How we produce the Signal.
The Signal is our weekly read on staffing, AI, and the rules around both. We produce it through the same discipline we build into Cassion: several independent inputs, adversarial checking, and a human accountable at the end.
The process.
Five steps, run every week, in this order.
Parallel research.
Several independent frontier AI models, run side by side.
Each week starts with one research brief run through several of the leading frontier AI models at once. Running them side by side surfaces what any single model misses and exposes where they disagree, which is usually where the interesting questions are. The set of models is not fixed; we expect it to change as the tools do.
Scoring and convergence.
One rubric, applied to every research pass.
We score each research pass on the same rubric: in-window coverage, factual accuracy, source quality, and honesty about what it could not verify. Then we converge the passes into one working source, keeping the strongest material and discarding anything stale or unsupported.
Adversarial review.
A deliberate attempt to break the draft.
The draft goes through adversarial review, a deliberate attempt to break its claims, its framing, and its sourcing. Then we audit the challenges themselves before accepting them, because a confident correction can also be wrong. We check the fix, not just the finding.
Verification.
Load-bearing facts checked against primary sources.
Facts, dates, figures, and rankings that carry weight are checked against primary sources: the agency release, the company filing, the regulator’s own text. When we cannot verify something, we say so or we leave it out. We do not round a source’s “one of the largest” up to “the largest.”
Human editing and accountability.
A person owns what goes out.
A person edits and approves every Signal before it publishes. The AI does the gathering and the first drafting. A human owns what goes out.
AI and people both make mistakes. This process is built to catch them, not to pretend they do not happen.
Why we show this.
We build governed, audited AI for staffing operators. It would be strange to produce our own writing any other way.
Found a mistake.
If something in a Signal is wrong, tell us. Use the interest form, select “Ask a question,” and include the piece and what you found. We treat corrections as priority.
Tell us.Last reviewed: July 6, 2026.