Full Fact, the UK's independent fact-checking charity, does not just publish verdicts. It publishes the trail behind each one — where a claim originated, how it moved between outlets, and what has changed about it along the way. That practice, more than any single verdict Full Fact has issued, is the model behind this Labs project: verification as a visible process, not a stamp of approval.
The common response to this problem elsewhere in AI tooling has been to add a chatbot that claims to fact-check on request, answering a query about whether something is true with another generated paragraph of prose. That response treats verification as a conversation to be had rather than as a process to be made visible, and it produces an answer with the same opacity as the claim it was meant to check.
A workbench, by contrast, is built around making the verification process itself inspectable: where did this claim originate, what has changed about it as it moved between sources, what is the actual confidence level behind the number being cited, and what would need to be true for the claim to hold up. The output is not a verdict delivered with false confidence — it is a trail a person can actually follow and disagree with, the same standard Full Fact holds itself to publicly.
This matters most for exactly the kind of writing this publication tries to produce: analysis that depends on getting specific facts right, where a single unverified number can undermine an otherwise sound argument, and where the honest answer to 'is this true' is sometimes 'we don't know yet' rather than a confident yes or no.
A workbench that quietly skipped the trail to save its users a click would be building exactly the chatbot-verdict tool this project exists to replace. The harder, slower version — show the sourcing, show the gaps, let the reader disagree with the reasoning — is the only version worth shipping, and it is the only version we are willing to put a Duck Times byline behind.