Workplace conditions at Tesla’s California factory have been scrutinized for months, after an employee in February , sparking a renewed effort to unionize the Fremont facility by the United Auto Workers. And claims the automaker’s own internal data shows that injury rates at Fremont exceed what’s found across the industry.

Tesla has , asserting that its safety record is “much better” than the industry average. The report, published by California-based nonprofit Worksafe, addresses several claims made in a Tesla blog post earlier this month, in which the company preemptively attempted to pushback against news stories about workplace conditions at the Fremont shop. “A few anecdotes in a factory of over 10,000 people can always be given,” , “but these are the facts.”

In particular, the post highlighted the factory’s Total Recordable Incident Rate (TRIR)—which it said is the leading metric for workplace safety—through the end of the first quarter of 2017. Tesla’s rate is 4.6, the automaker said, more than 30 percent better than the industry average of 6.7.

When it comes to earlier years, the “facts”—as Worksafe’s analysis indicates—aren’t as spectacular. Worksafe based its report off Tesla’s OSHA Form 300, which logs work-related injuries that companies must provide employees upon request. Worksafe says employees first requested copies of the logs in April. What the OSHA forms showed is that Tesla’s TRIR rate in 2014 and 2015 outpaced the industry, at 8.4 and 8.8 respectively, the latter more than 30 percent higher than the industry average.

And perhaps that explains the earnest, forward-looking phrasing of a statement given to Jalopnik by a Tesla spokesperson in response to the report.

“We may have had some challenges in the past as we were learning how to become a car company, but what matters is the future and with the changes we’ve made, we now have the lowest injury rate in the industry by far,” a Tesla spokesperson told Jalopnik by email. “Our goal is to have as close to zero injuries as humanly possible and to become the safest factory in the auto industry.”

But Worksafe’s report said Tesla’s injury data recorded through Q1 “is too preliminary to be considered accurate given Tesla’s somewhat erratic reporting patterns.” One example of the reporting patterns can be found in how Tesla handled its 2016 injury data, the report said. On February 1, according to the report, Tesla reported 705 injuries on its 2016 OSHA 300 log. But three months later, it revised the report, which showed 840 reportable injuries.

Amendments were also made to the 2017 logs, Worksafe said:

In April, Tesla provided a log showing 100 injuries, including 22 lost time injuries resulting in 164 lost work days and 80 restricted duty injuries resulting in 1852 restricted duty work days. A few weeks later Tesla produced a revised log for the same period showing 146 injuries, including 40 lost time injuries resulting in 632 lost work days and 96 restricted duty injuries resulting in 3,829 restricted duty days.10

While the 2017 data illustrates a reduction in injuries over previous years, “Relying on 2017 injury data to reach any conclusions about safety trends at the plant is premature and could have misleading results,” the report said. That’s why Worksafe asserted that one quarter of data isn’t “a sufficient length of time to accurately identify a meaningful and lasting trend in injury reduction.”

The release of Workplace’s report was publicized by a third-party spokesperson hired by the UAW to assist in coordinating media interviews with pro-union Tesla employees earlier this year.

Last month, the , claiming Tesla was illegally intimidating employees behind the union drive.