Grounded answers: how Graphify tags what it knows

As agents lean harder on codebase context, 'the model said so' stops being good enough. You don't need your tools to be infallible — you need them to be honest about what they actually know. That's the design principle behind Graphify's confidence tags.

Three kinds of knowing

Every relationship in the graph carries one of three tags. EXTRACTED means the edge came straight out of the tree-sitter AST: the parser found the call, the import, the class definition. It's deterministic and it doesn't hallucinate. INFERRED means your model connected the dots — a doc page linked to the service it describes, a Terraform resource matched to the code that consumes it. Usually right, but a judgment call, and labeled as one. AMBIGUOUS means Graphify found evidence it couldn't fully resolve: dynamic dispatch, imports built from strings, reflection. Rather than pick a guess and present it as fact, it keeps the edge and tells you it isn't sure.

Answers trace to a path

Because every answer is a traversal, every answer comes with its route: real files, real symbols, real edges — each with its tag. Run graphify explain on a node and you see not just what it connects to, but how each connection was established. GRAPH_REPORT.md rolls the same honesty up to the repo level, separating the architecture the parser extracted from the connections the model proposed.

The honest version of trust

Most tools flatten everything to the same confidence and let you find out the hard way which parts were guessed. We think the opposite is the only trust model that scales: keep the seams visible. When an assistant tells you the payment worker depends on the sessions table, you should be able to see in one glance whether that came from the AST or from an inference — and decide how much to lean on it accordingly.