JoinWitness

JoinWitness — Decision Atlas JoinWitness is a developer-built decision-robustness application that tests whether a conclusion survives uncertainty in the relationships between records. Traditional analysis often assumes that a delivery belongs to a particular order or an attendance record matches a particular registration. When those relationships are ambiguous, even a seemingly reliable conclusion can be fragile. JoinWitness makes this uncertainty inspectable. Through four interactive stages—Import, Review, Challenge and Verify—users examine records, explicitly authorise permissible alternative relationships, and test whether those alternatives can change a decision. Its deterministic verification engine searches the authorised scope for decision-changing counterexamples. An independent checker recomputes the witness, while evidence receipts support verification and replay. Two working investigations demonstrate the system: Event Investigation: Across 60 attendance records and 90 registrations, two permitted identity changes transform a 32:28 lead into a 30:30 tie. Commerce Investigation: Across orders, payments, deliveries and refunds, one permitted delivery reassignment changes the clean-fulfilment result from 2/3 to 1/3, reversing a threshold conclusion. The innovation lies in separating three responsibilities: AI may suggest possibilities, humans authorise relationships, and deterministic verification establishes the consequences. A suggestion never silently becomes an approved match. Created and directed by Kushaagra Gupta, JoinWitness is a functioning local prototype developed with assistance from Meta Muse Code and other AI development tools. It uses fictional demonstration datasets and remains under active development. Live Meta Model API inference and public deployment are not claimed.