Hi — I'm Ceri, author of ValiChord (github.com/topeuph-ai/ValiChord), a blind commit-reveal protocol built on Holochain for independent reproducibility verification. A researcher hashes their result before any validator starts work; validators each commit their verdict blind; everyone reveals simultaneously — making post-hoc coordination mathematically impossible. The outcome is a tamper-evident HarmonyRecord on a public distributed hash table, independently fetchable by anyone. There's a working demo at valichord-demo.onrender.com/demo if you want to see it in action.
Reading through ACG I think the two protocols are natural complements that cover different threat surfaces:
- ACG answers: did this AI faithfully derive its claims from its sources? (hallucination / source faithfulness)
- ValiChord answers: can independent parties reproduce that derivation blind, with no coordination? (independent reproducibility)
Neither solves the other's problem. Together they form a complete audit chain for AI-generated claims.
The concrete integration point I can see: the VAR (Veracity Audit Registry) that ACG produces is a hash-committed, machine-readable audit record — exactly the kind of structured output ValiChord was designed to validate. A VAR hash would be submitted as the data_hash in a ValiChord validation request. Independent validators each run the ACG verifier against the same input and commit their verdict before reveal. The resulting HarmonyRecord provides cryptographic proof that independent parties reached the same conclusion without coordinating.
Would you be open to exploring this? Happy to discuss the design, or sketch out what a joint worked example would look like.
Thank you
Hi — I'm Ceri, author of ValiChord (github.com/topeuph-ai/ValiChord), a blind commit-reveal protocol built on Holochain for independent reproducibility verification. A researcher hashes their result before any validator starts work; validators each commit their verdict blind; everyone reveals simultaneously — making post-hoc coordination mathematically impossible. The outcome is a tamper-evident HarmonyRecord on a public distributed hash table, independently fetchable by anyone. There's a working demo at valichord-demo.onrender.com/demo if you want to see it in action.
Reading through ACG I think the two protocols are natural complements that cover different threat surfaces:
Neither solves the other's problem. Together they form a complete audit chain for AI-generated claims.
The concrete integration point I can see: the VAR (Veracity Audit Registry) that ACG produces is a hash-committed, machine-readable audit record — exactly the kind of structured output ValiChord was designed to validate. A VAR hash would be submitted as the
data_hashin a ValiChord validation request. Independent validators each run the ACG verifier against the same input and commit their verdict before reveal. The resulting HarmonyRecord provides cryptographic proof that independent parties reached the same conclusion without coordinating.Would you be open to exploring this? Happy to discuss the design, or sketch out what a joint worked example would look like.
Thank you