-
Drop unreproducible design-distance multiple; correct undatabased non-host rate to 99%
-
Correct foundation-model gain description (promoters near-tie, histone marks mid-range); full-data fine-tune confirms
-
Comprehensive card rewrite; Markov-correlated generalization in the proof; correlation figure
-
Real foundation model (DNABERT) gain-vs-D, field-wide sweep, protein/RNA parity; boundary.py multiclass+regression
-
Add boundary.py (the D criterion as a standalone tool); tighten README
-
Composition boundary: machine-checked factorization/Bayes/JSD theorem + Nucleotide Transformer sweep showing D predicts solvability
-
Partition law and first-principles crossover; machine-checked robustness proof; provenance watermark
-
README: embed learned-space and composition-manifold figures
-
README: partition generalized to protein/RNA, MI crossover, certified frontier, dominant hybrid, manifold
-
Link atlas dataset instead of bundling
-
Move atlas to its own dataset (dna-origin-atlas)
-
Rich README + certify/attribute methods, design.py, bundled 8-mer atlas
-
Rich README + certify/attribute methods, design.py, bundled 8-mer atlas
-
Rich README + certify/attribute methods, design.py, bundled 8-mer atlas
-
Rich README + certify/attribute methods, design.py, bundled 8-mer atlas
-
Ship single k=8 discriminative model (host 0.993/0.990, engineered 0.919/0.896, 5-class 0.708); breaks adversary at order 7
-
Ship single k=8 discriminative model (host 0.993/0.990, engineered 0.919/0.896, 5-class 0.708); breaks adversary at order 7
-
Ship single k=8 discriminative model (host 0.993/0.990, engineered 0.919/0.896, 5-class 0.708); breaks adversary at order 7
-
Ship single k=8 discriminative model (host 0.993/0.990, engineered 0.919/0.896, 5-class 0.708); breaks adversary at order 7
-
Ship single k=8 discriminative model (host 0.993/0.990, engineered 0.919/0.896, 5-class 0.708); breaks adversary at order 7
-
Ship single k=8 discriminative model (host 0.993/0.990, engineered 0.919/0.896, 5-class 0.708); breaks adversary at order 7
-
Remove version label; flatten phrasing
-
Add defensive half: Carbon-8B resists the composition-matched adversary (0.53 -> 1.00 at order 5+)
-
Add base_model: Carbon-8B (derived by analysis) + clarify analytical derivation vs weight transfer
-
Add TOOL.md
-
Add ADVERSARIAL.md
-
v2: retrained on scaled+deduplicated benchmark (host 0.995/0.993, engineered 0.909/0.874, 5-class 0.69); add read-filter CLI
-
v2: retrained on scaled+deduplicated benchmark (host 0.995/0.993, engineered 0.909/0.874, 5-class 0.69); add read-filter CLI
-
v2: retrained on scaled+deduplicated benchmark (host 0.995/0.993, engineered 0.909/0.874, 5-class 0.69); add read-filter CLI
-
Build out card: dataset link, model-index eval results, references (Carbon-8B comparison, method lineage)
-
Replace with RC-folded multi-order spectrum (k=4,6,8) model: host 0.982, engineered 0.895, 5-class 0.65
-
Replace with RC-folded multi-order spectrum (k=4,6,8) model: host 0.982, engineered 0.895, 5-class 0.65
-
Replace with RC-folded multi-order spectrum (k=4,6,8) model: host 0.982, engineered 0.895, 5-class 0.65
-
Replace with RC-folded multi-order spectrum (k=4,6,8) model: host 0.982, engineered 0.895, 5-class 0.65
-
Add measured evaluation results
-
initial upload
-
initial commit
