Loopwright

Loopwright is a development platform that turns failures in AI-assisted production into quality controls for the next run. Every failure is recorded, aggregated, and — only with human approval — promoted into a deterministic gate or a review criterion that later productions must pass. The demo runs a job-posting production workflow for editorial staff: the source posting is pinned by hash, FACTs are extracted with a Meta model and reconciled against the source, and the run stops on any contradiction instead of generating. At each judgement point, the artifact and a question are delivered to a human; after they answer, the tool itself writes the approval record, resumes the run, and re-verifies the result — input identity, grounding of every sentence, and deterministic page checks. Items the source does not state are never generated as claims. The job postings are the test workload, not the product: the same loop applies to any AI-assisted production whose output can be checked. All data is fictional.