The seller says the software does X.
Does it?
We check what a piece of software says about itself against what its code actually does. Every finding names the file and the lines that settle it, so you can look for yourself. The same method serves a developer weighing an open-source dependency and a buyer weighing a company.
Read a finished review Talk to us
Three ways in
Try it
Free
Read a completed claims review of a small open-source project and ask it questions. If you have a repository of your own to evaluate, write to us and we will set you up to run one.
Claims review
From $500 · quoted after intake
Every claim in the documents you name, tested against the code, independently reviewed, and rated against what matters to you. A report you can check line by line, and a page where you can keep asking.
Attested due diligence
Quoted per engagement
The claims review, plus a reading of the whole record by someone who has sat on both sides of a deal. A signed opinion, and a person on call to your deal team while it matters.
Who it is for
Developers choosing between open-source projects, where the README is most of what you have to go on. Advisers and buyers of software businesses in the $1M to $5M range, at the point where the seller's description of the product is the main thing on the table. A claims review sits a rung below a full technical due diligence: narrower, quicker, and priced so it can be done before the letter of intent rather than after.
How it works
A short conversation about what you are evaluating and why: what you already know, what would change your mind, and, for a deal, what would move the price or end it.
The claims are listed and frozen before any evidence is read, and then each one is tested against the code and the documents supplied.
A second, independent pass re-reads the evidence behind every finding. Each gap is rated against your thresholds, twice, and looked at again when the two readings differ.
You receive the report and a private page where you can ask about any finding in your own time, with the working record behind every answer.
Why it holds up
The reading is done by language models, but not by asking one a question and taking its word. The models work inside a fixed procedure that keeps the list of claims separate from the verdicts, keeps every search and every file opened, and refuses a finding it cannot cite. A person reads the result and stands behind it. The report says on every line which part was machine and which was judgment.
Machine drafts. A person confirms and signs. The report tells you which is which.