Chapter, verse and checksum

BigBio does not summarize documents it has not read. In the translation linked below, the Chinese sits directly above every English paragraph, so any sentence can be checked against the source characters by anyone who can read them. The checks that follow are the ones a reader cannot do by eye.

The deliverable: the full side-by-side translation, all eighteen pages, the Chinese above every English paragraph. It is unofficial; the Chinese text is the sole authoritative version. What follows is how it was made and how it was checked.

The ask

On 18 September 2026 the Ministry of Industry and Information Technology and nine other ministries published the 15th Five-Year Plan for pharmaceutical industry development (医药工业发展"十五五"规划), signed 28 August. A week after publication, the English-language record consisted of broker notes and news summaries. No official English text exists; MIIT publishes in Chinese only. The last plan set the precedent. When the 14th plan (工信部联规〔2021〕217号) appeared in January 2022,11State Council of the PRC, policy library copy of 工信部联规〔2021〕217号, published 31 January 2022. https://www.gov.cn/zhengce/zhengceku/2022-01/31/content_5671480.htm (T1 -- government mirror) English readers got summaries: briefs from MERICS and Trivium China, and an English notice from China's medical-device association that linked back to the Chinese text.22MERICS, "China Industries" brief, February 2022 (published March 2022) -- a summary of the plan, not a translation. https://merics.org/sites/default/files/2022-03/MERICS%20February%202022%20China%20Industries.pdf (T3 -- think-tank summary)33Trivium China, "Pharma industry: target setting in the shadow of Covid-19", 9 February 2022 -- a summary of the plan. https://triviumchina.com/2022/02/09/pharma-industry-target-setting-in-the-shadow-of-covid-19/ (T3 -- consultancy summary)44China Association for Medical Devices Industry, English notice of the plan, January 2022, linking the Chinese text. http://en.camdi.cn/news/5045 (T2 -- industry association) We found no full English translation. Any US biotech deciding this quarter whether China's stated targets touch its pipeline, its CDMO, or its licensing partner had a summary of a summary to work from. That was the ask on the morning of 25 September: the original, in English, with proof.

The deliverable, first

The result before the method. Box 1 of the plan is the whole document in ten rows: the indicators MIIT will score itself against in 2030. Here is the table as translated, the Chinese beside the English so a reader can check the units.

Notes, as printed in the source: 〔〕 marks a five-year cumulative figure. The FIC share counts products in Phase I to III, at NDA/BLA, and approved; it excludes preclinical, terminated, and withdrawn programs. One term the plan does not define: "above designated size" (规模以上) is the National Bureau of Statistics threshold for an industrial enterprise, annual revenue of 20 million yuan and above since 2011 (NBS, statistical FAQ on industry, question 14: 2011年开始至今，统计范围为年主营业务收入2000万元及以上的工业法人单位),55National Bureau of Statistics of China, FAQ "六、工业统计", question 14 -- "2011年开始至今，统计范围为年主营业务收入2000万元及以上的工业法人单位". http://www.stats.gov.cn/hd/cjwtjd/202302/t20230207_1902277.html (T1 -- national statistics agency) so the 3.5 trillion target counts only those.

The plan states its money targets in yuan, except one in dollars, and converts none of them. We built the table below so the rows compare in one unit. Rate: 6.6975 yuan per US dollar, the Federal Reserve H.10 noon buying rate for 18 September 2026; H.10 gives 6.7110 for 25 September, about 0.2% higher.66Board of Governors of the Federal Reserve System, H.10 Foreign Exchange Rates, China historical series -- 18-SEP-26 6.6975; 25-SEP-26 6.7110. https://www.federalreserve.gov/releases/h10/hist/dat00_ch.htm (T1 -- central bank statistical release) The ECB reference cross for 25 September, 7.6551 CNY and 1.1403 USD per euro, gives 6.7132.77European Central Bank, euro foreign exchange reference rates, 2026-09-25 -- CNY 7.6551; USD 1.1403. https://data-api.ecb.europa.eu/service/data/EXR/D.CNY.EUR.SP00.A?startPeriod=2026-09-25&endPeriod=2026-09-25 (T1 -- central bank statistical release) We round to two significant figures; the 2030 figures are targets, so a third digit would be false precision.

The plan's shape, in one unit: fifty companies each above US$1.5 billion in revenue, twenty parks each above US$15 billion, five products each above US$1 billion in global sales, inside an industry the state expects to reach US$520 billion. The last row is the smallest firm the 3.5 trillion counts.

Every row above is on the translation page with its full context, the Chinese beside it. Getting to that table took four steps, and each one left a record.

Finding the original

The first step was finding the document at all. "Biotech five-year plan" returns nothing useful, because the document is not indexed under the reader's words. It is indexed under the issuer's: the ministry's own title, in Chinese, with its document-number convention. 医药工业 十五五 印发 returns the MIIT notice in one query.

