Organization’s audit of Federal Court decisions finds dozens of inaccurate citations

The Federal Court’s Chief Justice says the court will review the flagged cases

Organization’s audit of Federal Court decisions finds dozens of inaccurate citations
Tom Macintosh Zheng
By Jessica Mach
Aug 12, 2026 / Share

An analysis of nearly 2,500 Federal Court decisions published between January 2025 and August 2026 found that dozens contain incorrect neutral citations, the codes that courts assign to individual rulings for cataloguing purposes.

The analysis was conducted by a team at Courtready.ca, which provides legal research tools. Tom Macintosh Zheng, a former commercial litigator and co-founder of Courtready.ca, flagged the 38 errors, which were found across 38 decisions, to the Federal Court in a letter on Tuesday. The letter includes a list of the errors, which involve case names that are matched with the wrong neutral citations. 

While Macintosh Zheng emphasizes that 38 errors across thousands of decisions represents a very high accuracy rate, he told Canadian Lawyer there were several reasons his organization wanted to bring the mistakes to the Federal Court’s attention.

First, the inaccuracies could lead to confusion for self-represented parties and errors for third-party research databases that extract neutral citations from new decisions, he says. Second, given the breadth of recent stories about AI hallucinated case citations in the media, individuals who recognize these errors might mistakenly attribute the mistakes to the misuse of AI tools by the court.

The Federal Court published AI guidelines last year indicating it will not use AI – specifically automated decision-making tools – to help make judgments and orders without first consulting the public. The guidelines follow other notices on AI that the court issued in 2023 and 2024.

“What we don't want to happen is for anyone to look at an innocent mistake and make a judgment that somehow the court used AI when the court did not,” Macintosh Zheng says.

Finally, he argues that citations “are really a source of trust.”

“It is telling the parties… I’m ruling this way because other cases have ruled this way,” he says. “And if those other cases are incorrectly cited, then the trust in the system, in my view, is slowly being eroded.”

In response to Macintosh Zheng’s letter, Federal Court Chief Justice Alan Diner told Canadian Lawyer the court will review the cases and make the necessary corrections.

The thousands of Federal Court decisions analyzed by Courtready.ca come from Access to Algorithmic Justice, a joint project by York University’s Osgoode Hall Law School and Toronto Metropolitan University’s Lincoln Alexander School of Law that provides free access to more than 200,000 court rulings across Canada’s courts. The project is not affiliated with Courtready.ca.

Macintosh Zheng says Courtready.ca built an index of every Federal Court decision published between January 2025 and August 2026, as well as their corresponding neutral citation. The organization used a custom tool to scan the full text of each decision and compared the name of each cited case – for example, Tom v. Jerry – against the neutral citation it was paired with. The organization’s team then manually read both the decision itself, and the decision that the neutral citations actually led to.

The organization will conduct the same audit of decisions released by the Supreme Court of British Columbia next due to the large volume of rulings available via A2AJ, and plans to look at other courts’ decisions in the future.

“The fact that we were able to do this [analysis] shows that this is not a technology that is very difficult to [use],” Macintosh Zheng says. “We do encourage the courts to look into whether there are more technologically enhanced ways to carry out such exercises before a decision is released to the public.”

Related stories

Federal Court denies application under s. 41 of the Privacy Act for prematurity Canadian Lawyer survey: How Canada’s courts are regulating, using, and evaluating generative AI