Statewave Guide
An open-source project for AI code understanding and agent verification. By constructing an application graph and a Feature Scope mechanism, it addresses the verification blind spot in which an AI assertion is technically correct but functionally irrelevant, ensuring that verification is grounded in the correct functional context.
Timeline (last 90 days)
Product testing used the independent model OpenAI GPT-5.6 Sol to score 21 feature points one by one according to a fixed rubric, and the results showed a median usefulness of 1/3, with only 24% rated as "truly useful."
In the first round of experiments, features with verified capabilities averaged 1.43, while features without verified capabilities averaged 1.14, a negligible gap; after fixing the language issues, the same comparison became 2.00 versus 1.07.
Of the 21 outputs, 14 were judged as "too technical," and 13 were flagged as lacking some capability.
10 feature points directly output the original text "No capability, route or permission has been verified for this feature." to users, and all 10 of these feature points were rated as "too technical" (10/10).
The Statewave Guide architecture adds a Guidance Compiler layer, located between the Product Model and the Renderer, whose job is to translate verified product facts into guidance users can understand, without introducing any new unverified claims.
In a verification-driven pipeline, the renderer may not have prepared user-facing alternative copy for the "unverified" state, resulting in direct output of internal diagnostic records, exposing the lack of conversion logic between the verification layer and the presentation layer.
Safety and usefulness are two independent axes; a system can simultaneously achieve "zero false claims" and "almost entirely mediocre," and there is no automatic conversion relationship between the two.
Early experiments easily misjudged "poor expression" as "insufficient capability," because in the first round of experiments the capability signal was almost invisible, and the signal only became clear after fixing the language issues.
The Statewave Guide team used the anthropic-opus-5 model to run hundreds of tests on its in-house application, and the results showed zero fabricated routes reaching the production model, zero fabricated permissions, and zero unknown topics erroneously introduced.
The verifier only answers "whether this statement is supported by real evidence," not "whether the evidence belongs to the current feature," which leads to technically correct but completely irrelevant answers.
7 more timeline events
All Facts (12)
Early experiments easily misjudged "poor expression" as "insufficient capability," because in the first round of experiments the capability signal was almost invisible, and the signal only became clear after fixing the language issues.
50%Unverified10 feature points directly output the original text "No capability, route or permission has been verified for this feature." to users, and all 10 of these feature points were rated as "too technical" (10/10).
50%UnverifiedThe Statewave Guide architecture adds a Guidance Compiler layer, located between the Product Model and the Renderer, whose job is to translate verified product facts into guidance users can understand, without introducing any new unverified claims.
50%UnverifiedIn a verification-driven pipeline, the renderer may not have prepared user-facing alternative copy for the "unverified" state, resulting in direct output of internal diagnostic records, exposing the lack of conversion logic between the verification layer and the presentation layer.
50%UnverifiedSafety and usefulness are two independent axes; a system can simultaneously achieve "zero false claims" and "almost entirely mediocre," and there is no automatic conversion relationship between the two.
50%UnverifiedProduct testing used the independent model OpenAI GPT-5.6 Sol to score 21 feature points one by one according to a fixed rubric, and the results showed a median usefulness of 1/3, with only 24% rated as "truly useful."
50%UnverifiedIn the first round of experiments, features with verified capabilities averaged 1.43, while features without verified capabilities averaged 1.14, a negligible gap; after fixing the language issues, the same comparison became 2.00 versus 1.07.
50%UnverifiedOf the 21 outputs, 14 were judged as "too technical," and 13 were flagged as lacking some capability.
50%UnverifiedStatewave Guide's design flow was changed from "model → fabricate assertions → verifier" to "graph → deterministic assertion opportunities → model adds meaning → verifier," downgrading the model's role from fact producer to expresser.
50%UnverifiedStatewave Guide is built on top of Statewave, which is an open-source, self-hosted AI agent memory runtime emphasizing governance and provenance, under the Apache 2.0 license.
50%UnverifiedThe verifier only answers "whether this statement is supported by real evidence," not "whether the evidence belongs to the current feature," which leads to technically correct but completely irrelevant answers.
50%UnverifiedFeature Scope is an intermediate layer introduced by the Statewave Guide team to fix the verification blind spot: the graph first determines the feature currently being analyzed and its corresponding nodes and edges, and then generates candidate assertions only from within that subgraph.
50%