Core corpus size is taken from the parser log on the build machine. Verify the source line:
grep "DB var" /srv/beyin/kaynaklar/sira.log → [adapter] cc100_tr: DB var (109279716 madde) - ATLANDI
Comparable proof commands for sister corpora are documented in the pipeline repo (README → corpus manifest).
| Status | Corpus | Proof |
|---|---|---|
| 🟢 built | wikipedia (EN + 10 langs), wiktionary/wikiquote/wikisource (TR+EN), arxiv, tatoeba TR/EN | ls /srv/beyin/wiki/*.faiss |
| 🟡 running | cc100-tr (109.3M docs, embedding in progress) | live embed log |
| 🔵 queued | faiss indices for newscrawl, gutenberg, wet, paracrawl, caselaw | roadmap table |
Index build config: IVF-PQ (PQ32), nprobe 128, k*64 candidates → full-vector rerank. Config lives in faiss_kur.py in the repo.
Full license texts ship with the corpus manifest repository. Commercial use excludes non-commercial corpora (WikiHow) at the product layer — physically separate indexes. See Corpus Manifest table for per-corpus positions.
On-chain verifiable. Address and QR on the Donate section are identical. Verify on any block explorer:
mempool.space/address/bc1qelnjsg2m7nzedlwc49wcc2ugpsh9mpul9pj34g
LICENSE in repo (fork allowed)No settlements have occurred yet. The first settlement will be published here with: epoch, amount, miner count, transaction ID, and explorer link. All future payouts will remain public and append-only. ⚪ at network launch
Status page is planned at status.nemes-x.space. Until then, component health is visible in the Network Status table (updated as we ship) and in GitHub commit activity.
Found a discrepancy between what we say and what the code does? Open an issue. The burden of proof is on us.