The July 2026 technical report is backed by local checks, ALB matrices, ablation results, an official Mem0 external baseline, and a gated OpenAI real-LLM brutal artifact.
python -m build --wheel --sdist
twine check dist/*
python -m venv /tmp/scm-smoke
/tmp/scm-smoke/bin/pip install dist/scm_memory-*.whl
cd /tmp
/tmp/scm-smoke/bin/python -c "from scm import SCMEngine; print('ok')"
/tmp/scm-smoke/bin/scm --helpscm serve --host 127.0.0.1 --port 8000
curl -s http://127.0.0.1:8000/health
curl -s http://127.0.0.1:8000/v1/health
curl -s http://127.0.0.1:8000/v1/tools?format=openaiRun from the repository root with the project virtual environment:
venv/bin/python research/scripts/review_publication_artifacts.py --strict
venv/bin/python research/scripts/validate_publication_readiness.py --strict
venv/bin/python research/scripts/run_publication_local_checks.py \
--output-json research/PUBLICATION_LOCAL_CHECKS_2026_07_09.jsonExpected state for the July 2026 report:
- Artifact review:
blocker_count=0 - Readiness:
ready_for_top_claims=true - Local checks: all commands pass and
secret_hits=[]
The external path must read API keys only from environment variables. Do not put secrets in command-line arguments or repository files.
venv/bin/python research/scripts/run_publication_blockers.py \
--allow-paid \
--apply-paper-claimsFull raw benchmark outputs are intentionally distributed as GitHub Release assets, not committed directly to Git. See ARTIFACTS.md for the expected bundle names, hashes, row counts, and evidence role.