x.klickd v4.1.0: Portable, Encrypted and Governed Memory for Long-Running AI Workflows

x.klickd v4.1 is an open portable, encrypted and governed memory architecture for long-running AI workflows.

This DOI evidence pack includes scientific benchmark analysis, behavioral analysis, automatic quality scoring, benchmark data, scripts, metadata, and claim-boundary documentation for the v4.1 controlled long-project protocol.

The completed four-bundle benchmark reports 7,189 valid outputs over 7,200 expected outputs, with a 99.85% completion rate. In this protocol, static x.klickd reduced repeated input-token overhead by approximately 76.49% compared with prompt-history memory. With optional compressed_memory, the reduction reached approximately 93.34%.

The benchmark also includes automatic long-project quality scoring. Under this benchmark-specific rubric, x.klickd conditions achieved a mean score of 86.24/100, compared with 58.51/100 for standard AI usage without x.klickd (no portable memory file: prompt-history memory, project documents, or no memory).

Bundle 5 (b05_drone_mission_ops) is documented separately as a provider quota/rate-limit stress scenario and is excluded from the completed-bundle aggregate.

These results are benchmark-specific. They do not claim universal native support across all AI systems, automatic GDPR or EU AI Act compliance, or superiority over external public memory benchmarks such as LongMemEval or MemoryArena.
Future work includes external comparative evaluation.