# Rachel Sandoval, Founder at Cloudbyte Systems — read of LexiRisk, June 16 2026

> "11 years in B2B SaaS, currently running an 8-person shop building inventory tooling for small manufacturers. I sign vendor agreements, contractor SOWs, and the occasional partnership doc roughly once a month."

## How I got here

Someone in the Indie Hackers Slack dropped this link in the #tools channel with the comment "anyone tried this?" No context, no endorsement. I had 12 minutes before I needed to pick up my kid from soccer so I opened it on my phone. That's about the most charitable condition this page is going to get from me.

## What I clicked first

The headline "Stop Signing Blind Contracts" landed. I actually said "yeah" out loud. The subhead "LexiRisk reads your contracts, flags the risky clauses, and explains what you're really signing" is about as clear as this category gets. I didn't have to guess what the product does, which is already better than two-thirds of the tools I've looked at this month.

The pain section also worked on me more than I expected. The "The Damage Compounds Later" scenario about the uncapped liability clause was specific enough to feel real, not generic. I've signed something like that. It stuck.

## Where I paused

Annual plan. It lists "AI training on your contracts" as a feature. But two sections earlier, the FAQ says: "We don't train on your data or share it with anyone. Privacy is not negotiable."

Those two things cannot both be true. Either you train on my contracts (annual plan) or you don't (FAQ). I read it twice. I'm not misreading it. That is either a copywriting error or it's a signal that nobody on the team actually proofread this before shipping. Neither answer makes me feel better about uploading a signed term sheet.

## What I distrusted

"LexiRisk flags 95%+ of common dangerous clauses." That number is doing a lot of work with no support. 95% of what? Measured how? Against which corpus? "Common" is doing even more work. The sentence sounds like a stat but it's a claim without a denominator.

Also: "Our ML model scores every clause as high, medium, or low risk." Is this GPT-4 with a system prompt? A fine-tuned model? Something the team actually built? "ML model" in 2026 tells me nothing. Every wrapper calls itself an ML model.

And I cannot find a single example of the actual output. Not one screenshot. Not a redacted PDF sample. Not a "here's what a risk report looks like." I'm being asked to upload a confidential business document to a product I cannot actually see work. That's a hard ask.

## What would convince me

One real report. Redacted, anonymized, whatever you need to do. Show me what the flagged output looks like on an actual SOW. Not a demo video with fake contract text. A real clause, a real risk score, the actual plain-English explanation the AI produces.

One named customer I could look up. Not a logo. A name, a company, one sentence about what they caught. Even a Reddit comment from a real founder who ran a contract through it and said "it flagged X and they were right."

Fix the training contradiction. That alone would bring my trust up a level.

## What I'd ask in an email reply

1. The annual plan lists "AI training on your contracts" but the FAQ says you never train on user data. Which is accurate, and can you point me to a data processing agreement I could read?

2. What does the actual report look like? Can you send me a sample output on a generic NDA or contractor agreement?

3. Is this analyzing contracts under US law only? I have one vendor in Ontario and another in the UK. Do those just get skipped, or does the model take a swing anyway and I just won't know where it's wrong?

## Verdict: on-the-fence

The concept is right and the copy is sharper than most. But the training/privacy contradiction is a genuine red flag for a product that's asking for my confidential docs, and without a single example of real output, I have no idea if the analysis is useful or just pattern-matched boilerplate. Fix those two things and I'd probably pay $19 to try it.

---
*Memo by skeptic persona, generated 2026-06-16. Studio breaks own self-grading loop.*