Every claim checked. Every receipt kept.
Three AI reviewers, one trail of receipts. The system was not asked to guess. It read the data, ran the queries, and refused any claim the data could not support. The verdict is below.
What the labels mean.
The same convention appears on every accountability check.
The table or dataset the check read. Source names are clickable where they appear, so a reviewer can jump back to the review-lead index.
Whether the data fields were verified, the query ran, a prepared dataset is still needed, or an outside source is required.
The receipt: the exact query, how many records came back, how long it took, and a fingerprint another reviewer can re-run.
Three reviewers. Three jobs.
Each AI is used for the job it is good at, then checked against the others. The discipline is in the hand-off, not the model.
First reviewer
Reads each accountability question and drafts a research brief. Eleven questions, all at once.
Second reviewer
Connects to the Alberta database. Checks every reference, runs the query, and captures the receipt: runtime, record count, and a fingerprint another reviewer can re-run.
Final reviewer
Reviews the evidence, writes the explanation, refuses any claim the data does not support, and names what is missing.
Per-check verdict
One row per accountability check. Click to open the deep dive.