The notice page does not link the PDF. The file sits inside the page's embedded PDF viewer (viewer.html?file=/cms_files/.../attach/....pdf), a URL no search engine indexes. The URL surfaces in the page's HTML, not in the search results.

We fixed provenance before translating anything:

The citation on every downstream artifact is the ministry, never the search engine that located it. With the file hashed and the mirrors noted, there was something to translate.

Two registers, two renderings

The second step was the translation, and the first decision inside it was that one method would not do. A Chinese plan is written in two registers, and they fail in opposite ways when translated the same way.

The binding layer is the targets, thresholds, currencies, dates, and verbs of obligation. It must be rendered character-literally and then checked by machine, because a fluent reader cannot see the errors that matter here: 100亿元 is 10 billion yuan, not 100 billion; 〔>= 200〕 is a five-year cumulative count, not a per-year one; 不低于 is "no less than", which includes the number, where 超过 is "exceeding", which does not.

The slogan layer is the four-character parallel couplets that head every task section. They carry rank order and funding intent, and a literal rendering is accurate and unreadable. Task (1) is headed 引领产业向新向优，锻造创新驱动新引擎: literally "lead the industry toward the new and the superior, forge a new innovation-driven engine". Our gloss: first-in-class innovation is the first-funded priority, and 向新向优 means novel over incremental, quality over volume. A gloss is an interpretation, so the page carries both renderings, literal above and gloss beside, with the characters in view. A reader who disagrees with the gloss has the literal to disagree from.

We split the plan into four chunks at its section headings, so no paragraph straddled a boundary, and translated the chunks in four parallel passes under the same brief: reassemble paragraphs without altering characters, quote every Chinese paragraph directly above its English, render the boxes as two-column tables, hold a fixed glossary for the policy terms that have settled English forms (new quality productive forces, manufacturing powerhouse, Healthy China, first-in-class), never summarize, never omit a sentence, and add a translator's note only where the source is inconsistent with itself. Four passes gave a complete draft inside the hour. A draft is where the errors live, and the next two steps were for finding them.

The checks a reader never catches

The third step was mechanical. A parity script pairs every Chinese paragraph with its English and flags four classes of defect. A careful bilingual reader misses all four on a long document, because each turns on a single character or a dropped sentence.

A flag is a question, not a verdict. The translator adjudicated all 26 against the characters, and the review ledger published with the translation records each adjudication.

The plan itself contains four scaled figures, all in Box 1 (the fifth row of the conversion table above is the statistics bureau's definition, not the plan's), and we also converted each by hand. 亿 (10^8) has no English unit, and re-basing it to millions and billions is where press coverage of these plans most often goes wrong.

The script clears the numbers. It cannot judge meaning. For that we needed a reader who wanted us to be wrong.

Adversarial review, ledger shown

The fourth step has a rule behind it: a review must cross model families, because a translator does not grade its own work. Every reviewer available from another family came from a Chinese lab, which is the right blind spot to test for a PRC political document: the brief asked each reviewer to confirm that the ideological formulations (Xi Jinping Thought, the Fourth Plenum, the national security barrier, a global community of health for all) were rendered without softening. All three received the identical brief. We ran DeepSeek and Qwen at temperature zero, listing defects only, each anchored to an exact Chinese fragment.

The four accepted flags, in order: (1) 超限制造, now "beyond-limit manufacturing" rather than "extreme manufacturing"; (2) 全球首创, now "world-first", so that first-in-class stays the document's defined term for 首创新药 and does not bleed into a regional cluster's slogan; (3) and (4) the two places where the source disagrees with itself, taken up in the next section. The raw ten, with each verdict, are in the review ledger published with the translation.

We record Kimi's refusal as a finding about the reviewer, not the document. The error class is the API's own, content filter, not a rate limit or a malformed request, and the same prompt ran unchanged on the other two reviewers. A model from a Chinese lab declined to review official PRC text for softening. That is the blind spot the cross-family rule exists to expose, and a review process that quietly dropped the refusal would hide its most informative result. A flag a reviewer raises is evidence; a flag it fails to raise is not, so a separate pass also read the politically loaded passages line by line against the characters. None needed a change.

Four flags survived the review. Two of them pointed at the source, not at the translation.

What we left broken on purpose

The source disagrees with itself twice, and both times the abbreviation is the established term while the characters say something else.

• 热缩性弹性体（TPE）: the characters say heat-shrinkable, the abbreviation says thermoplastic elastomer.
• 多功能诱导干细胞（iPSC）: the characters say multifunctional, the abbreviation says induced pluripotent stem cell.

