DeepSeek V3.2Evidence firstEN + 中文5–10 audits / day
Runtime configuration · deepseek/deepseek-v3.2

DeepSeek reads the claim.Evidence decides what survives.

RealitySift is a DeepSeek-powered China reality auditor. It reads official statistics, policy documents, speeches, state-media reports and corporate disclosures — then separates what was said from what the evidence actually establishes.

RealitySift 由 DeepSeek 驱动。系统持续读取中国官方数据、政策文件、讲话、国家媒体报道和企业披露,并把“来源说了什么”与“证据真正证明了什么”分开。

Continuous intake ACTIVE
Every 3 hours
Source scan cadence
1–2 claims
Maximum per scan
5–10 / day
Target audit throughput
Human approved
External publication

The monitor collects candidates automatically. DeepSeek performs the interactive audit. Numbers and words are both testable objects; authority is never used as a substitute for evidence.

What it watches

Not just numbers.

A GDP figure, a ministry target, a sentence in a speech and a causal claim in state media enter the same evidence pipeline.

State Council / Xinhua
National Bureau of Statistics
General Administration of Customs
People's Bank of China
NDRC
MOFCOM
MIIT
Ministry of Foreign Affairs
Plus laws, central and local policy documents, listed-company disclosures, research institutions and independent evidence used for corroboration.
Daily Watch · initial production batch

Seven claims. Seven different evidence problems.

The point is not to assume official Chinese claims are true or false. The point is to classify exactly what kind of claim was made, test what can be tested, and leave the unresolved part visible.

8 Sep 2026 intake
DW-001
2026-09-08
GAC via State Council / Xinhua
STATISTICAL CLAIMKEEP

China's Jan–Aug goods trade reached 34.78 trillion yuan, up 17.6% year on year.

中国1—8月货物贸易进出口总值34.78万亿元,同比增长17.6%。

The official release supports the number, period and unit. It does not by itself prove the headline's causal story about overseas demand.
Open source
DW-002
2026-09-08
NHSA notice via State Council / Xinhua
POLICY COMMITMENTQUALIFY

China will seek to enroll more gig, migrant and flexible workers in basic medical insurance over the next three years.

未来三年,中国将推动更多灵活就业、新就业形态和农民工群体参加基本医疗保险。

The notice establishes a policy commitment. It does not establish how many people will actually gain coverage or whether the measures will succeed.
Open source
DW-003
2026-09-08
15th Five-Year IP plan via State Council / Xinhua
POLICY INTENTQUALIFY

China will refine IP rules for AI, algorithms, AI-generated content and platform-economy issues during 2026–2030.

“十五五”期间,中国将完善人工智能、算法、生成式内容和平台经济等领域的知识产权规则。

The plan confirms intent and scope. Future implementation, enforcement quality and practical effects remain separate empirical questions.
Open source
DW-004
2026-09-07
MIIT guideline via State Council / Xinhua
TARGETQUALIFY

By 2028, China targets 50 classical-industry enterprises above 10 billion yuan annual revenue and 100 national consumer brands.

到2028年,政策目标包括培育50家年营收超100亿元企业和100个全国性消费品牌。

The targets are source-confirmed. They are not present-day achievements and should not be reported as outcomes before the target dates.
Open source
DW-005
2026-09-08
Ministry of Foreign Affairs
DIPLOMATIC CHARACTERIZATIONQUALIFY

Xi Jinping said China and the U.K. have more common interests than disagreements.

习近平表示,中英两国“共同利益大于分歧”。

The MFA record verifies that the statement was made. The proposition itself is a diplomatic judgment, not a directly measurable count of interests versus disagreements.
Open source
DW-006
2026-09-05
Vice Premier Ding via State Council / Xinhua
CAUSAL CLAIMUNVERIFIED

China–Russia energy cooperation has played an important role in global energy-market stability.

丁薛祥表示,中俄能源合作对全球能源市场稳定发挥了重要作用。

The speech proves the claim was made. Establishing the causal effect requires market evidence and a credible counterfactual; the speech alone is insufficient.
Open source
DW-007
2026-09-07
Multi-agency rural plan via State Council / Xinhua
FUTURE POLICY OUTCOMEQUALIFY

China aims to basically establish a rural-revitalization investment mechanism aligned with development conditions by 2030.

到2030年,中国目标是基本建立与农业农村发展相适应的乡村振兴投入机制。

The plan establishes the stated goal and instruments. Whether the mechanism is actually established by 2030 remains a future test.
Open source
Production Case 001 · NBS · 1 September 2026

Manufacturing improved. A broad recovery was not established.

QUALIFY
Primary source: National Bureau of Statistics of China
49.8
Manufacturing PMI
July: 49.2
50.4
Production
Expansion
50.6
New orders
Expansion
49.5
Composite PMI
Below 50
Executive finding

49.2 → 49.8 means the manufacturing diffusion index improved while remaining below 50. Production and new orders moved above 50, but the composite remained below 50. None of those readings is a direct measurement of real GDP growth.

Discriminating test

A durable recovery claim should survive several later releases and broaden across output, orders, employment and non-manufacturing activity. One improving month is not enough.

Claim ledger

Corrections are data.

Bad calls are not erased. RealitySift records what changed, why it changed and what evidence triggered the change.

RETRACT / REPLACE
Previous claim
Non-manufacturing employment was not reported
NBS reported an August employment diffusion index of 45.4 and new orders of 44.1. Diffusion indexes are not jobs counts.
RETRACT / REPLACE
Previous claim
PMI 49.5 proves real GDP contracted
PMI is a diffusion survey of activity, not a direct measure of real GDP growth.
QUALIFY
Previous claim
One improving month proves a turning point
One release can support an early-stabilization hypothesis, but a trend requires additional sourced observations.
Same standard. Every claim.

An audit trail you can attack.

Chinese, Western, official, academic, state, private: origin never substitutes for evidence quality.

01

Extract the exact claim

A number, target, speech line, causal assertion or institutional description is captured before interpretation begins.

02

Trace the source

Date, document, wording, unit, denominator, sample and reporting period stay attached to the claim.

03

Test both directions

Evidence for the claim and the strongest counterevidence are evaluated under the same standard.

04

Publish the delta

KEEP, QUALIFY, RETRACT and UNVERIFIED remain visible instead of being silently rewritten away.

DeepSeek inside. Evidence in public.

Give it a statistic, a speech, a policy document or a headline.

RealitySift will tell you what the source actually says, what can be verified, what remains unproven, what would falsify the leading explanation and how confident the verdict should be.

Open DeepSeek auditor