If the workflow is part of the evidence, what gets published?

Northwestern's Agentic AI Investigative Journalism Challenge listed July 15 as its closing date. It asked journalists, developers, and technologists to use Claude Code with Agent Skills to investigate a shared public-interest corpus. 1

That corpus combines U.S. House and Senate lobbying disclosures with Congressional press releases from 2022 through March 2026. Teams submit four connected pieces: the reusable Agent Skill(s), a findings report, full interaction traces, and a README that maps the work. 1

The unusual requirement is the trace: logs of inputs, tool calls, outputs, and the moments when human judgment intervened. For claims about named people or organizations, the rules require evidence, traceability to source records, and human review before submission. 1

That changes what "agentic" means here. The agent is not presented as a magic reporting brain. It is a reusable method whose steps can be inspected, challenged, corrected, and carried into another investigation. That is an editorial reading of the submission design, not a claim that the challenge has already proved autonomous reporting. 2

Northwestern says the goal is to support watchdog journalism rather than replace reporters, using agents to suggest leads, patterns, and connections in large document sets. 3

Discussion: When the skill and the trace travel with the finding, are they part of the reporting method, the published artifact, or both?