Raw sources
GitHub, papers, blogs, and public discussion material remain the claim source.
Resource library / English mirror
Read the corpus as an evidence chain, not as a pile of links.
The useful question is not how many repositories or papers were collected. The useful question is whether a public claim can move backward from a site page to a report, an index, a paper, or a source repository.
Count boundaries
Different numbers answer different questions.
| Count | Scope | How to read it |
|---|---|---|
| 686 | raw GitHub captures | Source captures used before classification and analysis. |
| 686 | classified GitHub repos | Repository rows that have entered the evidence taxonomy. |
| 300 | analyzed model-card reports | Processed project reports tracked by the governance index. |
| 491 | review-gated public report snapshots | Readable source-tracing pages, not per-page quality certification. |
| 196 | survey-display papers | Papers currently shown by the site survey view. |
| 97 | public discussion signals | Public semantic signals, not controlled interview results. |
Pipeline
From source material to public claims.
Processed analysis
Classification, model cards, paper reviews, triage queues, and worksheet scores turn sources into comparable evidence.
Survey synthesis
The survey organizes mechanisms into working loops and keeps claim limits visible.
Public results
The site, reports, graph, PDF, and topic pages help readers move from conclusion back to source.