| title: Mnemosyne Memory Lab | |
| emoji: 🧭 | |
| colorFrom: gray | |
| colorTo: indigo | |
| sdk: static | |
| app_file: index.html | |
| pinned: false | |
| # Mnemosyne Memory Lab v0.2.1 | |
| Compare how memory architectures handle changing beliefs, stale procedures, | |
| duplicated evidence, privacy boundaries, and historical reconstruction. | |
| Forty deterministic synthetic scenarios. Five memory strategies. All | |
| processing runs locally in your browser; uploaded files are not transmitted or | |
| persisted. | |
| This is synthetic fixture behavior, not production memory accuracy. Twenty-five | |
| of 40 scenarios are currently non-discriminating; no production superiority is | |
| claimed. Local imports stay in the browser. | |
| Dataset: https://huggingface.co/datasets/solsticestudioai/mnemosyne-memory-lifecycle-benchmark | |
| Collection: https://huggingface.co/collections/solsticestudioai/solstice-agent-reliability-lab | |
| Methodology and validity: https://huggingface.co/datasets/solsticestudioai/mnemosyne-memory-lifecycle-benchmark/tree/main/docs | |
| Validity summary: https://huggingface.co/datasets/solsticestudioai/mnemosyne-memory-lifecycle-benchmark/blob/main/reports/benchmark-validity-summary.md | |
| Tooling v0.2.1 adds a provider-neutral external strategy protocol. Load a local | |
| result bundle with “Compare your memory system”; no hosted memory upload is | |
| performed. | |