v0.1.0 — Research Preview

ABI v0.1.0 — Research Preview

Frozen-Module Domain Transfer Across LLM Architectures — verified experimental results.

What is validated in this release

Claim	Key metric	Result file
T5-large same-backbone ABI reconstruction	top-5 = 0.8725	\cross_arch_t5_nib_v53_results.json\
GPT-2 → Qwen2.5 cross-family (Exp 32)	top-5 = 0.8701	\cross_family_nib_results.json\
T5-large → GPT-2-medium enc-dec → dec-only (Exp 39)	top-5 = 0.8699	\cross_arch_enc_dec_nib_results.json\
T5-large backbone-update invariance (Exp 40)	efficacy = 304.3%	\cross_arch_t5_succession_results.json\

All four NIB criteria pass simultaneously on every validated claim.

Verify in < 5 seconds

\\�ash
git clone https://github.com/Yoder23/abi
cd abi
pip install -e .
python verify_result.py
\\

Expected final line: \Path 2C -- T5-large ABI domain reconstruction -- VERIFIED\

New in this release (v0.1.0 vs initial push)

• Badges, Start Here block, claim ladder, claim-to-file map in README
• \LICENSE\ (Apache 2.0), \CITATION.cff, \pyproject.toml\ (installable package)
• \CLAIMS.md\ — canonical claim map (validated, not tested, explicitly not claimed)
• \SKEPTICS.md\ — direct answers to 9 hard questions
• \FAQ.md, \CONTRIBUTING.md, \ROADMAP.md, \SECURITY.md\
• \Dockerfile\ (pytorch/pytorch:2.1.0-cuda12.1 base)
• GitHub Actions: \�erify.yml\ (runs \�erify_result.py\ on push/PR), \lint.yml\ (ruff)
• Issue templates: reproduction report, new model result, bug report
• PR template with locked-result checklist
• 
  otebooks/ABI_Verify.ipynb\ — Colab-ready verification notebook
• \ ests/\ — pytest suite (NIB threshold regression, package import tests)

What this does NOT claim

• Universal transfer across all architectures and domains
• 7B+ scale (not yet tested)
• Production readiness
• Theorem-level proof of universality

See \CLAIMS.md\ for the complete list.