded, and the impeller designations in the plaintiffs' dataset followed a
naming pattern that corresponded recognizably to the impeller designations on
the defendants' products. So:

1. Take an impeller designation from a defendant's product.
2. Map it to the corresponding basic-stage data record in the claimed dataset.
3. Load that record into the plaintiffs' own software, together with specified
   initial conditions and some of the performance figures printed in the
   defendants' own product literature.
4. Compare the software's output against the remaining performance figures in
   that same literature.

Across eight centrifugal compressors, the outputs came back identical or highly
similar to the defendants' published performance data, once the inevitable
margin of reasonable manual intervention in product design was excluded. The
SPC held this sufficient to preliminarily establish that the defendants had
used the plaintiffs' basic-stage data in impeller selection design — discharging
the plaintiffs' burden under Article 32(2)(i), now Article 39(2)(i).

Nothing here required access to the defendants' software or dataset. The
evidence came from the plaintiffs' own program and the defendants' sales
literature.

## Obstruction, and how the number was built

The defendants then declined to cooperate with on-site inspection, declined to
produce research and design materials, and offered no reasonable explanation of
the reverse-derivation results. Under Chinese civil evidence doctrine that is
**obstruction of proof** (举证妨碍): a party who controls the evidence and
withholds it invites an adverse inference. Having failed to prove it had not
infringed, the defendants lost.

The damages arithmetic, which the commentary skips, is the second lesson. Per
the SPC IP Tribunal's own case page, the award was built from the defendants'
profits over 2013–2021 of approximately RMB 309.55 million, multiplied by a
**technical contribution rate** of 30% to isolate the portion attributable to
the misappropriated secret, with a **punitive multiplier of two** applied to
the conduct occurring after April 2019 — when the 2019 AUCL amendment made
punitive damages available for trade-secret infringement.

That structure is worth internalizing. The contribution rate cuts the base
down; the punitive multiplier drives it back up; and because the infringement
ran for more than a decade, the profit base was enormous to begin with. A
long-running, quiet misappropriation is the most expensive kind.

## What this means for data held in China

DCC has covered the registration side of enterprise data rights at length —
including [the first Chinese ruling on the evidentiary weight of a data-IP
registration certificate](/posts/datatang-v-yinmu-data-ip-registration-case/),
which held that such a certificate is prima facie evidence of a property-type
interest and lawful sourcing, but not an absolute property right. Registration
helps you prove you had something. It does not, by itself, give you a cause of
action with a damages engine behind it.

The Anti-Unfair Competition Law now offers two distinct routes, and they suit
different assets. **Article 13**, the data clause added in the 2025 revision,
protects data an operator *lawfully holds* against improper acquisition or use —
the theory the Beijing Internet Court applied in
[the first published data-clause judgment](/posts/aucl-data-clause-first-case-platform-scraping/),
where the decisive question was whether the platform lawfully held the scraped
career data. That route does not require secrecy, which is why it reaches
scraping of data users can see. **Trade secrecy under Articles 10 and 39** is the
opposite trade: it demands that the data never became generally known, and pays
for that constraint with a much stronger evidentiary position and punitive
damages. Data published to users takes the first route. Data held internally —
training sets, parameter tables, engineering data — takes the second, and the
centrifuge case is what the second looks like when it works.

Trade secrecy also has a recently widened perimeter. SAMR's
[Provisions on the Protection of Trade Secrets](/laws/trade-secret-protection-provisions/)
(Order No. 126, effective 1 June 2026) fold algorithms, data and source code
expressly into protectable technical information, and recognize tiered access,
data masking and audit-log retention as adequate confidentiality measures. Read
alongside this judgment, the practical guidance for a company holding valuable
datasets in China comes to four points.

**Delimit the secret before you need to.** The plaintiffs lost their software
claim on delimitation alone. Identify the specific secret points — which tables,
which fields, which parameters, which functions — and record them while nothing
is in dispute. A functional description of a system is not a claim.

**Confidentiality measures are judged reasonably, not perfectly.** Management
systems, handbooks and signed employee confidentiality agreements carried the
element here. The standard is what suffices in the ordinary case to prevent
disclosure, and the newer SAMR Provisions confirm that access tiering and audit
logging count. Departing-employee exit procedures are the weak point in most
foreign-invested subsidiaries.

**Design datasets so that use leaves a signature.** The plaintiffs prevailed
because their impeller designations corresponded recognizably to output that
appeared in the defendants' own marketing. That was partly luck and partly
structure. Deliberate internal conventions — designation schemes, seeded
records, derived values that propagate into downstream output — turn an
unprovable claim into a provable one. This is a data-architecture decision with
litigation consequences, and it should be made by people who know both.

**Non-cooperation is not a defense.** Refusing inspection and withholding design
files did not create doubt in the defendants' favor; it supplied the inference
that finished them.

For overseas counsel the summary is short. If a Chinese counterparty, joint
venture partner or departing team takes your data, the AUCL route is available,
the burden of proof tilts toward you once you clear a modest threshold, and the
damages are real. Whether you can use it is decided years earlier, by how
precisely you defined what you owned.

---

**Source:** 李彦莹 (Li Yanying), 《视点丨涉软件数据商业秘密纠纷的举证责任分配规则解析——最高院案例解读系列⑨》,
安杰世泽律师事务所 WeChat Official Account, 7 August 2026 —
[original](https://mp.weixin.qq.com/s/qGBYslBSaqs2cssSUTaPag). Case details,
the split holding on software versus dataset, and the damages calculation are
taken from the Supreme People's Court Intellectual Property Tribunal's own
case page for (2022)最高法知民终1592号 and the SPC's 8 September 2025 release of
the 2025 typical anti-unfair-competition cases. The caseload figures the
original cites (343 technical-secret substantive cases accepted and 334
concluded since the Tribunal's establishment, 51 concluded in 2025) are
attributed to the *Annual Report of the Intellectual Property Tribunal of the
Supreme People's Court (2025)* as reported in the source; DCC has not
independently verified them. Article 39 is quoted from
[the Anti-Unfair Competition Law](/laws/anti-unfair-competition-law/) as revised
in 2025; the judgment itself applied the identically worded Article 32 of the
2019 version.

— Not legal advice.