framework
Open source eval framework named alongside Inspect AI in the post's framing but not exercised or version-pinned in this installment.
ArgoCD now hands diff calculation to the API server and accepts OCI registries as first-class sources. The same write-up still calls Git the sole source of truth, so teams pulling charts from Harbor have to decide which system records intent.
Reality
Berkeley's RDI center attacked the step where each benchmark computes its score, and without solving a task its own scorecard reports 100% on five of the eight, about 98% on GAIA and 73% on OSWorld.
Reality
A 300-trial study swapped Goose, OpenCode and OpenHands-SDK under Qwen 3.6 Plus and MiniMax M2.5, and reports that the scaffold sets tokens per solved task and the failure mode while the score barely moves.
Reality
The credential was in history[].created_by, expanded there by a RUN command that took a build argument, inside a Harbor project anyone could pull anonymously. Baseten rotated it about 17 hours after the disclosure email.
Reality
An RKE2 hub-and-spoke design runs Prometheus, Harbor, Vault and ArgoCD once for four clusters and documents in-cluster failover carefully, while the hub's own sizing and outage behaviour stay unwritten.
Reality
An arXiv tracing study of Claude Code agents on Gemma and Qwen measured prefix-cache hit rates between 84.6 and 99.5 percent, which moves the serving bottleneck to how long you can keep KV blocks resident between tool calls.
Reality
Dynamic 3.0 ships Qwen3.8-27B GGUFs from 6.2GB up, with an unreproduced accuracy claim attached. The number that matters is the one that decides where the file fits.
Reality
A Google AI series on dev.to shows how Inspect AI turns "is this MCP server worth my tokens" into a measured question, using a cheap grader model and three runs per test.
Reality