We render both as written, with a translator's note inline. A translation that corrected either would be more readable and less faithful, and a reader who later compares the English to the characters would find a discrepancy the translator introduced. The defect belongs to the ministry's drafters. It stays visible.

What this generalizes to

By late morning there was a translation, a ledger, and a hash. Nothing in the way of getting there was specific to China or to pharmaceuticals. The method applies to any non-English regulatory or policy text that a US biotech has to act on before a think tank gets around to translating it: a Japanese PMDA notice, a Korean MFDS guideline, an EU document whose only current version is in French. The steps are the same. Find the issuer's own page in the issuer's own words and fix the provenance before reading a line. Render the binding layer literally and check it by machine, because the errors that change a decision are single characters. Gloss the rhetoric beside the literal, with the source in view. Have a reviewer from a different family attack the work, keep the ledger, and record the refusals with the findings. Publish the translation with the source above every paragraph, so that a reader can check the translator instead of trusting one.

Method: original PDF from the MIIT authority page, sha256 e4ba3a4c737f9ed1deafde79fb0d7cf63b4bd47aac93f68502aadaa95a25da8e, downloaded 2026-09-25. Translation assembled 2026-09-25 by BigBio's Claude-based pipeline under the brief above, UNOFFICIAL; the Chinese text is the sole authoritative version. Mechanical parity checks by script over every paragraph pair. Adversarial review: DeepSeek deepseek-v4-pro at temperature 0; Qwen3.6-35B 4-bit, local; Kimi (Moonshot) refused. Exchange rate: Federal Reserve H.10, 18 September 2026. Every number, threshold, and negation is machine-matched to the Chinese; meaning-level fidelity rests on one cross-family review and the translator's own reading of the politically loaded passages.

Footnotes

1. State Council of the PRC, policy library copy of 工信部联规〔2021〕217号, published 31 January 2022. https://www.gov.cn/zhengce/zhengceku/2022-01/31/content_5671480.htm (T1 -- government mirror) ↩

2. MERICS, "China Industries" brief, February 2022 (published March 2022) -- a summary of the plan, not a translation. https://merics.org/sites/default/files/2022-03/MERICS%20February%202022%20China%20Industries.pdf (T3 -- think-tank summary) ↩

3. Trivium China, "Pharma industry: target setting in the shadow of Covid-19", 9 February 2022 -- a summary of the plan. https://triviumchina.com/2022/02/09/pharma-industry-target-setting-in-the-shadow-of-covid-19/ (T3 -- consultancy summary) ↩

4. China Association for Medical Devices Industry, English notice of the plan, January 2022, linking the Chinese text. http://en.camdi.cn/news/5045 (T2 -- industry association) ↩

5. National Bureau of Statistics of China, FAQ "六、工业统计", question 14 -- "2011年开始至今，统计范围为年主营业务收入2000万元及以上的工业法人单位". http://www.stats.gov.cn/hd/cjwtjd/202302/t20230207_1902277.html (T1 -- national statistics agency) ↩

6. Board of Governors of the Federal Reserve System, H.10 Foreign Exchange Rates, China historical series -- 18-SEP-26 6.6975; 25-SEP-26 6.7110. https://www.federalreserve.gov/releases/h10/hist/dat00_ch.htm (T1 -- central bank statistical release) ↩

7. European Central Bank, euro foreign exchange reference rates, 2026-09-25 -- CNY 7.6551; USD 1.1403. https://data-api.ecb.europa.eu/service/data/EXR/D.CNY.EUR.SP00.A?startPeriod=2026-09-25&endPeriod=2026-09-25 (T1 -- central bank statistical release) ↩

8. Ministry of Industry and Information Technology, notice publishing 工信部联规〔2026〕210号, 2026-09-18 -- authority page. https://www.miit.gov.cn/zwgk/zcwj/wjfb/tz/art/2026/art_9463adac4db040159f6cbf482675eeea.html (T1 -- issuing ministry) ↩

9. Ministry of Industry and Information Technology, 《医药工业发展"十五五"规划》 (original PDF, 18 pages, sha256 e4ba3a4c737f9ed1deafde79fb0d7cf63b4bd47aac93f68502aadaa95a25da8e). https://www.miit.gov.cn/cms_files/filemanager/1226211233/attach/20269/c9ef9c990ec84347a7a87999e23e26e0.pdf (T1 -- issuing ministry, primary document) ↩

10. State Council of the PRC, policy library copy of 工信部联规〔2026〕210号. https://www.gov.cn/zhengce/zhengceku/202609/content_7081457.htm (T1 -- government mirror) ↩

11. Hunan Provincial People's Government, copy of the plan PDF. https://www.hunan.gov.cn/zqt/zcsd/202609/34067914/files/0eeed358d8864446b24ef51a7548c742.pdf (T1 -- government mirror) ↩