Promises, not features
Most software tells you what it can do. That list is easy to write and impossible to check. This is the other list. Every line is something Cal refuses to do, and every line is something you can hold us to.
It will tell you what the guideline says, which section it came from, and when that section took effect. What it will never do is make the call. You hold the license. You sign the file. A tool that says "approved" is a tool that has quietly moved your liability onto a server, and no software should be allowed to do that.
Cal returns a verdict of SUPPORTED, OVERREACH, or AMBIGUOUS on a guideline position. It does not return APPROVED.
When a document does not state its own effective date, most systems stamp the day they imported it. It looks tidy. It is also a lie, and it is the kind of lie that gets caught in an audit rather than in a demo.
Cal stores nothing instead, flags the record, and tells you the date is unconfirmed. You will see that badge more often than you would like. That is the point.
From the ingest code, verbatim: "we must never present it as one. Only a genuine, strict ISO effectiveDate counts as verified; anything else is UNVERIFIED and we store null and a flag rather than fabricating today's date."
This one is worth being blunt about, because it is common in this category and it is usually invisible. Several tools that match your scenario to a lender are free to you because a lender is paying on the other side. That payment has to show up somewhere, and it shows up in the order of your results.
Cal has no lender-paid tier. Not a placement fee, not a lead fee, not a sponsored slot, not a revenue share. The order you see is the order the guidelines produced. If that ever changes, this page changes first and in public.
There is no code path in Cal by which a payment can alter a match result, because there is no lender billing at all.
Not to a lender, not to a marketing partner, not to an aggregator, not as anonymized and enriched anything. Your pipeline is not our product. It is difficult to overstate how much of the mortgage technology business model is quietly the opposite of this sentence.
The affordability calculator does not have a database connection. Not a policy about one, not a promise to delete later. It was built without the ability to write, so a visitor's income and debts cannot be retained even by accident, even by a future change we did not think through.
Privacy that depends on us remembering to behave is not privacy. Where we can make the wrong thing impossible instead of forbidden, we do.
From the source: "Statelessness is the privacy design: registerCalQualifyRoutes takes NO db."
No invented advisor with a headshot. No "your dedicated specialist" that is a prompt. When a borrower is talking to software, Cal says so, and when the conversation turns into advice it hands off to a licensed human. That is not a courtesy. It is what the license means.
We run a nightly regression suite. We could quote the score. We would rather publish the methodology, including the parts that flatter it, and let you decide what the number is worth.
Every product in this category claims to be accurate. Almost none of them will show you a question they got wrong.
See What Cal got wrong, which lists our misses, what we changed, and where the evaluation is weaker than the headline suggests.
What that leaves
Cal reads the guidelines so you do not spend your afternoon on hold. It shows its work, it names its source, and it stops at the edge of your license. That is a smaller promise than most software makes. It is also one we can keep on the day someone checks.
Chris Black built Cal after twenty years originating loans. Every refusal on this page exists because he watched the alternative cost somebody a file.
NMLS 275073 · Alamo, California