What we caught when we walked every Chinese-policy quote back to its primary source.

Eleven of twelve anchors, primary-source verified

The founding corpus of twelve doctrinal anchor passages shipped with the working English carrying editorial paraphrase rather than primary-source quotation. Speaker, venue, and date were independently verified; the wording was not. This release closes that gap. Eleven of the twelve anchor passages now carry primary-source verificationagainst PRC official mirrors (Foreign Languages Press, Xinhua, Qiushi, gov.cn, the State Council Information Office, China Daily) and Western translations of record (Stanford DigiChina, CSIS Interpret, Rogier Creemers' China Copyright and Media). The twelfth — an August 2020 People's Daily editorial on dual circulation — remains a placeholder pending print-archive retrieval. The audit surfaced six translation discrepancies along the way; the two heaviest both sit in Xi's 2 January 2019 Taiwan address. Separately, the December 2024 U.S. chip controls event-response pair was enriched with China's same-week retaliatory export ban and a coordinated statement from the country's four major chip-industry trade groups urging members to avoid U.S. silicon.

A falsifiable-prediction instrument cannot ship paraphrase

The premise of the predictor is that PRC responses to foreign events route through a specific doctrinal vocabulary. If the system is going to be wrong in public on a specific date, the corpus it retrieves from has to be worded the way the responses are actually worded — not the way a working summary made plausible. Paraphrase is a forecast-quality liability twice over: it strips the lexical hooks any retrieval system can match on, and it implies certainty about wording the underlying record has not yet confirmed.

The placeholder gate at first release was deliberate — better visible placeholders than fabricated quotation — but it was always a release-blocking debt. The audit pass below is what pays it down.

Walk each passage back to a primary or canonical source

For each of the twelve anchors, we identified the primary publisher of record (a PRC mirror where one exists, or the first Western translation of record where it does not) and resolved the working English against the source text. Where the working English diverged from the source, we adopted the source wording. Where two sources diverged from each other in doctrinally relevant ways, we logged the discrepancy rather than picking a winner.

Eleven passages resolved cleanly. Mao's 1949 and 1956 texts came from the Foreign Languages Press Selected Worksvolumes; Deng's 1978 and 1992 passages from the same series; Jiang Zemin's 1995 Eight Points from the PRC embassy and Taiwan Affairs Office mirrors; Hu Jintao's 2007 Scientific Outlook from the PRC Consulate Manchester mirror; Xi's 2012 Chinese Dream passage from Creemers' China Copyright and Media; Xi's 2018 Chinese Academy of Sciences keynote from Stanford DigiChina, the academic translation of record; Xi's 2 January 2019 Taiwan address from Xinhua directly and from CSIS Interpret; the August 2022 Taiwan white paper from the State Council Information Office, China Daily, and CSIS Interpret; and Xi's 2024 essay on new quality productive forces from the official Qiushi English edition.

The twelfth anchor — the 17 August 2020 People's Daily editorial on dual circulation, signed under the Ren Zhongping byline — remains a placeholder. The underlying event (the Politburo's introduction of dual circulation on 30 July 2020) is well-attested; the specific commentary requires People's Daily print-archive retrieval the lab has not yet completed.

Six translation discrepancies — two doctrinally heavy

The audit surfaced six translation discrepancies across five anchors, of which two carry direct doctrinal weight. Both sit in Xi's 2 January 2019 address marking the 40th anniversary of the “Message to Compatriots in Taiwan.”

China must be reunified, and will surely be reunified.

The CSIS Interpret rendering carries 必然 as deterministic, almost teleological — the “surely” doing serious doctrinal work. Xinhua's own English highlights flatten the same Chinese to “China must be and will be reunified” — resolute, but not metaphysical. The flatter wording has become the post-2022 PRC-English convention. A retrieval system that matches only on the Xinhua phrasing will silently miss every Western analyst piece that quotes the “surely” variant, and vice versa.

The second heavy discrepancy is in the same speech's use-of-force formulation. The literal translation of the Chinese (不承諾放棄) is “do not promise to renounce.” The PRC-English record shows a three-variant progressive hardening across the 2019–2022 doctrinal window: “make no promise to renounce” (Xinhua, 2019) → “do not renounce” (CSIS Interpret) → “will not renounce” (2022 Taiwan white paper).

That shading — from passive restraint not waived, to active retention of the force option — maps exactly onto the post-Pelosi doctrinal escalation the 2022 white paper itself codifies (the white paper drops the 1993 and 2000 commitments to refrain from sending troops or administrators to Taiwan post-reunification). The divergence is itself the signal. The four lighter discrepancies (in Mao 1956, Jiang 1995, Xi 2012, and the 2024 productive-forces essay) carry medium or low doctrinal load.

Why divergence is sometimes the finding

The naive view of a translation audit is housekeeping: pick the right wording, ship the corrected text, move on. The Xi 2019 use-of-force hardening is what happens when the housekeeping turns out to be substantive. The PRC-English record is not a single object; it is a curated sequence, and the curation itself encodes a doctrinal direction-of-travel. The three variants between 2019 and 2022 are not translation noise. They are PRC official translators tracking a hardening in the Chinese original — or in the official interpretation of the Chinese original — that the structure of the English sentence absorbs.

For the predictor, this means every retrieval against the Xi 2019 anchor should carry the full variant set, not just a single working text. A Taiwan-incident event that surfaces “reunified” language in 2026 maps to a different doctrinal posture depending on whether the PRC-English uses the 2019 wording or the 2022 wording. The schema work to make that distinction queryable — rather than buried in editorial notes — comes in the next release.

Separately on the event-response side, the audit enriched the December 2024 U.S. chip-controls entry. The original credited the PRC response with a single subsidy lever inferred from the broader industrial-policy cycle. The audit added two pieces of evidence the founding dossier had missed: China's commerce ministry issued a same-week ban on dual-use exports of gallium, germanium, and antimony to the United States, and within 48 hours the country's four major chip-industry trade groups jointly urged Chinese firms to avoid U.S. silicon. That pair joins the eight prior pairs in the corpus and tightens the observed response window from a week to under two days.

What this audit does not yet close

One anchor remains placeholder.The August 2020 People's Daily dual-circulation editorial requires print-archive retrieval; until that lands, the corpus runs at eleven of twelve verified. The dual-circulation event itself is verified through the Politburo readout; the specific commentary is not.

Translation discrepancies are documented but not yet queryable.The six discrepancies are logged as narrative editorial notes. Code that runs retrieval still sees only a single working English per anchor. The schema extension that makes variants queryable from code — with a paired surfacing of the heaviest discrepancies on the live page — lands in the next release.

Three Xi-era anchors stay deferred.Xi's 2017 19th Congress political report, the August 2021 common-prosperity speech, and the 2024 Third Plenum communique remain outside the live retrieval index. The audit pass focused on what was already in the index; loading those three is scheduled for the release after next.