About the show
How China Reports Review is made
A machine-generated podcast should show its workings. This page is generated from the same configuration and rules as the episodes.
Production
The pipeline uses gpt-5.6-sol, configured at xhigh reasoning effort, to discover reports, frame an episode and write its script. Speech is generated by gemini with gemini-3.1-flash-tts-preview, alternating between 2 configured voices at chapter boundaries.
Report selection
Each episode aims for 5 reports. Reports published within the latest 30 days are preferred, as are reports from the configured pool of 30 organisations below. A report must be primarily about China and available to read in full. PDF reports must clear the 15-page floor; the code also recognises a documented long-form web equivalent.
When the preferred pool is too thin, the selection code can use older reports up to 60 days old or substantial work from a comparable organisation outside the list. These are explicit fallback paths: the script is instructed to tell the listener when an episode reaches outside the usual window or source pool, and why.
Configured source pool
- CSIS (Center for Strategic and International Studies)
- Brookings Institution
- RAND Corporation
- Council on Foreign Relations
- Carnegie Endowment for International Peace
- Chatham House
- MERICS (Mercator Institute for China Studies)
- Rhodium Group
- ASPI (Australian Strategic Policy Institute)
- Lowy Institute
- IISS (International Institute for Strategic Studies)
- Peterson Institute for International Economics
- Hoover Institution
- American Enterprise Institute
- Bruegel
- European Council on Foreign Relations
- Stimson Center
- Center for a New American Security
- SIPRI
- National Bureau of Asian Research
- German Marshall Fund of the United States
- Foundation for Defense of Democracies (FDD)
- Central European Institute of Asian Studies (CEIAS)
- Jamestown Foundation
- Atlantic Council
- Asia Society Policy Institute
- German Institute for International and Security Affairs (SWP)
- Royal United Services Institute (RUSI)
- Wilson Center
- German Council on Foreign Relations (DGAP)
What “and challenged” means
Each report is handled on its own. The script identifies its claim, follows the reasoning, names the evidence, surfaces its forecast and tests the weakest link. A challenge must turn on the report's actual data, sources, assumptions or omissions; it must grant what the report gets right and say what evidence would settle the question. Unsupported contrarianism is excluded by the writing rules.
For a narrow factual check that could settle a challenge, the writer is instructed to look outside the Western report pool, attribute the source in the episode, and say when it could not find an independent figure. Those checks are evidence for a challenge, not additional reports under review.
Transcripts, sources and errors
Every episode publishes its full transcript in HTML and plain text. The podcast feed also carries the reports discussed in each episode's show notes. The site does not state a separate corrections policy. To report an error, identify the episode and passage, include the supporting source, and use the operator contact carried in the RSS feed.