Braid Zero
A small open collection that proves your BRAID pipeline works before valuable data enters it.
- Size
- 43 accepted
- Lineage
- Independent conformance fixture
The corpus catalog
10 locked entries, 1 candidate, and one clear status for each.
Locked means the artifact identity is fixed. Candidate means it is visible for planning but is not a frozen release.
A small open collection that proves your BRAID pipeline works before valuable data enters it.
A compact English knowledge corpus for training tests, search, embeddings, and full pipeline checks.
Pinned English Wikipedia prose that expands every C1 topic into a training-scale text corpus.
An evidence-linked curriculum of claims, cloze tasks, and source-retrieval contrasts derived from C2.
A repaired ZERO.5 512-token target with mirrored claims, unchanged cloze records, and paired retrieval.
A target-aware expansion of C3 with grounded evidence bundles and separate 512- and 1024-token memberships.
The smooth, pair-atomic 512-token target combining C3.2 repairs with C4 evidence bundles.
Natural model text and structured objective sidecars with C4.1 membership kept exactly intact.
Nested fresh-source natural prose at exactly 100 million and 1 billion ZERO.5 training tokens.
Deterministic structured-state worlds with executable proofs and aligned text and symbolic training views.
A private, source-grounded relational candidate over the 999 C1 entities for Zero E1.