Why BeforeJD Checks Every Read
A read of a contract is only as good as the check on it. BeforeJD has a second reader check every read, and every quoted finding is checked against your contract.
Most AI contract review tools feed your document into a single large language model, ask it to find problems, and return whatever it generates. That approach is fast and cheap. It is also fragile in ways that matter. One pass has to understand the document, judge the risk, and check its own work all at once, with nothing to verify the result.
BeforeJD checks every read before it reaches you, and every quoted finding in the final report is checked against your document.
The Problem with a Single Pass
Too Many Jobs at Once
Understanding what a contract says and judging whether it is risky are different tasks. When one pass tries to do both at the same time, it tends to take shortcuts. It reacts to high-profile clause types like indemnification and non-competes, and it can gloss over quieter provisions like audit rights, data retention, or force majeure definitions.
The Hallucination Problem
Large language models hallucinate. They generate confident, plausible-sounding analysis that is factually wrong. In contract review, hallucination takes specific forms: citing clauses that are absent, attributing language to the wrong party, generating risk scores that drift from what the contract actually says. When a read has no check of its own, those errors go directly to the user.
General-purpose AI chatbots are especially exposed here. As explored in our post on why general AI falls short for contract review, a tool that was built for general conversation lacks the structural guardrails that legal documents call for.
A Checked Read
Every Read Is Checked
A second reader checks every read before it reaches you. The rules applied are written for each of the 18 contract types BeforeJD reviews, from employment agreements to SaaS terms of service, so a lease is read as a lease and an NDA is read as an NDA.
Every Quotation Is Checked
Checked Against Your Document
When a report shows you the contract’s own words, those words are checked against your document. Anything that cannot be matched is never shown as the contract's words.
We measured this across 265 public contracts and two full runs. About 95 percent of 1,260 high-severity quotations matched the contract word for word, and none was fabricated. The few that could not be matched are never shown as the contract's words (how it was measured).
The Evidence: CUAD Benchmark Results
What the Numbers Show
The CUAD dataset is a peer-reviewed legal benchmark produced by the Atticus Project, a multi-institution academic collaboration. It contains more than 13,000 clauses marked up by legal experts across 41 categories. It is the closest thing the field has to a ground truth for contract analysis.
Tested on a public benchmark of more than 13,000 expert-labeled clauses, it finds about nine in ten material risks (how it was measured). That performance represents the balance between catching real risks (recall) and avoiding false alarms (precision).
The story of how BeforeJD came together, including the challenges and surprises along the way, is told in how BeforeJD was built.
What Benchmarks Leave Out
Benchmarks measure accuracy on annotated datasets. They leave out how clearly a risk is explained and how ready a proposed redline is to send. The CUAD result describes how many material risks the read finds, and a read can still miss a risk. It is a first read to build on, made to work alongside your attorney.
What This Means for Your Contract Review
Practical Implications
When you upload a contract to BeforeJD, a second reader checks your read, and the findings that result are checked against your contract before they reach the report.
The design goal is a read that is less likely to miss a real risk and less likely to invent one. It takes a few minutes, where a single pass takes about 30 seconds. In contract review, care is worth the wait, and the extra time goes to checks a single pass skips.
The Tradeoff Is Worth It
A checked read costs more to run than a single pass. It is the right trade, because a missed risk can be financial, legal, or career-altering, and analysis you can trust is worth more than analysis you get in thirty seconds.
We built BeforeJD to produce results worth acting on, with every quoted finding checked against your contract.
Every AI read has blind spots. Checking the work and verifying every quotation against the document is how BeforeJD compensates, and why it checks every read.
See the difference for yourself. Upload any contract to BeforeJD and get a risk analysis in minutes.
Questions people ask
- Why does BeforeJD check every read?
- One pass tends to take shortcuts when it has to do every job at once, with nothing to check the result. BeforeJD has a second reader check every read, and every quoted finding in the report is checked against your document.
- What does a BeforeJD read check?
- The read weighs each clause against rules written for each of 18 contract types, and a second reader checks it before the report is written. Every quoted finding is checked against your contract, and anything that cannot be matched is never shown as the contract's words.
- How does BeforeJD check the quotations in a report?
- Every quotation in a report is checked against your document, and anything that cannot be matched is never shown as the contract's words. Across 265 public contracts and two full runs, about 95 percent of 1,260 high-severity quotations matched the contract word for word, and none was fabricated.
- How accurate is BeforeJD on the CUAD benchmark?
- Tested on a public benchmark of more than 13,000 expert-labeled clauses, it finds about nine in ten material risks. The benchmark is CUAD, a peer-reviewed legal benchmark produced by the Atticus Project, with clauses marked up by legal experts across 41 categories. False alarms are measured too.
- Is a checked contract read slower than a single pass?
- Yes. It typically takes a few minutes instead of thirty seconds, and it costs more to run. Contract review rewards care over speed, and the extra time is spent on checks a single pass does not make.