One Rule, Three Names: A Search for China’s Capital-Acceleration Doctrine Finds 44 Per Cent of It

A creditor holding a judgment against a Chinese limited liability company with no assets has one question that matters: can the shareholders be made to pay the capital they subscribed but were not yet due to contribute? That question has an answer in Chinese law, and the answer moved in July 2024. It also has an answer in the case law, and that is where a retrieval system, an evaluation set or a research memo quietly goes wrong, because Chinese courts describe this single rule in at least three wordings, and the best-known one finds 44 per cent of the records.

This is the company-law sibling of our category-recall audit. That piece measured a label that arrived years after the cases. This one measures vocabulary that changed under a statute, and a cause-of-action filter that was never going to work. Every figure below is an exact count taken on 13 September 2026 against our own judgment index, with total-hit tracking on, unless it is labelled as a sample.

The rule, in four steps

Since the 2014 amendment, PRC company registration has run on subscribed rather than paid-in capital: shareholders commit to an amount and to a date, often decades away, and enjoy the benefit of that term. Whether a creditor could collapse the term was unsettled until the Supreme People’s Court’s national commercial adjudication minutes of November 2019 (widely called the Nine Minutes) stated a default — shareholders keep their term — and two exceptions: a company that is a judgment debtor with enforcement exhausted and no assets, which meets the conditions for bankruptcy but has not filed; and a shareholders’ resolution that extended the contribution term after the debt arose. Judgments of that period anchor the result in Article 13(2) of the Third Company Law Interpretation and, by analogy, Article 35 of the Enterprise Bankruptcy Law.

The Company Law revised on 29 December 2023 and in force from 1 July 2024 replaced the exceptions with a rule. Article 54: where a company cannot pay its debts as they fall due, the company or a creditor whose claim has fallen due may require shareholders who have subscribed but whose term has not expired to pay their contributions early. The revision also capped subscription periods for limited liability companies at five years. In enforcement, creditors typically pair Article 54 with Article 17 of the Supreme People’s Court provisions on changing and adding parties in civil enforcement to ask that shareholders be added as judgment debtors.

For data purposes the consequence is simple. The doctrine kept its name in commentary, but the statute that now governs it uses different words.

Three wordings, each confirmed by sampling

We counted three phrase families and then checked each by reading. For each group we drew 25 records with a fixed random seed (20260913, random_score on _seq_no), because an unseeded top-N from a merged index returns whichever ingest batch happens to sit first.

Group A, the name every practitioner would type, is 44.37 per cent of the union. Adding the statutory wording lifts that to 50.93 per cent. Group C is the largest single family and consists of ordinary judicial prose: “the shareholders’ contribution term should accelerate”, “subscribed capital accelerates”. Nothing in it is exotic. It simply does not contain the four characters of the label.

The union is a floor, not a census. We did not sample the bare phrase 加速到期 (acceleration) on its own. It appears in 114,087 records, and loan-acceleration clauses are an obvious competing sense, so we paired it with 认缴. Judgments that describe the rule in other words are not counted here.

The statute changed the words faster than it changed the doctrine

Splitting A and B by judgment date around the day the revision took effect shows the handover directly.

In the first half of 2025 the statutory wording overtook the doctrinal label for the first time, and it has stayed level with it since. For the 2025 judgment year as a whole, group A reaches 1,864 of the 4,425 records in the union, which means a query on the doctrine’s name misses 57.9 per cent of that year. In 2022 the same query missed 53.3 per cent. The loss was already large and has grown. The pre-statute B-only records are a reminder that early payment was also a phrase before it became a statute: between 85 and 201 records a year used it without the label from 2020 to 2022.

Temporal application shows up in the text. Courts applying Article 54 to older facts cite the Supreme People’s Court provisions on the temporal effect of the Company Law. Those provisions, as the judgments quote them, apply the new law to legal facts arising after it took effect and, for earlier facts, the law in force at the time where that law had a rule. Records that contain both the statutory wording and the phrase 时间效力 (temporal effect) fell from 19.7 per cent of statutory-wording records in 2024 H2 to 14.6 per cent in 2025 H1 and 9.9 per cent in 2025 H2. Treat that as an upper bound: in a sample of ten, one cited the Civil Code’s temporal provisions, not the Company Law’s.

The obvious query for the revised statute returns zero

Before settling on the statute’s operative words, we tried the two searches most people would run. The phrase 2023年修订 (revised 2023) returns 0 records. The citation 第五十四条 (Article 54) also returns 0. That is plainly false, because sampled judgments quote 《中华人民共和国公司法》第五十四条 verbatim.

The cause is the analyzer pair. Our body field is indexed with one Chinese tokenizer and searched with another. At index time 第五十四条 becomes five overlapping tokens (第五十, 第五, 五十四, 十四, 四条). At query time it becomes two (第五十, 四条). Phrase matching requires consecutive positions, the positions never agree, and the engine returns a clean, silent zero. 2023年修订 fails the same way (2023 / 年 / 修订 at index time, 2023年 / 修订 at query time). The phrases we relied on, 出资加速到期, 提前缴纳出资, 加速到期 and 认缴, tokenise identically on both sides. We checked each with both analyzers before counting.

