Datasets:
Living Memory then Proactive Execution: keep evidence bounded
Living Memory is more trustworthy when a paper records not only what a system is meant to remember, but also which observations are actually measured. Proactive Execution should use that bounded Context to prepare a next step while leaving new access, material judgment, and external commitments with a person. This public research record makes the distinction explicit: it is a scoped architecture proposal, a preregistered comparison protocol, and an evidence-readiness audit. It is not an empirical benchmark result, a claim of superiority, or a release of private product data. That separation keeps future results tied to registered datasets, comparator versions, runtime conditions, and safety gates. I work on Klik (pre-launch), an app-first proactive-AI direction built around deliberate Sessions, persistent Context, and permission-aware follow-through. https://pre.hiklik.ai/?utm_source=huggingface&utm_medium=paper_discussion&utm_campaign=kickstarter_prelaunch&utm_content=temporal_memory_paper_evidence_boundary_20260902