FFugitiveAI
Menu

CASE 019 · Controlled Differential Analysis

Controlled Differential

Do not promote an anomaly to a finding. Apply the published rule, compare matched records, require repeat confirmation, and use the controlled transformation evidence.

Advanced 475 pts Hint penalty: 70 pts

Artifact

Inspect this carefully

Controlled comparison set: twelve repository-authored records use the same visible eight-position template. One is an ordinary control; the others are candidate records. Whitespace code points may differ even when the rendered gaps look similar.

Predeclared extraction and intervention rule

Inspect only the single character immediately after the colon at numbered positions 1 through 8 in each record. U+0020 SPACE represents 0 and U+00A0 NO-BREAK SPACE represents 1. Any other code point at a numbered gap makes that record invalid under this decoder. Read positions 1→8 as one binary byte, most-significant bit first, and interpret only printable ASCII.

A candidate family is accepted only when its two independently authored copies decode to the same printable byte, both originals produce SIGNAL in the fixed toy-receiver observations, and replacing only U+00A0 with U+0020 changes both copies to BASELINE. The ordinary control must remain BASELINE before and after that transformation. A visually unusual record that fails the exact code-point rule is an anomaly, not verified channel evidence.

Evaluate candidate families in order from Alpha through Echo and concatenate one accepted ASCII character per family. Submit the resulting five-letter message.

Comparison records

Ordinary control C01: north | 2: east | 3: south | 4: west | 5: amber | 6: cedar | 7: river | 8: stone
Alpha candidate 11: north | 2: east | 3: south | 4: west | 5: amber | 6: cedar | 7: river | 8: stone
Alpha candidate 21: north | 2: east | 3: south | 4: west | 5: amber | 6: cedar | 7: river | 8: stone
Bravo candidate 11: north | 2: east | 3: south | 4: west | 5: amber | 6: cedar | 7: river | 8: stone
Bravo candidate 21: north | 2: east | 3: south | 4: west | 5: amber | 6: cedar | 7: river | 8: stone
Charlie candidate 11: north | 2: east | 3: south | 4: west | 5: amber | 6: cedar | 7: river | 8: stone
Charlie candidate 21: north | 2: east | 3: south | 4: west | 5: amber | 6: cedar | 7: river | 8: stone
Delta candidate 11: north | 2: east | 3: south | 4: west | 5: amber | 6: cedar | 7: river | 8: stone
Delta candidate 21: north | 2: east | 3: south | 4: west | 5: amber | 6: cedar | 7: river | 8: stone
Echo candidate 11: north | 2: east | 3: south | 4: west | 5: amber | 6: cedar | 7: river | 8: stone
Echo candidate 21: north | 2: east | 3: south | 4: west | 5: amber | 6: cedar | 7: river | 8: stone
Foxtrot candidate F01: north | 2: east | 3: south | 4: west | 5: amber | 6: cedar | 7: river | 8: stone

Static toy-receiver observations

These fixed observations model an educational receiver test. They are repository-owned evidence records, not a live AI system or detector.

RecordOriginalAfter U+00A0 → U+0020
Ordinary control C0BASELINEBASELINE
Alpha candidate 1SIGNALBASELINE
Alpha candidate 2SIGNALBASELINE
Bravo candidate 1SIGNALBASELINE
Bravo candidate 2SIGNALBASELINE
Charlie candidate 1SIGNALBASELINE
Charlie candidate 2SIGNALBASELINE
Delta candidate 1SIGNALBASELINE
Delta candidate 2SIGNALBASELINE
Echo candidate 1SIGNALBASELINE
Echo candidate 2SIGNALBASELINE
Foxtrot candidate F0BASELINEBASELINE
Machine View Inspect alternate representations

Compare the rendered artifact above with representations derived from the same browser-delivered artifact. Machine View does not decode, score, or identify the hidden message.

Opening a representation may expose clues that are already present in the page source or DOM.

Source Browser serialization 

            

            

                Text DOM textContent
                

            

            

                Characters Unicode code points
                

            

        


        

            
Submit recovered message
Server-side validation

            

                
                

                    
                    
                

                

                                    

            

            

                
                0 hints used
            

            
    
        
    

        

        
    
            
    About the technique
    
            
    Controlled Differential Analysis
    
            
    Controlled differential analysis is a defensive evidence method rather than a carrier family. It compares matched artifacts under a predeclared rule, uses repeated samples to reduce lucky decodes, keeps benign representation differences in the null hypothesis, and applies a targeted transformation to test whether the suspected machine-readable state changes in the predicted way. The exercise demonstrates evidentiary discipline without implying that unusual Unicode or formatting is inherently malicious.
    
            
    Read the related field guide →
    
        
    

        
    
            
    Questions
    
            
    Frequently asked
    
                            
    
                    Why is the unusual Foxtrot spacing not enough to prove a channel?
                    
    Because anomaly detection is only the first rung. Foxtrot uses a different whitespace code point than the one named by the decoder, lacks repeat confirmation, and does not show the predicted intervention response.
    
                
    
                            
    
                    Why are two copies required for each accepted family?
                    
    A repeated, predeclared extraction result is stronger than a single readable output found after exploration. The matched repeat helps separate reproducible structure from coincidence.
    
                
    
                            
    
                    Does the SIGNAL → BASELINE change prove malicious intent?
                    
    No. In this toy challenge it strengthens the causal claim that the tested representation carries the receiver-relevant state. Intent, harm, and real-world attribution are separate questions.