This is not peculiar to us. Any Chinese full-text system whose index-time and query-time analyzers differ can do it. A zero is the most quotable result a search can produce and the least trustworthy one, so for any Chinese phrase that returns nothing, run the analyzer on both sides before believing it.

Raw counts say it peaked in 2022; per record held, it has risen every year

Plot the second column and you would write that acceleration litigation peaked in 2022 and fell by 61 per cent. Plot the last column and you would write that it doubled after 2022. Neither is a fact about filings. The 2022 peak is inflated by duplication: 28.0 per cent of that year’s records share a case number with another record, against effectively none in 2024 and 2025. The fall after 2022 follows the corpus-wide recency cliff, which comes from publication volume, not court activity. The rising rate is the less misleading reading, but it cannot rule out that the judgments still published after 2023 are a different mix from those published before. A credible trend would need an external denominator. We do not have one, so we do not claim a trend.

A cause-of-action filter finds one record in six

The natural shortcut is to filter on the causes built for this: shareholder capital disputes, recovery of unpaid capital, and shareholder liability for damaging creditors’ interests. The union shows why that fails.

The largest bucket is the ordinary claim. Creditors raise acceleration inside the sales-contract, general-contract, lease, labor or private-lending case that produced the debt, or in the enforcement ruling that follows it, and the record keeps that original label. In the A-or-B subset, leaving aside blank labels, the most common labels are the shareholder-creditor cause (2,337), enforcement objection suits (2,300) and sales contracts (1,822). The blank-label share has fallen sharply in recent batches, to 4.45 per cent of 2025 A-or-B records. The underlying-dispute share has grown instead: 49.45 per cent of 2025 A-or-B records, against 41.81 per cent across all years.

Where the fight moved after July 2024

A statute that states a rule does not end argument about the route. The records show two recurring disputes, and each can be checked against the court’s own case number:

• Whether an enforcement court can add shareholders directly. In (2025)鄂0606执异74号 (Xiangyang Fancheng District People’s Court, 22 April 2025), the court quotes Article 54 as in force from 1 July 2024 and reasons that shareholders who have not paid no longer enjoy the benefit of their term. In (2025)京0108执异41号 (Beijing Haidian District People’s Court, 10 January 2025), the court finds no clear legal basis for requiring early contribution within that proceeding and declines an application to add shareholders under Article 17.
• Whether inability to pay is enough without bankruptcy. Appellants continue to argue the pre-2024 position. In (2025)渝01民终9351号 (Chongqing No. 1 Intermediate People’s Court, 11 November 2025), an appellant contends that inferring acceleration from inability to pay substitutes civil litigation for bankruptcy procedure. In (2024)新01民终8998号 (Urumqi Intermediate People’s Court, 27 February 2025), an appellant argues that the new provision lacks implementing rules.

These are the arguments in the text, not the outcomes. The last two describe what a party argued, and we have not counted how either question is resolved across courts. That count needs the reasoning read, not phrases matched.

How to reproduce this against any corpus

1. List every wording of the rule you can find in a first sample: the doctrinal label, the statute’s operative words, and the plain-prose pairing. Read at least 25 hits of each before counting anything.
2. Run every phrase through both the index-time and query-time analyzers. Discard any phrase whose token positions differ, and do not publish its zero.
3. Count each family exclusive of the ones before it, then the union, with total-hit tracking on. Never read a capped total as a count.
4. Split by judgment date around the effective date of the statute, and report the share reachable only through the new wording.
5. Report distinct case numbers beside records, and normalise by records held per year before describing any trend.
6. Break the union down by cause label, including blank. That table tells you whether a cause filter is usable for this doctrine. Here it is not.

What we would hand a buyer

If your legal AI answers questions on PRC shareholder liability, the test takes an afternoon. Ask the system for acceleration cases decided in 2025, run the three phrase families above against the same corpus, and compare. A system tuned on pre-2024 terminology, or one that routes by cause of action, will show the losses in the tables above. A retrieval evaluation set built from the doctrinal label alone will score a system as complete when it has missed most of the post-revision case law. We can supply the record identifiers behind each figure so the comparison runs on identical records.

What we are not claiming

We did not measure any other provider’s data, so nothing here is comparative. The union is a floor, since wordings outside the three families are uncounted. A 25-record sample per family establishes that the families are on-topic, not a precise precision rate. The temporal-effect share is an upper bound. The per-million rate is normalised by records held, not cases filed, and published judgments after 2023 may not be a representative mix. The four cases above illustrate the arguments and do not describe the distribution of results. We fixed nothing in the corpus. The analyzer mismatch that produces the false zeros is still in place, and we have written it down instead of silently changing search behaviour that existing integrations depend on.