Almost every review here is already a complaint

Among the reviews carriers left on CDLScan, 94.8% carry the lowest rating. That figure describes those reviews, not drivers in general: carriers write after a bad run, so the share tells you what shows up once someone bothers to report it. This is the most important number on the site, and it is the easiest one to misuse.

It does not mean almost every driver is a problem. It means this corpus is a record of problems. Every share you read on these pages — abandonment, paid travel, no-shows, late loads — is a share of complaints, not a share of the workforce.

If you walk away from this article with only one habit, make it that one.

What a complaint corpus can and cannot say

It can say which failures carriers bother to write down, and how those write-ups have shifted year to year. Abandoned-truck mentions rose from 18.8% of reviews in 2018 to 40.4% in 2025. That is a change in the mix of complaints. It is a real change. It is not a census of CDL holders.

It can say which failures travel together. Paid travel and a no-show sit far above chance. An abandoned truck and a no-show sit far below it. Those are facts about paragraphs, not about every driver who ever quit.

It cannot say how often a random applicant will fail. Happy employers rarely write. A driver who finished a year, turned in the keys, and got a decent check does not generate a page here. The silence is not a clean record. It is a missing one.

How we count the shares, and why we never publish the size of the corpus, is on the methodology page. Compare percentages inside a year. Do not compare year totals. The pile of reviews grew. A bigger pile is not a finding.

Why the floor is so high

People write reviews when they are still angry and still unpaid for the week they just lost. A truck sitting at a stop, a ticket that was never used, a phone that died on a live load — those generate a lowest rating because the author is documenting a loss, not grading a semester.

That is why 33.5% of the reviews mention an abandoned truck and 32.2% carry verdict language — a recommendation that the next employer not bother. Those two are not the same thing. One is an event. The other is the author’s mood toward you. We keep them apart on purpose.

The 5.2% that are not the floor still are not praise. They are a slightly different mix of complaints, unpacked in not every bad review means the same thing. Walk-offs and no-shows concentrate on the floor. Late loads and damage barely move. Even that contrast is a contrast inside a complaint pile.

I stop trend lines at 2025. 2026 is open. Using a partial year as proof that “things got better” would be the same class of error as treating 94.8% as a driver rate.

What this does to official reports

Nothing directly. PSP, an MVR, and the Clearinghouse do not know about this floor. They store filings. The bias of this corpus is the opposite of their bias. They under-report employment messes because nobody files them. We over-represent employment messes because that is when someone writes.

Read together, they are useful. A clean packet plus a quiet name in the reviews is a stronger sentence than either half. A clean packet plus three write-ups about a left truck is a warning the packet cannot give. A pile of lowest ratings with no event in them is a pile of mood. Read the sentence.

The mistake I see desks make — and I am describing a habit, not a case from my own book — is to treat this site like a score and the official packet like a score, then average them. Neither is a score. One is a filing system. The other is a complaint log. Your job is to know which question each one answers.

How I want you to read every article here

First number, first paragraph: it is a share of reviews. If a page ever sounds like it is talking about all drivers, it has failed, including this one.

Second: look for the companion, not just the headline share. 33.5% abandonment is the start of an argument, not the end of one. Who else is in the paragraph — loaded trailer, silence, a ticket, a test — tells you which screen to run.

Third: do not hunt for comfort in the 5.2%. There is not enough mass there to save a hire, and there is still a 21.0% abandonment mention rate inside it.

Fourth: go read the name. The site exists so you can check a person, not so you can quote a percentage in a meeting. Check it on CDLScan before you spend. It takes less than a minute. The percentage told you the problem is common in what carriers write. The name tells you whether this carrier already wrote it about this person.

I write these pages from data CDLScan shared with me. That is a material tie. It is also why I will keep repeating the ugly number at the top: 94.8% of these reviews are already the floor. If we ever stop saying that, stop trusting the shares.

Frequently asked questions

Does 94.8% mean almost every truck driver gets a 1?

No. It means almost every review in this corpus carries the lowest rating. Carriers write after a bad run. Drivers who finish a job quietly are mostly not in the file.

Then why believe any of the other percentages?

Because they are percentages of the same complaint pile, counted the same way each year. They are good at describing what goes wrong when someone reports it. They are bad at describing the whole workforce. Use them for the first job, not the second.

Is a name with no reviews a good sign?

It is the absence of a complaint, not the presence of a recommendation. Combine it with the packet you already buy. Do not promote silence into a reference.

Why don’t you publish how many reviews there are?

Because the useful facts are shares and years, and the size of the corpus is not something we put on the public site. The method is on the methodology page.

Should I ignore reviews that are only a lowest rating and a warning?

Ignore them as events. Do not ignore them as heat. Verdict language is the author’s advice to you, not a description of what the driver did. Go find an incident sentence before you treat it as a walk-off.

Do PSP or an MVR correct for this bias?

They have a different bias. They miss unfiled employment messes. This corpus over-represents them. Read both. Do not average them.

What is the first thing I should do after reading this?

Pick a live candidate and look the name up on CDLScan. The floor is a warning about how to read. The name is the work.