TTTokenTest.io
76
trust score

Router truth is partially verified

Request-side evidence is strong, but backend billing is still pending. Nonce checks and usage details are the first blockers before procurement.

OKmodel identity
OKusage fields
Watchnonce replay
Pendingreal billing
No run yet Run once to produce an evidence report.
Evidenceready
CheckStatusEvidenceProcurement impact

Prompt packet


              

              

                
Evidence digest

                

              

            

          

        

      

    


    

      Methodcctest plus billing truth
      

        

          

            
M

            
Model Truth

            
Requested model, actual model, alias policy, identity answer, and unsupported-model behavior are recorded as separate facts.

          

          

            
N

            
Nonce Guard

            
Every matrix run inserts a unique nonce. Missing or stale nonce marks replay, prompt interception, or model obedience failure.

          

          

            
U

            
Usage Ledger

            
Input, output, cache creation, and cache read tokens stay separate so billing reconciliation can be audited later.

          

          

            
B

            
Billing Boundary

            
Published prices never become real-cost conclusions. Reports stay pending until ledger, balance delta, or backend rows exist.

          

        

        

          
1. Discover
List models, protocol, pricing metadata, and visible aliases.

          
2. Probe
Run identity, long output, nonce, usage, and Claude Code behavior checks.

          
3. Reconcile
Match request window to backend ledger rows or wallet balance delta.

          
4. Compare
Calculate official expected charge only after real billing exists.

          
5. Publish
Generate a shareable HTML evidence report with secrets redacted.