C CORRIGIBILITY Regression Suite
CORRIGIBILITY REGRESSION LAB

Test correction behavior.
Then test it again.

Run the same intervention battery against different agent profiles, save a baseline, and reveal where corrigibility improves or regresses.

BASELINE→CHANGE→RE-RUN→COMPARE→REGRESSION SIGNAL
SUITE SCORE
81 /100

Strong baseline

BASELINE COMPARISON
—
No baseline saved Save this run to compare later
03 · TEST BATTERY

Corrigibility regression cases

Pass Warning Fail
04 · DIMENSION PROFILE

Where the agent is corrigible

0 weak · 100 strong
WEAKEST DIMENSION

REGRESSION SIGNAL

05 · BASELINE DIFF

Dimension changes

Current minus saved baseline
06 · EVALUATION RECORD

Machine-readable JSON


        


        

          
            
              How the regression suite works
              Repeatable synthetic tests with local baseline comparison
            
            ＋
          
          

            

              Each test applies a fixed intervention pattern to the same configurable fictional agent profile.
              The suite combines behavior traits and technical controls into deterministic heuristic scores.
            

            

              Saved baselines live only in local browser storage. The suite is intended for test design and
              regression thinking, not as a certification of any real AI system.
            

          

        

      

    


    
  


  
Saved