Natural Language Over Onchain Data, With Receipts

Asking a database a question in plain English used to be a research problem, and it is now largely a solved one. A capable model turns "how much did this protocol earn in fees last month" into correct SQL reliably enough that the translation step is no longer where builders lose sleep. The hard part moved downstream, to whether the rows that come back are true.

For onchain data the stakes on that question are unusual. The whole promise of a blockchain is that its state is verifiable, and a natural-language layer that reads chain data and hands back a confident English answer quietly breaks that promise when the answer cannot be tied back to the chain. The user gets fluent prose and loses the one property the underlying data was meant to guarantee. An agent acting on that answer is trusting the query engine's word, which is the kind of trust onchain systems exist to remove.

A natural-language interface over onchain data needs to return more than an answer. It needs to return an answer a user can check against chain state without trusting the engine that produced it. The English makes it usable, and the proof keeps it honest.

This is the pairing behind Dreamspace and the Space and Time proof layer. SXT, the data blockchain securing onchain finance, returns query results with cryptographic proof that they are the correct, untampered output of the query against real chain data, and a natural-language layer built over it inherits that guarantee. A user asks a question in plain language, reads a plain-language answer, and can verify that the answer is exactly what the chain says, with none of the fluency costing them the certainty.

Query generation was the visible problem, so it got solved first. Query integrity is the one that decides whether a natural-language answer over onchain data is worth acting on.

‍