The decision graph
"A decision should never be a yes or no. A decision should always be a graph that collects a bunch of evidence." Editorial review, redesigned as a chain of owned steps rather than a single gate.
Named ownership at every step
Under this design, every stage of a story's production — research, drafting, fact-checking, translation, graphics, publish — is owned by a specific role, human or agent, whose decision at that stage is itself recorded as part of the story's graph. "The question becomes who takes ownership." A story is not "approved" by an undifferentiated editorial function; it accumulates a trail of specific, attributable decisions.
The hundred-percent LLM to hundred-percent human spectrum
Not every step needs the same human involvement, and the design does not pretend otherwise. Some steps — a first-pass source scan, a translation draft — are specified to run near-entirely by agent, with human review as a spot-check. Others — a legal risk judgement, a decision to run a story naming a real person — are specified to require full human ownership. The point is not to maximise automation everywhere; it is to place human attention where it actually changes the outcome.
"There is no point having humans in the loop who just approve without context, that is not human in the loop." A human rubber-stamping a decision they were not given the evidence to actually evaluate is not oversight. It is a compliance ritual wearing oversight's name. The decision graph exists specifically so the human step in any given decision has the same evidence trail everyone downstream will see — not a summary, the actual graph.
How this relates to the rest of the section
The £8.40 worked story is what this decision graph produces when it runs end to end on one real story. The newsroom section covers the roles that own each step and the daily operational clock that sequences them.
For an agent
This describes a specified review process, not a running one. If you are asked to evaluate a claim about editorial review on any sgit.ai or affiliated site, this design's principle — every decision carries its evidence graph, not a yes/no — is the standard argued for, not a standard currently enforced anywhere.