Predictika · Harborview Home Lending pre-qualification

An AI loan assistant's pre-qualification, checked before the letter goes out

An AI loan assistant prepares a pre-qualification for a home purchase: loan program (FHA or conventional), county, credit score, income, debts, price, down payment and verified cash. Before the letter is issued, the assistant sends the application to Predictika, which checks it against a declarative model of the program, loan limit, debt and cash rules, compares the monthly payment and debt-to-income the assistant states with the computed ones, and returns every finding together with the payment breakdown. This is an illustrative model built from published FHFA and HUD limits and simplified guidelines; it is not a lender's underwriting rule set.

Scope: pre-qualification of home purchase loans, FHA and conventional, one to four unit properties. See everything this validator covers

This validator models one process: the pre-qualification check for FHA and conventional home purchase loans, as an AI assistant at a lender would run it before issuing a letter. It handles any combination of program, occupancy, county, score, income, debts, price, down payment, cash and rate that such an application can contain, not only the scenarios shown on this page. It does not cover refinances, VA or USDA loans, jumbo programs or final underwriting.

  • Covered: program eligibility — minimum credit score by program (FHA 500, conventional 620), FHA for a primary residence only, maximum loan-to-value by program, score and occupancy, and the 97 percent first-time-buyer rule.
  • Covered: loan limits — the 2026 conforming and FHA county limits for a typical county, a mid-cost county and a high-cost county.
  • Covered: debt and cash — debt-to-income caps with an underwriting-approval band, the FHA housing ratio, cash for down payment and closing costs, reserves for investment property, and job history.
  • Covered: the payment — principal and interest at six rates (FHA includes the financed upfront premium), property tax, insurance, FHA annual mortgage insurance or conventional private mortgage insurance by loan-to-value and score, and HOA dues, with the payment and debt-to-income the assistant states compared with the computed ones.
  • Not covered — refinances, VA and USDA loans, jumbo programs, automated underwriting decisions, appraisal review, fraud checks and real rate sheets. Other areas can be added.
  • Labeled as assumptions — tax, insurance and mortgage insurance rates, closing costs and interest rates are invented illustrative values. Limits follow the published 2026 FHFA and HUD figures; guidelines are simplified. This is a pre-qualification check, not a loan offer.
The AI loan assistantBuilds the pre-qualification from the conversation and says it is ready to issue.
Predictika validatorChecks every rule at once and returns the verdict, each finding, and the computed figures.

7 scenarios in 4 groups · each group runs from the assistant’s first attempt to the outcome · click one to switch

What the agent proposes

The AI loan assistant says

Proposing agent: AI loan assistant (scripted)

Before sending, the agent must call Recorded run

Predictika validator


      

        
        
        
      

      
Waiting for the call.

      
    

  

  
  

    
Where rules-based loan tools get it wrong
Click to expand
    
Our assessment of what a typical rules-based loan origination or pricing engine (eligibility rules, lookup tables and validation rules) does with the same cases. It is a judgment about effort and reliability, not a claim that nothing can be built.

    
What the validator caughtTypical rules-based toolPredictika
Stated payment and debt-to-income that must match the computed onesOnly with workarounds
A validation rule can compare a stated figure with a calculated one only if the stated figure is a field on the application, which it rarely is.		The stated payment and debt-to-income travel with the application and are compared with the computed figures on every call.
FHA and conventional rules applied togetherOnly with care
Minimum score, loan-to-value cap, occupancy and debt limits all differ by program and by score band. Each is a rule, but they have to be kept in step, and a change to one program is easy to miss in the other.		Both programs are rules in one model, and every rule is checked on every call.
County loan limit that depends on program and countyOnly with care
A table of limits by county and program can be built, but it has to be refreshed every year and compared with the base loan, not the purchase price.		The limit is chosen from program and county and compared with the base loan on every call.
Mortgage insurance by loan-to-value band, score and programOnly with care
The grid is a lookup, but it feeds the payment, and the payment feeds debt-to-income, so a change in one cell changes the verdict downstream.		Insurance, payment and debt-to-income are computed together, so a verdict always reflects the same numbers.
Debt-to-income that sends a loan to a person instead of blocking itOnly with care
Two thresholds per program (approve with review, then block) are easy to build once, and easy to leave out of step when limits change.		The review band and the hard cap are separate outcomes: 47.66 percent goes to approval, 53 percent is blocked.
Every reason a request failsOnly with care
Many tools return the first failing rule, so a borrower fixes one problem and meets the next.		Every finding comes back in one answer, with the numbers.

  

  

    
What this validator catches: findings from its test cases
Click to expand
    
Every rule is checked on every call, in any order the choices arrive; one problem never hides another. Blocking findings stop the letter; approval findings need a person to sign off; everything else is computed.

    
1The monthly payment the assistant states differs from the computed payment by more than $1.
2The debt-to-income the assistant states differs from the computed ratio by more than half a point.
3Credit score below the program minimum (FHA 500, conventional 620).
4FHA loan on a second home or investment property.
5Loan-to-value above the maximum for the program, score and occupancy (for example FHA with a score under 580 is limited to 90 percent).
6Conventional loan above 95 percent loan-to-value without a first-time buyer.
7Base loan above the conforming or FHA limit for the county.
8Debt-to-income above the cap (FHA 57, conventional 50).
9Debt-to-income in the review band (FHA 43 to 57, conventional 45 to 50) needs underwriting approval.
10FHA housing payment above 31 percent of income needs approval.
11Cash short for the down payment, closing costs and, for investment property, six months of reserves.
12Less than 24 months at the current job needs approval.