The product repo tracks the paper PDF and this concise artifact manifest. Full raw benchmark directories should be attached to the corresponding GitHub Release so the Git history stays clean.
Paper artifact release: v0.9.0-product-runtime
Product validation release: v1.1.0
Paper DOI: 10.5281/zenodo.21323877
| File | Bytes | SHA256 |
|---|---|---|
research/SCM_Final_Paper.pdf |
599760 | e7a700537adf946deddaacfdb77a3793b247507290c8269ca8b5c8db810efa6e |
| Asset | Bytes | SHA256 | Source directory | Evidence role |
|---|---|---|---|---|
scm-alb-v02-canonical-1500.tar.gz |
741981 | 16ed4ef250fd1ee57a6e61b3719d24c5f46c4d67ae655d27c7cf046ae938fad8 |
research/benchmarks/alb/results_v02_50p_5seed/ |
1500-run canonical ALB matrix. |
scm-alb-v02-reference-4000.tar.gz |
1846971 | 0eb041d0d93418646d96375076ccf587846d1d7e3438f83e3ea47ac66ff82b53 |
research/benchmarks/alb/results_v02_50p_5seed_with_refs/ |
4000-run reference-adapter matrix. |
scm-ablation-1200.tar.gz |
528185 | 378e31f4beebb0b92337681464979f027b95294f223d93cab670893c930e6126 |
research/benchmarks/alb/results_scm_ablation_20p_5seed/ |
1200-run component ablation matrix. |
scm-mem0-official-500.tar.gz |
13359936 | f256f3a182dbc68d45cd3d9074a56873a28de5f388f0272ea8e9a518ef0db369 |
research/benchmarks/alb/results_mem0_official_openai_v02_50p_5seed_plan/ |
500-row official upstream Mem0 external baseline. |
scm-real-llm-openai.tar.gz |
2288 | b052e1fee170f89b606937daf48e0cf346f9df0b913c99c9773c9a4b992aca44 |
research/metrics/real_llm/ |
OpenAI gpt-5.4-mini real-LLM brutal artifacts. |
| File | Rows / status | SHA256 |
|---|---|---|
results_v02_50p_5seed/MANIFEST.json |
manifest | f76c67e3091000f4f1adac7eeb90e52d17ef55fd0bfa4c1f3bd68538d7e86b4a |
results_v02_50p_5seed/aggregate.json |
1500 / complete | 062d07e4b8b12482a359eab3ef38475209043c307aceae4317b054c556455797 |
results_v02_50p_5seed_with_refs/MANIFEST.json |
manifest | a3a6a09de7af66188a0156b6cae10783aa858fc93ac2f9246d1994e643d11785 |
results_v02_50p_5seed_with_refs/aggregate.json |
4000 / complete | c26a60861ff6757ae74b138af5772ee7303fe0a1bfb80d2ea00fb25afb45b2da |
results_scm_ablation_20p_5seed/MANIFEST.json |
manifest | 598020873db2114930d4bbf035a3653d4cfd19a83a64109e3b4728b31534dab1 |
results_scm_ablation_20p_5seed/aggregate.json |
1200 / complete | 953893f92879bd9335792bc63b0250e146c114e447e02ed708ffd479e797143c |
results_mem0_official_openai_v02_50p_5seed_plan/MANIFEST.json |
manifest | 68952d05e24c47bdb12d1c2599626140410f6cdbdb5e29513cc112ee74033f0c |
results_mem0_official_openai_v02_50p_5seed_plan/aggregate.json |
500 / complete | f91308cc3052162123d86485ec48faaba31718b0d2ea3a8d391124fee2cb2890 |
results_mem0_official_openai_v02_50p_5seed_plan/RUN_STATUS.json |
complete matrix, zero crashes | 4ed960a19ab834205f67997213a5f5be903b958b270b6213047e285854bed6af |
results_mem0_official_openai_v02_50p_5seed_plan/CRASHES.jsonl |
0 bytes | e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 |
real_llm/brutal_real_llm_openai_latest.json |
6/6, pass_rate=1.0 | dab4a17221f9efe898baf3b6582190b56507fc062119d84479c58364e9cde94c |
These artifacts support the paper's lifecycle-memory claim: SCM is useful when memory must transform during idle time. They do not claim universal superiority over all retrieval systems on all memory workloads.
The v1.1.0 release adds
scm-v1.1.0-attribution-evidence.tar.gz, a credential-free aggregate archive
containing the tracked machine-readable attribution summary. It records 2,400
audited paid-model responses, zero crashes, 100% cross-user isolation, and no
shared-memory increase in false self-attribution under the tested protocol.
Raw paid responses are intentionally excluded. See
Safety Validation for the scope and limitations.