enigma-memory v0.1.5

Official standard benchmark harness release.

Install:

npm install -g enigma-memory

Benchmark datasets:

npm run benchmark:datasets -- --dry-run
npm run benchmark:datasets -- --execute --dataset all --out-dir .enigma/benchmarks/datasets --manifest .enigma/benchmarks/dataset-manifest.json
npm run benchmark:standard -- --locomo .enigma/benchmarks/datasets/locomo10.json --longmemeval .enigma/benchmarks/datasets/longmemeval_s_cleaned.json --max-locomo-qa 25 --max-longmemeval-items 25 --top-k 5 --out .enigma/benchmarks/standard-memory-benchmark-sample.json

Highlights:

• Adds official LoCoMo downloader support.
• Adds LongMemEval oracle/S/M downloader support with streaming large-file downloads.
• Adds public-safe retrieval/evidence proxy benchmark runner.
• Adds enhanced deterministic Enigma relevance over simple keyword filtering.
• Adds reproducibility docs and optional CI example.

Observed sample result:

• LoCoMo sample hit@5: keyword 0.28, Enigma relevance 0.52.
• LoCoMo sample exact coverage: keyword 0.28, Enigma relevance 0.36.
• LongMemEval-S sample estimated tokens: keyword 60,969, Enigma relevance 55,689.
• LongMemEval Oracle sample estimated tokens: keyword 54,066, Enigma relevance 46,939.

Verification:

• standard benchmark downloader/runner tests pass
• datasets downloaded with SHA-256 manifest
• standard sample reports generated
• npm run check
• npm test
• npm pack --dry-run
• npm publish --access public --dry-run
• npm publish --access public
• npm run registry:verify -- --execute
• registry install smoke
• GitHub CI passed on Ubuntu and Windows

Boundaries:
Retrieval/evidence proxy only. No LLM answer accuracy, provider deletion proof, model forgetting proof, ROI/savings guarantee, compliance certification, third-party competitor score, or benchmark-leadership claim.