A click is not proof. Explore a validated GPT-6 Astra run and evidence-bounded software requiremLiveProof turns software requirements into executable proof. GPT-6 Astra operates the visible UI, while LiveProof checks the resulting postconditions and returns bounded Proof Receipts. In our validated five-requirement run, 2 were PROVEN, 2 PARTIAL, and 1 CONTRADICTED. The recovery scenario restored the project but failed to preserve collaborator permissions.ents.
What became possible in your product with Astra that was not practical before?
Maker
Before Astra, it was practical to generate test cases or automate clicks, but much harder to turn a real UI run into reviewable proof of what actually became true.
With GPT-6 Astra, LiveProof can execute the visible interface, follow a frozen requirement, and return observed postconditions, action traces, evidence references, and bounded verdicts that the application revalidates into Proof Receipts.
In our validated run, this exposed an important distinction: a deleted project was restored, but its collaborator permissions were not preserved. The action succeeded, while the requirement was still CONTRADICTED.
Astra made it practical for LiveProof to connect software requirements to real interface execution and evidence-bounded verification, rather than treating successful clicks as successful outcomes.
Report
Maker
📌
Hi Product Hunt! I built LiveProof around a simple question: what actually remained true after the click?
AI agents can perform actions, but an action itself is not proof that the required outcome happened. LiveProof connects software requirements, Proof Contracts, GPT-6 Astra UI execution, postcondition checks, and bounded Proof Receipts.
In our validated five-requirement run, two requirements were PROVEN, two PARTIAL, and one CONTRADICTED. The clearest example was a recovery scenario: the project itself was restored, but the collaborator permission state was not preserved.
That is the distinction LiveProof is built to surface.
The public demo replays a validated GPT-6 Astra Work run rather than launching a new Astra session from your browser. Evidence limitations are shown explicitly, including screenshot references whose persistent upload failed.
I’d love feedback on whether this kind of executable evidence would make AI-driven software testing easier to trust and review.