From the source
Apple researchers introduced SCLATE, an execution substrate for continual-learning agent training and evaluation that allows benchmarks and unmodified agents to share a single event scheduler via an adapter.
- The system compresses month-long scenarios into hours using a hybrid simulated clock.
- The researchers ported seven benchmarks to SCLATE and compared ten unmodified harness and memory configurations on ten models, finding that an added memory system does not reliably beat the harness's native memory and that models differ in how they use the same harness and memory.
- Post-training Qwen3.5-4B through unmodified harnesses and memory systems resulted in the model reading 6.8x fewer file lines with a 16.7-point higher SWE-bench Verified pass rate and writing richer memory records, while held-out MetaClaw accuracy rose by up to 11.8 points.